Loading...
1.. SPDX-License-Identifier: GPL-2.0
2
3===================================
4Linux Ethernet Bonding Driver HOWTO
5===================================
6
7Latest update: 27 April 2011
8
9Initial release: Thomas Davis <tadavis at lbl.gov>
10
11Corrections, HA extensions: 2000/10/03-15:
12
13 - Willy Tarreau <willy at meta-x.org>
14 - Constantine Gavrilov <const-g at xpert.com>
15 - Chad N. Tindel <ctindel at ieee dot org>
16 - Janice Girouard <girouard at us dot ibm dot com>
17 - Jay Vosburgh <fubar at us dot ibm dot com>
18
19Reorganized and updated Feb 2005 by Jay Vosburgh
20Added Sysfs information: 2006/04/24
21
22 - Mitch Williams <mitch.a.williams at intel.com>
23
24Introduction
25============
26
27The Linux bonding driver provides a method for aggregating
28multiple network interfaces into a single logical "bonded" interface.
29The behavior of the bonded interfaces depends upon the mode; generally
30speaking, modes provide either hot standby or load balancing services.
31Additionally, link integrity monitoring may be performed.
32
33The bonding driver originally came from Donald Becker's
34beowulf patches for kernel 2.0. It has changed quite a bit since, and
35the original tools from extreme-linux and beowulf sites will not work
36with this version of the driver.
37
38For new versions of the driver, updated userspace tools, and
39who to ask for help, please follow the links at the end of this file.
40
41.. Table of Contents
42
43 1. Bonding Driver Installation
44
45 2. Bonding Driver Options
46
47 3. Configuring Bonding Devices
48 3.1 Configuration with Sysconfig Support
49 3.1.1 Using DHCP with Sysconfig
50 3.1.2 Configuring Multiple Bonds with Sysconfig
51 3.2 Configuration with Initscripts Support
52 3.2.1 Using DHCP with Initscripts
53 3.2.2 Configuring Multiple Bonds with Initscripts
54 3.3 Configuring Bonding Manually with Ifenslave
55 3.3.1 Configuring Multiple Bonds Manually
56 3.4 Configuring Bonding Manually via Sysfs
57 3.5 Configuration with Interfaces Support
58 3.6 Overriding Configuration for Special Cases
59 3.7 Configuring LACP for 802.3ad mode in a more secure way
60
61 4. Querying Bonding Configuration
62 4.1 Bonding Configuration
63 4.2 Network Configuration
64
65 5. Switch Configuration
66
67 6. 802.1q VLAN Support
68
69 7. Link Monitoring
70 7.1 ARP Monitor Operation
71 7.2 Configuring Multiple ARP Targets
72 7.3 MII Monitor Operation
73
74 8. Potential Trouble Sources
75 8.1 Adventures in Routing
76 8.2 Ethernet Device Renaming
77 8.3 Painfully Slow Or No Failed Link Detection By Miimon
78
79 9. SNMP agents
80
81 10. Promiscuous mode
82
83 11. Configuring Bonding for High Availability
84 11.1 High Availability in a Single Switch Topology
85 11.2 High Availability in a Multiple Switch Topology
86 11.2.1 HA Bonding Mode Selection for Multiple Switch Topology
87 11.2.2 HA Link Monitoring for Multiple Switch Topology
88
89 12. Configuring Bonding for Maximum Throughput
90 12.1 Maximum Throughput in a Single Switch Topology
91 12.1.1 MT Bonding Mode Selection for Single Switch Topology
92 12.1.2 MT Link Monitoring for Single Switch Topology
93 12.2 Maximum Throughput in a Multiple Switch Topology
94 12.2.1 MT Bonding Mode Selection for Multiple Switch Topology
95 12.2.2 MT Link Monitoring for Multiple Switch Topology
96
97 13. Switch Behavior Issues
98 13.1 Link Establishment and Failover Delays
99 13.2 Duplicated Incoming Packets
100
101 14. Hardware Specific Considerations
102 14.1 IBM BladeCenter
103
104 15. Frequently Asked Questions
105
106 16. Resources and Links
107
108
1091. Bonding Driver Installation
110==============================
111
112Most popular distro kernels ship with the bonding driver
113already available as a module. If your distro does not, or you
114have need to compile bonding from source (e.g., configuring and
115installing a mainline kernel from kernel.org), you'll need to perform
116the following steps:
117
1181.1 Configure and build the kernel with bonding
119-----------------------------------------------
120
121The current version of the bonding driver is available in the
122drivers/net/bonding subdirectory of the most recent kernel source
123(which is available on http://kernel.org). Most users "rolling their
124own" will want to use the most recent kernel from kernel.org.
125
126Configure kernel with "make menuconfig" (or "make xconfig" or
127"make config"), then select "Bonding driver support" in the "Network
128device support" section. It is recommended that you configure the
129driver as module since it is currently the only way to pass parameters
130to the driver or configure more than one bonding device.
131
132Build and install the new kernel and modules.
133
1341.2 Bonding Control Utility
135---------------------------
136
137It is recommended to configure bonding via iproute2 (netlink)
138or sysfs, the old ifenslave control utility is obsolete.
139
1402. Bonding Driver Options
141=========================
142
143Options for the bonding driver are supplied as parameters to the
144bonding module at load time, or are specified via sysfs.
145
146Module options may be given as command line arguments to the
147insmod or modprobe command, but are usually specified in either the
148``/etc/modprobe.d/*.conf`` configuration files, or in a distro-specific
149configuration file (some of which are detailed in the next section).
150
151Details on bonding support for sysfs is provided in the
152"Configuring Bonding Manually via Sysfs" section, below.
153
154The available bonding driver parameters are listed below. If a
155parameter is not specified the default value is used. When initially
156configuring a bond, it is recommended "tail -f /var/log/messages" be
157run in a separate window to watch for bonding driver error messages.
158
159It is critical that either the miimon or arp_interval and
160arp_ip_target parameters be specified, otherwise serious network
161degradation will occur during link failures. Very few devices do not
162support at least miimon, so there is really no reason not to use it.
163
164Options with textual values will accept either the text name
165or, for backwards compatibility, the option value. E.g.,
166"mode=802.3ad" and "mode=4" set the same mode.
167
168The parameters are as follows:
169
170active_slave
171
172 Specifies the new active slave for modes that support it
173 (active-backup, balance-alb and balance-tlb). Possible values
174 are the name of any currently enslaved interface, or an empty
175 string. If a name is given, the slave and its link must be up in order
176 to be selected as the new active slave. If an empty string is
177 specified, the current active slave is cleared, and a new active
178 slave is selected automatically.
179
180 Note that this is only available through the sysfs interface. No module
181 parameter by this name exists.
182
183 The normal value of this option is the name of the currently
184 active slave, or the empty string if there is no active slave or
185 the current mode does not use an active slave.
186
187ad_actor_sys_prio
188
189 In an AD system, this specifies the system priority. The allowed range
190 is 1 - 65535. If the value is not specified, it takes 65535 as the
191 default value.
192
193 This parameter has effect only in 802.3ad mode and is available through
194 SysFs interface.
195
196ad_actor_system
197
198 In an AD system, this specifies the mac-address for the actor in
199 protocol packet exchanges (LACPDUs). The value cannot be a multicast
200 address. If the all-zeroes MAC is specified, bonding will internally
201 use the MAC of the bond itself. It is preferred to have the
202 local-admin bit set for this mac but driver does not enforce it. If
203 the value is not given then system defaults to using the masters'
204 mac address as actors' system address.
205
206 This parameter has effect only in 802.3ad mode and is available through
207 SysFs interface.
208
209ad_select
210
211 Specifies the 802.3ad aggregation selection logic to use. The
212 possible values and their effects are:
213
214 stable or 0
215
216 The active aggregator is chosen by largest aggregate
217 bandwidth.
218
219 Reselection of the active aggregator occurs only when all
220 slaves of the active aggregator are down or the active
221 aggregator has no slaves.
222
223 This is the default value.
224
225 bandwidth or 1
226
227 The active aggregator is chosen by largest aggregate
228 bandwidth. Reselection occurs if:
229
230 - A slave is added to or removed from the bond
231
232 - Any slave's link state changes
233
234 - Any slave's 802.3ad association state changes
235
236 - The bond's administrative state changes to up
237
238 count or 2
239
240 The active aggregator is chosen by the largest number of
241 ports (slaves). Reselection occurs as described under the
242 "bandwidth" setting, above.
243
244 The bandwidth and count selection policies permit failover of
245 802.3ad aggregations when partial failure of the active aggregator
246 occurs. This keeps the aggregator with the highest availability
247 (either in bandwidth or in number of ports) active at all times.
248
249 This option was added in bonding version 3.4.0.
250
251ad_user_port_key
252
253 In an AD system, the port-key has three parts as shown below -
254
255 ===== ============
256 Bits Use
257 ===== ============
258 00 Duplex
259 01-05 Speed
260 06-15 User-defined
261 ===== ============
262
263 This defines the upper 10 bits of the port key. The values can be
264 from 0 - 1023. If not given, the system defaults to 0.
265
266 This parameter has effect only in 802.3ad mode and is available through
267 SysFs interface.
268
269all_slaves_active
270
271 Specifies that duplicate frames (received on inactive ports) should be
272 dropped (0) or delivered (1).
273
274 Normally, bonding will drop duplicate frames (received on inactive
275 ports), which is desirable for most users. But there are some times
276 it is nice to allow duplicate frames to be delivered.
277
278 The default value is 0 (drop duplicate frames received on inactive
279 ports).
280
281arp_interval
282
283 Specifies the ARP link monitoring frequency in milliseconds.
284
285 The ARP monitor works by periodically checking the slave
286 devices to determine whether they have sent or received
287 traffic recently (the precise criteria depends upon the
288 bonding mode, and the state of the slave). Regular traffic is
289 generated via ARP probes issued for the addresses specified by
290 the arp_ip_target option.
291
292 This behavior can be modified by the arp_validate option,
293 below.
294
295 If ARP monitoring is used in an etherchannel compatible mode
296 (modes 0 and 2), the switch should be configured in a mode
297 that evenly distributes packets across all links. If the
298 switch is configured to distribute the packets in an XOR
299 fashion, all replies from the ARP targets will be received on
300 the same link which could cause the other team members to
301 fail. ARP monitoring should not be used in conjunction with
302 miimon. A value of 0 disables ARP monitoring. The default
303 value is 0.
304
305arp_ip_target
306
307 Specifies the IP addresses to use as ARP monitoring peers when
308 arp_interval is > 0. These are the targets of the ARP request
309 sent to determine the health of the link to the targets.
310 Specify these values in ddd.ddd.ddd.ddd format. Multiple IP
311 addresses must be separated by a comma. At least one IP
312 address must be given for ARP monitoring to function. The
313 maximum number of targets that can be specified is 16. The
314 default value is no IP addresses.
315
316ns_ip6_target
317
318 Specifies the IPv6 addresses to use as IPv6 monitoring peers when
319 arp_interval is > 0. These are the targets of the NS request
320 sent to determine the health of the link to the targets.
321 Specify these values in ffff:ffff::ffff:ffff format. Multiple IPv6
322 addresses must be separated by a comma. At least one IPv6
323 address must be given for NS/NA monitoring to function. The
324 maximum number of targets that can be specified is 16. The
325 default value is no IPv6 addresses.
326
327arp_validate
328
329 Specifies whether or not ARP probes and replies should be
330 validated in any mode that supports arp monitoring, or whether
331 non-ARP traffic should be filtered (disregarded) for link
332 monitoring purposes.
333
334 Possible values are:
335
336 none or 0
337
338 No validation or filtering is performed.
339
340 active or 1
341
342 Validation is performed only for the active slave.
343
344 backup or 2
345
346 Validation is performed only for backup slaves.
347
348 all or 3
349
350 Validation is performed for all slaves.
351
352 filter or 4
353
354 Filtering is applied to all slaves. No validation is
355 performed.
356
357 filter_active or 5
358
359 Filtering is applied to all slaves, validation is performed
360 only for the active slave.
361
362 filter_backup or 6
363
364 Filtering is applied to all slaves, validation is performed
365 only for backup slaves.
366
367 Validation:
368
369 Enabling validation causes the ARP monitor to examine the incoming
370 ARP requests and replies, and only consider a slave to be up if it
371 is receiving the appropriate ARP traffic.
372
373 For an active slave, the validation checks ARP replies to confirm
374 that they were generated by an arp_ip_target. Since backup slaves
375 do not typically receive these replies, the validation performed
376 for backup slaves is on the broadcast ARP request sent out via the
377 active slave. It is possible that some switch or network
378 configurations may result in situations wherein the backup slaves
379 do not receive the ARP requests; in such a situation, validation
380 of backup slaves must be disabled.
381
382 The validation of ARP requests on backup slaves is mainly helping
383 bonding to decide which slaves are more likely to work in case of
384 the active slave failure, it doesn't really guarantee that the
385 backup slave will work if it's selected as the next active slave.
386
387 Validation is useful in network configurations in which multiple
388 bonding hosts are concurrently issuing ARPs to one or more targets
389 beyond a common switch. Should the link between the switch and
390 target fail (but not the switch itself), the probe traffic
391 generated by the multiple bonding instances will fool the standard
392 ARP monitor into considering the links as still up. Use of
393 validation can resolve this, as the ARP monitor will only consider
394 ARP requests and replies associated with its own instance of
395 bonding.
396
397 Filtering:
398
399 Enabling filtering causes the ARP monitor to only use incoming ARP
400 packets for link availability purposes. Arriving packets that are
401 not ARPs are delivered normally, but do not count when determining
402 if a slave is available.
403
404 Filtering operates by only considering the reception of ARP
405 packets (any ARP packet, regardless of source or destination) when
406 determining if a slave has received traffic for link availability
407 purposes.
408
409 Filtering is useful in network configurations in which significant
410 levels of third party broadcast traffic would fool the standard
411 ARP monitor into considering the links as still up. Use of
412 filtering can resolve this, as only ARP traffic is considered for
413 link availability purposes.
414
415 This option was added in bonding version 3.1.0.
416
417arp_all_targets
418
419 Specifies the quantity of arp_ip_targets that must be reachable
420 in order for the ARP monitor to consider a slave as being up.
421 This option affects only active-backup mode for slaves with
422 arp_validation enabled.
423
424 Possible values are:
425
426 any or 0
427
428 consider the slave up only when any of the arp_ip_targets
429 is reachable
430
431 all or 1
432
433 consider the slave up only when all of the arp_ip_targets
434 are reachable
435
436arp_missed_max
437
438 Specifies the number of arp_interval monitor checks that must
439 fail in order for an interface to be marked down by the ARP monitor.
440
441 In order to provide orderly failover semantics, backup interfaces
442 are permitted an extra monitor check (i.e., they must fail
443 arp_missed_max + 1 times before being marked down).
444
445 The default value is 2, and the allowable range is 1 - 255.
446
447coupled_control
448
449 Specifies whether the LACP state machine's MUX in the 802.3ad mode
450 should have separate Collecting and Distributing states.
451
452 This is by implementing the independent control state machine per
453 IEEE 802.1AX-2008 5.4.15 in addition to the existing coupled control
454 state machine.
455
456 The default value is 1. This setting does not separate the Collecting
457 and Distributing states, maintaining the bond in coupled control.
458
459downdelay
460
461 Specifies the time, in milliseconds, to wait before disabling
462 a slave after a link failure has been detected. This option
463 is only valid for the miimon link monitor. The downdelay
464 value should be a multiple of the miimon value; if not, it
465 will be rounded down to the nearest multiple. The default
466 value is 0.
467
468fail_over_mac
469
470 Specifies whether active-backup mode should set all slaves to
471 the same MAC address at enslavement (the traditional
472 behavior), or, when enabled, perform special handling of the
473 bond's MAC address in accordance with the selected policy.
474
475 Possible values are:
476
477 none or 0
478
479 This setting disables fail_over_mac, and causes
480 bonding to set all slaves of an active-backup bond to
481 the same MAC address at enslavement time. This is the
482 default.
483
484 active or 1
485
486 The "active" fail_over_mac policy indicates that the
487 MAC address of the bond should always be the MAC
488 address of the currently active slave. The MAC
489 address of the slaves is not changed; instead, the MAC
490 address of the bond changes during a failover.
491
492 This policy is useful for devices that cannot ever
493 alter their MAC address, or for devices that refuse
494 incoming broadcasts with their own source MAC (which
495 interferes with the ARP monitor).
496
497 The down side of this policy is that every device on
498 the network must be updated via gratuitous ARP,
499 vs. just updating a switch or set of switches (which
500 often takes place for any traffic, not just ARP
501 traffic, if the switch snoops incoming traffic to
502 update its tables) for the traditional method. If the
503 gratuitous ARP is lost, communication may be
504 disrupted.
505
506 When this policy is used in conjunction with the mii
507 monitor, devices which assert link up prior to being
508 able to actually transmit and receive are particularly
509 susceptible to loss of the gratuitous ARP, and an
510 appropriate updelay setting may be required.
511
512 follow or 2
513
514 The "follow" fail_over_mac policy causes the MAC
515 address of the bond to be selected normally (normally
516 the MAC address of the first slave added to the bond).
517 However, the second and subsequent slaves are not set
518 to this MAC address while they are in a backup role; a
519 slave is programmed with the bond's MAC address at
520 failover time (and the formerly active slave receives
521 the newly active slave's MAC address).
522
523 This policy is useful for multiport devices that
524 either become confused or incur a performance penalty
525 when multiple ports are programmed with the same MAC
526 address.
527
528
529 The default policy is none, unless the first slave cannot
530 change its MAC address, in which case the active policy is
531 selected by default.
532
533 This option may be modified via sysfs only when no slaves are
534 present in the bond.
535
536 This option was added in bonding version 3.2.0. The "follow"
537 policy was added in bonding version 3.3.0.
538
539lacp_active
540 Option specifying whether to send LACPDU frames periodically.
541
542 off or 0
543 LACPDU frames acts as "speak when spoken to".
544
545 on or 1
546 LACPDU frames are sent along the configured links
547 periodically. See lacp_rate for more details.
548
549 The default is on.
550
551lacp_rate
552
553 Option specifying the rate in which we'll ask our link partner
554 to transmit LACPDU packets in 802.3ad mode. Possible values
555 are:
556
557 slow or 0
558 Request partner to transmit LACPDUs every 30 seconds
559
560 fast or 1
561 Request partner to transmit LACPDUs every 1 second
562
563 The default is slow.
564
565max_bonds
566
567 Specifies the number of bonding devices to create for this
568 instance of the bonding driver. E.g., if max_bonds is 3, and
569 the bonding driver is not already loaded, then bond0, bond1
570 and bond2 will be created. The default value is 1. Specifying
571 a value of 0 will load bonding, but will not create any devices.
572
573miimon
574
575 Specifies the MII link monitoring frequency in milliseconds.
576 This determines how often the link state of each slave is
577 inspected for link failures. A value of zero disables MII
578 link monitoring. A value of 100 is a good starting point.
579 The use_carrier option, below, affects how the link state is
580 determined. See the High Availability section for additional
581 information. The default value is 100 if arp_interval is not
582 set.
583
584min_links
585
586 Specifies the minimum number of links that must be active before
587 asserting carrier. It is similar to the Cisco EtherChannel min-links
588 feature. This allows setting the minimum number of member ports that
589 must be up (link-up state) before marking the bond device as up
590 (carrier on). This is useful for situations where higher level services
591 such as clustering want to ensure a minimum number of low bandwidth
592 links are active before switchover. This option only affect 802.3ad
593 mode.
594
595 The default value is 0. This will cause carrier to be asserted (for
596 802.3ad mode) whenever there is an active aggregator, regardless of the
597 number of available links in that aggregator. Note that, because an
598 aggregator cannot be active without at least one available link,
599 setting this option to 0 or to 1 has the exact same effect.
600
601mode
602
603 Specifies one of the bonding policies. The default is
604 balance-rr (round robin). Possible values are:
605
606 balance-rr or 0
607
608 Round-robin policy: Transmit packets in sequential
609 order from the first available slave through the
610 last. This mode provides load balancing and fault
611 tolerance.
612
613 active-backup or 1
614
615 Active-backup policy: Only one slave in the bond is
616 active. A different slave becomes active if, and only
617 if, the active slave fails. The bond's MAC address is
618 externally visible on only one port (network adapter)
619 to avoid confusing the switch.
620
621 In bonding version 2.6.2 or later, when a failover
622 occurs in active-backup mode, bonding will issue one
623 or more gratuitous ARPs on the newly active slave.
624 One gratuitous ARP is issued for the bonding master
625 interface and each VLAN interfaces configured above
626 it, provided that the interface has at least one IP
627 address configured. Gratuitous ARPs issued for VLAN
628 interfaces are tagged with the appropriate VLAN id.
629
630 This mode provides fault tolerance. The primary
631 option, documented below, affects the behavior of this
632 mode.
633
634 balance-xor or 2
635
636 XOR policy: Transmit based on the selected transmit
637 hash policy. The default policy is a simple [(source
638 MAC address XOR'd with destination MAC address XOR
639 packet type ID) modulo slave count]. Alternate transmit
640 policies may be selected via the xmit_hash_policy option,
641 described below.
642
643 This mode provides load balancing and fault tolerance.
644
645 broadcast or 3
646
647 Broadcast policy: transmits everything on all slave
648 interfaces. This mode provides fault tolerance.
649
650 802.3ad or 4
651
652 IEEE 802.3ad Dynamic link aggregation. Creates
653 aggregation groups that share the same speed and
654 duplex settings. Utilizes all slaves in the active
655 aggregator according to the 802.3ad specification.
656
657 Slave selection for outgoing traffic is done according
658 to the transmit hash policy, which may be changed from
659 the default simple XOR policy via the xmit_hash_policy
660 option, documented below. Note that not all transmit
661 policies may be 802.3ad compliant, particularly in
662 regards to the packet mis-ordering requirements of
663 section 43.2.4 of the 802.3ad standard. Differing
664 peer implementations will have varying tolerances for
665 noncompliance.
666
667 Prerequisites:
668
669 1. Ethtool support in the base drivers for retrieving
670 the speed and duplex of each slave.
671
672 2. A switch that supports IEEE 802.3ad Dynamic link
673 aggregation.
674
675 Most switches will require some type of configuration
676 to enable 802.3ad mode.
677
678 balance-tlb or 5
679
680 Adaptive transmit load balancing: channel bonding that
681 does not require any special switch support.
682
683 In tlb_dynamic_lb=1 mode; the outgoing traffic is
684 distributed according to the current load (computed
685 relative to the speed) on each slave.
686
687 In tlb_dynamic_lb=0 mode; the load balancing based on
688 current load is disabled and the load is distributed
689 only using the hash distribution.
690
691 Incoming traffic is received by the current slave.
692 If the receiving slave fails, another slave takes over
693 the MAC address of the failed receiving slave.
694
695 Prerequisite:
696
697 Ethtool support in the base drivers for retrieving the
698 speed of each slave.
699
700 balance-alb or 6
701
702 Adaptive load balancing: includes balance-tlb plus
703 receive load balancing (rlb) for IPV4 traffic, and
704 does not require any special switch support. The
705 receive load balancing is achieved by ARP negotiation.
706 The bonding driver intercepts the ARP Replies sent by
707 the local system on their way out and overwrites the
708 source hardware address with the unique hardware
709 address of one of the slaves in the bond such that
710 different peers use different hardware addresses for
711 the server.
712
713 Receive traffic from connections created by the server
714 is also balanced. When the local system sends an ARP
715 Request the bonding driver copies and saves the peer's
716 IP information from the ARP packet. When the ARP
717 Reply arrives from the peer, its hardware address is
718 retrieved and the bonding driver initiates an ARP
719 reply to this peer assigning it to one of the slaves
720 in the bond. A problematic outcome of using ARP
721 negotiation for balancing is that each time that an
722 ARP request is broadcast it uses the hardware address
723 of the bond. Hence, peers learn the hardware address
724 of the bond and the balancing of receive traffic
725 collapses to the current slave. This is handled by
726 sending updates (ARP Replies) to all the peers with
727 their individually assigned hardware address such that
728 the traffic is redistributed. Receive traffic is also
729 redistributed when a new slave is added to the bond
730 and when an inactive slave is re-activated. The
731 receive load is distributed sequentially (round robin)
732 among the group of highest speed slaves in the bond.
733
734 When a link is reconnected or a new slave joins the
735 bond the receive traffic is redistributed among all
736 active slaves in the bond by initiating ARP Replies
737 with the selected MAC address to each of the
738 clients. The updelay parameter (detailed below) must
739 be set to a value equal or greater than the switch's
740 forwarding delay so that the ARP Replies sent to the
741 peers will not be blocked by the switch.
742
743 Prerequisites:
744
745 1. Ethtool support in the base drivers for retrieving
746 the speed of each slave.
747
748 2. Base driver support for setting the hardware
749 address of a device while it is open. This is
750 required so that there will always be one slave in the
751 team using the bond hardware address (the
752 curr_active_slave) while having a unique hardware
753 address for each slave in the bond. If the
754 curr_active_slave fails its hardware address is
755 swapped with the new curr_active_slave that was
756 chosen.
757
758num_grat_arp,
759num_unsol_na
760
761 Specify the number of peer notifications (gratuitous ARPs and
762 unsolicited IPv6 Neighbor Advertisements) to be issued after a
763 failover event. As soon as the link is up on the new slave
764 (possibly immediately) a peer notification is sent on the
765 bonding device and each VLAN sub-device. This is repeated at
766 the rate specified by peer_notif_delay if the number is
767 greater than 1.
768
769 The valid range is 0 - 255; the default value is 1. These options
770 affect only the active-backup mode. These options were added for
771 bonding versions 3.3.0 and 3.4.0 respectively.
772
773 From Linux 3.0 and bonding version 3.7.1, these notifications
774 are generated by the ipv4 and ipv6 code and the numbers of
775 repetitions cannot be set independently.
776
777packets_per_slave
778
779 Specify the number of packets to transmit through a slave before
780 moving to the next one. When set to 0 then a slave is chosen at
781 random.
782
783 The valid range is 0 - 65535; the default value is 1. This option
784 has effect only in balance-rr mode.
785
786peer_notif_delay
787
788 Specify the delay, in milliseconds, between each peer
789 notification (gratuitous ARP and unsolicited IPv6 Neighbor
790 Advertisement) when they are issued after a failover event.
791 This delay should be a multiple of the MII link monitor interval
792 (miimon).
793
794 The valid range is 0 - 300000. The default value is 0, which means
795 to match the value of the MII link monitor interval.
796
797prio
798 Slave priority. A higher number means higher priority.
799 The primary slave has the highest priority. This option also
800 follows the primary_reselect rules.
801
802 This option could only be configured via netlink, and is only valid
803 for active-backup(1), balance-tlb (5) and balance-alb (6) mode.
804 The valid value range is a signed 32 bit integer.
805
806 The default value is 0.
807
808primary
809
810 A string (eth0, eth2, etc) specifying which slave is the
811 primary device. The specified device will always be the
812 active slave while it is available. Only when the primary is
813 off-line will alternate devices be used. This is useful when
814 one slave is preferred over another, e.g., when one slave has
815 higher throughput than another.
816
817 The primary option is only valid for active-backup(1),
818 balance-tlb (5) and balance-alb (6) mode.
819
820primary_reselect
821
822 Specifies the reselection policy for the primary slave. This
823 affects how the primary slave is chosen to become the active slave
824 when failure of the active slave or recovery of the primary slave
825 occurs. This option is designed to prevent flip-flopping between
826 the primary slave and other slaves. Possible values are:
827
828 always or 0 (default)
829
830 The primary slave becomes the active slave whenever it
831 comes back up.
832
833 better or 1
834
835 The primary slave becomes the active slave when it comes
836 back up, if the speed and duplex of the primary slave is
837 better than the speed and duplex of the current active
838 slave.
839
840 failure or 2
841
842 The primary slave becomes the active slave only if the
843 current active slave fails and the primary slave is up.
844
845 The primary_reselect setting is ignored in two cases:
846
847 If no slaves are active, the first slave to recover is
848 made the active slave.
849
850 When initially enslaved, the primary slave is always made
851 the active slave.
852
853 Changing the primary_reselect policy via sysfs will cause an
854 immediate selection of the best active slave according to the new
855 policy. This may or may not result in a change of the active
856 slave, depending upon the circumstances.
857
858 This option was added for bonding version 3.6.0.
859
860tlb_dynamic_lb
861
862 Specifies if dynamic shuffling of flows is enabled in tlb
863 or alb mode. The value has no effect on any other modes.
864
865 The default behavior of tlb mode is to shuffle active flows across
866 slaves based on the load in that interval. This gives nice lb
867 characteristics but can cause packet reordering. If re-ordering is
868 a concern use this variable to disable flow shuffling and rely on
869 load balancing provided solely by the hash distribution.
870 xmit-hash-policy can be used to select the appropriate hashing for
871 the setup.
872
873 The sysfs entry can be used to change the setting per bond device
874 and the initial value is derived from the module parameter. The
875 sysfs entry is allowed to be changed only if the bond device is
876 down.
877
878 The default value is "1" that enables flow shuffling while value "0"
879 disables it. This option was added in bonding driver 3.7.1
880
881
882updelay
883
884 Specifies the time, in milliseconds, to wait before enabling a
885 slave after a link recovery has been detected. This option is
886 only valid for the miimon link monitor. The updelay value
887 should be a multiple of the miimon value; if not, it will be
888 rounded down to the nearest multiple. The default value is 0.
889
890use_carrier
891
892 Specifies whether or not miimon should use MII or ETHTOOL
893 ioctls vs. netif_carrier_ok() to determine the link
894 status. The MII or ETHTOOL ioctls are less efficient and
895 utilize a deprecated calling sequence within the kernel. The
896 netif_carrier_ok() relies on the device driver to maintain its
897 state with netif_carrier_on/off; at this writing, most, but
898 not all, device drivers support this facility.
899
900 If bonding insists that the link is up when it should not be,
901 it may be that your network device driver does not support
902 netif_carrier_on/off. The default state for netif_carrier is
903 "carrier on," so if a driver does not support netif_carrier,
904 it will appear as if the link is always up. In this case,
905 setting use_carrier to 0 will cause bonding to revert to the
906 MII / ETHTOOL ioctl method to determine the link state.
907
908 A value of 1 enables the use of netif_carrier_ok(), a value of
909 0 will use the deprecated MII / ETHTOOL ioctls. The default
910 value is 1.
911
912xmit_hash_policy
913
914 Selects the transmit hash policy to use for slave selection in
915 balance-xor, 802.3ad, and tlb modes. Possible values are:
916
917 layer2
918
919 Uses XOR of hardware MAC addresses and packet type ID
920 field to generate the hash. The formula is
921
922 hash = source MAC[5] XOR destination MAC[5] XOR packet type ID
923 slave number = hash modulo slave count
924
925 This algorithm will place all traffic to a particular
926 network peer on the same slave.
927
928 This algorithm is 802.3ad compliant.
929
930 layer2+3
931
932 This policy uses a combination of layer2 and layer3
933 protocol information to generate the hash.
934
935 Uses XOR of hardware MAC addresses and IP addresses to
936 generate the hash. The formula is
937
938 hash = source MAC[5] XOR destination MAC[5] XOR packet type ID
939 hash = hash XOR source IP XOR destination IP
940 hash = hash XOR (hash RSHIFT 16)
941 hash = hash XOR (hash RSHIFT 8)
942 And then hash is reduced modulo slave count.
943
944 If the protocol is IPv6 then the source and destination
945 addresses are first hashed using ipv6_addr_hash.
946
947 This algorithm will place all traffic to a particular
948 network peer on the same slave. For non-IP traffic,
949 the formula is the same as for the layer2 transmit
950 hash policy.
951
952 This policy is intended to provide a more balanced
953 distribution of traffic than layer2 alone, especially
954 in environments where a layer3 gateway device is
955 required to reach most destinations.
956
957 This algorithm is 802.3ad compliant.
958
959 layer3+4
960
961 This policy uses upper layer protocol information,
962 when available, to generate the hash. This allows for
963 traffic to a particular network peer to span multiple
964 slaves, although a single connection will not span
965 multiple slaves.
966
967 The formula for unfragmented TCP and UDP packets is
968
969 hash = source port, destination port (as in the header)
970 hash = hash XOR source IP XOR destination IP
971 hash = hash XOR (hash RSHIFT 16)
972 hash = hash XOR (hash RSHIFT 8)
973 hash = hash RSHIFT 1
974 And then hash is reduced modulo slave count.
975
976 If the protocol is IPv6 then the source and destination
977 addresses are first hashed using ipv6_addr_hash.
978
979 For fragmented TCP or UDP packets and all other IPv4 and
980 IPv6 protocol traffic, the source and destination port
981 information is omitted. For non-IP traffic, the
982 formula is the same as for the layer2 transmit hash
983 policy.
984
985 This algorithm is not fully 802.3ad compliant. A
986 single TCP or UDP conversation containing both
987 fragmented and unfragmented packets will see packets
988 striped across two interfaces. This may result in out
989 of order delivery. Most traffic types will not meet
990 this criteria, as TCP rarely fragments traffic, and
991 most UDP traffic is not involved in extended
992 conversations. Other implementations of 802.3ad may
993 or may not tolerate this noncompliance.
994
995 encap2+3
996
997 This policy uses the same formula as layer2+3 but it
998 relies on skb_flow_dissect to obtain the header fields
999 which might result in the use of inner headers if an
1000 encapsulation protocol is used. For example this will
1001 improve the performance for tunnel users because the
1002 packets will be distributed according to the encapsulated
1003 flows.
1004
1005 encap3+4
1006
1007 This policy uses the same formula as layer3+4 but it
1008 relies on skb_flow_dissect to obtain the header fields
1009 which might result in the use of inner headers if an
1010 encapsulation protocol is used. For example this will
1011 improve the performance for tunnel users because the
1012 packets will be distributed according to the encapsulated
1013 flows.
1014
1015 vlan+srcmac
1016
1017 This policy uses a very rudimentary vlan ID and source mac
1018 hash to load-balance traffic per-vlan, with failover
1019 should one leg fail. The intended use case is for a bond
1020 shared by multiple virtual machines, all configured to
1021 use their own vlan, to give lacp-like functionality
1022 without requiring lacp-capable switching hardware.
1023
1024 The formula for the hash is simply
1025
1026 hash = (vlan ID) XOR (source MAC vendor) XOR (source MAC dev)
1027
1028 The default value is layer2. This option was added in bonding
1029 version 2.6.3. In earlier versions of bonding, this parameter
1030 does not exist, and the layer2 policy is the only policy. The
1031 layer2+3 value was added for bonding version 3.2.2.
1032
1033resend_igmp
1034
1035 Specifies the number of IGMP membership reports to be issued after
1036 a failover event. One membership report is issued immediately after
1037 the failover, subsequent packets are sent in each 200ms interval.
1038
1039 The valid range is 0 - 255; the default value is 1. A value of 0
1040 prevents the IGMP membership report from being issued in response
1041 to the failover event.
1042
1043 This option is useful for bonding modes balance-rr (0), active-backup
1044 (1), balance-tlb (5) and balance-alb (6), in which a failover can
1045 switch the IGMP traffic from one slave to another. Therefore a fresh
1046 IGMP report must be issued to cause the switch to forward the incoming
1047 IGMP traffic over the newly selected slave.
1048
1049 This option was added for bonding version 3.7.0.
1050
1051lp_interval
1052
1053 Specifies the number of seconds between instances where the bonding
1054 driver sends learning packets to each slaves peer switch.
1055
1056 The valid range is 1 - 0x7fffffff; the default value is 1. This Option
1057 has effect only in balance-tlb and balance-alb modes.
1058
10593. Configuring Bonding Devices
1060==============================
1061
1062You can configure bonding using either your distro's network
1063initialization scripts, or manually using either iproute2 or the
1064sysfs interface. Distros generally use one of three packages for the
1065network initialization scripts: initscripts, sysconfig or interfaces.
1066Recent versions of these packages have support for bonding, while older
1067versions do not.
1068
1069We will first describe the options for configuring bonding for
1070distros using versions of initscripts, sysconfig and interfaces with full
1071or partial support for bonding, then provide information on enabling
1072bonding without support from the network initialization scripts (i.e.,
1073older versions of initscripts or sysconfig).
1074
1075If you're unsure whether your distro uses sysconfig,
1076initscripts or interfaces, or don't know if it's new enough, have no fear.
1077Determining this is fairly straightforward.
1078
1079First, look for a file called interfaces in /etc/network directory.
1080If this file is present in your system, then your system use interfaces. See
1081Configuration with Interfaces Support.
1082
1083Else, issue the command::
1084
1085 $ rpm -qf /sbin/ifup
1086
1087It will respond with a line of text starting with either
1088"initscripts" or "sysconfig," followed by some numbers. This is the
1089package that provides your network initialization scripts.
1090
1091Next, to determine if your installation supports bonding,
1092issue the command::
1093
1094 $ grep ifenslave /sbin/ifup
1095
1096If this returns any matches, then your initscripts or
1097sysconfig has support for bonding.
1098
10993.1 Configuration with Sysconfig Support
1100----------------------------------------
1101
1102This section applies to distros using a version of sysconfig
1103with bonding support, for example, SuSE Linux Enterprise Server 9.
1104
1105SuSE SLES 9's networking configuration system does support
1106bonding, however, at this writing, the YaST system configuration
1107front end does not provide any means to work with bonding devices.
1108Bonding devices can be managed by hand, however, as follows.
1109
1110First, if they have not already been configured, configure the
1111slave devices. On SLES 9, this is most easily done by running the
1112yast2 sysconfig configuration utility. The goal is for to create an
1113ifcfg-id file for each slave device. The simplest way to accomplish
1114this is to configure the devices for DHCP (this is only to get the
1115file ifcfg-id file created; see below for some issues with DHCP). The
1116name of the configuration file for each device will be of the form::
1117
1118 ifcfg-id-xx:xx:xx:xx:xx:xx
1119
1120Where the "xx" portion will be replaced with the digits from
1121the device's permanent MAC address.
1122
1123Once the set of ifcfg-id-xx:xx:xx:xx:xx:xx files has been
1124created, it is necessary to edit the configuration files for the slave
1125devices (the MAC addresses correspond to those of the slave devices).
1126Before editing, the file will contain multiple lines, and will look
1127something like this::
1128
1129 BOOTPROTO='dhcp'
1130 STARTMODE='on'
1131 USERCTL='no'
1132 UNIQUE='XNzu.WeZGOGF+4wE'
1133 _nm_name='bus-pci-0001:61:01.0'
1134
1135Change the BOOTPROTO and STARTMODE lines to the following::
1136
1137 BOOTPROTO='none'
1138 STARTMODE='off'
1139
1140Do not alter the UNIQUE or _nm_name lines. Remove any other
1141lines (USERCTL, etc).
1142
1143Once the ifcfg-id-xx:xx:xx:xx:xx:xx files have been modified,
1144it's time to create the configuration file for the bonding device
1145itself. This file is named ifcfg-bondX, where X is the number of the
1146bonding device to create, starting at 0. The first such file is
1147ifcfg-bond0, the second is ifcfg-bond1, and so on. The sysconfig
1148network configuration system will correctly start multiple instances
1149of bonding.
1150
1151The contents of the ifcfg-bondX file is as follows::
1152
1153 BOOTPROTO="static"
1154 BROADCAST="10.0.2.255"
1155 IPADDR="10.0.2.10"
1156 NETMASK="255.255.0.0"
1157 NETWORK="10.0.2.0"
1158 REMOTE_IPADDR=""
1159 STARTMODE="onboot"
1160 BONDING_MASTER="yes"
1161 BONDING_MODULE_OPTS="mode=active-backup miimon=100"
1162 BONDING_SLAVE0="eth0"
1163 BONDING_SLAVE1="bus-pci-0000:06:08.1"
1164
1165Replace the sample BROADCAST, IPADDR, NETMASK and NETWORK
1166values with the appropriate values for your network.
1167
1168The STARTMODE specifies when the device is brought online.
1169The possible values are:
1170
1171 ======== ======================================================
1172 onboot The device is started at boot time. If you're not
1173 sure, this is probably what you want.
1174
1175 manual The device is started only when ifup is called
1176 manually. Bonding devices may be configured this
1177 way if you do not wish them to start automatically
1178 at boot for some reason.
1179
1180 hotplug The device is started by a hotplug event. This is not
1181 a valid choice for a bonding device.
1182
1183 off or The device configuration is ignored.
1184 ignore
1185 ======== ======================================================
1186
1187The line BONDING_MASTER='yes' indicates that the device is a
1188bonding master device. The only useful value is "yes."
1189
1190The contents of BONDING_MODULE_OPTS are supplied to the
1191instance of the bonding module for this device. Specify the options
1192for the bonding mode, link monitoring, and so on here. Do not include
1193the max_bonds bonding parameter; this will confuse the configuration
1194system if you have multiple bonding devices.
1195
1196Finally, supply one BONDING_SLAVEn="slave device" for each
1197slave. where "n" is an increasing value, one for each slave. The
1198"slave device" is either an interface name, e.g., "eth0", or a device
1199specifier for the network device. The interface name is easier to
1200find, but the ethN names are subject to change at boot time if, e.g.,
1201a device early in the sequence has failed. The device specifiers
1202(bus-pci-0000:06:08.1 in the example above) specify the physical
1203network device, and will not change unless the device's bus location
1204changes (for example, it is moved from one PCI slot to another). The
1205example above uses one of each type for demonstration purposes; most
1206configurations will choose one or the other for all slave devices.
1207
1208When all configuration files have been modified or created,
1209networking must be restarted for the configuration changes to take
1210effect. This can be accomplished via the following::
1211
1212 # /etc/init.d/network restart
1213
1214Note that the network control script (/sbin/ifdown) will
1215remove the bonding module as part of the network shutdown processing,
1216so it is not necessary to remove the module by hand if, e.g., the
1217module parameters have changed.
1218
1219Also, at this writing, YaST/YaST2 will not manage bonding
1220devices (they do not show bonding interfaces on its list of network
1221devices). It is necessary to edit the configuration file by hand to
1222change the bonding configuration.
1223
1224Additional general options and details of the ifcfg file
1225format can be found in an example ifcfg template file::
1226
1227 /etc/sysconfig/network/ifcfg.template
1228
1229Note that the template does not document the various ``BONDING_*``
1230settings described above, but does describe many of the other options.
1231
12323.1.1 Using DHCP with Sysconfig
1233-------------------------------
1234
1235Under sysconfig, configuring a device with BOOTPROTO='dhcp'
1236will cause it to query DHCP for its IP address information. At this
1237writing, this does not function for bonding devices; the scripts
1238attempt to obtain the device address from DHCP prior to adding any of
1239the slave devices. Without active slaves, the DHCP requests are not
1240sent to the network.
1241
12423.1.2 Configuring Multiple Bonds with Sysconfig
1243-----------------------------------------------
1244
1245The sysconfig network initialization system is capable of
1246handling multiple bonding devices. All that is necessary is for each
1247bonding instance to have an appropriately configured ifcfg-bondX file
1248(as described above). Do not specify the "max_bonds" parameter to any
1249instance of bonding, as this will confuse sysconfig. If you require
1250multiple bonding devices with identical parameters, create multiple
1251ifcfg-bondX files.
1252
1253Because the sysconfig scripts supply the bonding module
1254options in the ifcfg-bondX file, it is not necessary to add them to
1255the system ``/etc/modules.d/*.conf`` configuration files.
1256
12573.2 Configuration with Initscripts Support
1258------------------------------------------
1259
1260This section applies to distros using a recent version of
1261initscripts with bonding support, for example, Red Hat Enterprise Linux
1262version 3 or later, Fedora, etc. On these systems, the network
1263initialization scripts have knowledge of bonding, and can be configured to
1264control bonding devices. Note that older versions of the initscripts
1265package have lower levels of support for bonding; this will be noted where
1266applicable.
1267
1268These distros will not automatically load the network adapter
1269driver unless the ethX device is configured with an IP address.
1270Because of this constraint, users must manually configure a
1271network-script file for all physical adapters that will be members of
1272a bondX link. Network script files are located in the directory:
1273
1274/etc/sysconfig/network-scripts
1275
1276The file name must be prefixed with "ifcfg-eth" and suffixed
1277with the adapter's physical adapter number. For example, the script
1278for eth0 would be named /etc/sysconfig/network-scripts/ifcfg-eth0.
1279Place the following text in the file::
1280
1281 DEVICE=eth0
1282 USERCTL=no
1283 ONBOOT=yes
1284 MASTER=bond0
1285 SLAVE=yes
1286 BOOTPROTO=none
1287
1288The DEVICE= line will be different for every ethX device and
1289must correspond with the name of the file, i.e., ifcfg-eth1 must have
1290a device line of DEVICE=eth1. The setting of the MASTER= line will
1291also depend on the final bonding interface name chosen for your bond.
1292As with other network devices, these typically start at 0, and go up
1293one for each device, i.e., the first bonding instance is bond0, the
1294second is bond1, and so on.
1295
1296Next, create a bond network script. The file name for this
1297script will be /etc/sysconfig/network-scripts/ifcfg-bondX where X is
1298the number of the bond. For bond0 the file is named "ifcfg-bond0",
1299for bond1 it is named "ifcfg-bond1", and so on. Within that file,
1300place the following text::
1301
1302 DEVICE=bond0
1303 IPADDR=192.168.1.1
1304 NETMASK=255.255.255.0
1305 NETWORK=192.168.1.0
1306 BROADCAST=192.168.1.255
1307 ONBOOT=yes
1308 BOOTPROTO=none
1309 USERCTL=no
1310
1311Be sure to change the networking specific lines (IPADDR,
1312NETMASK, NETWORK and BROADCAST) to match your network configuration.
1313
1314For later versions of initscripts, such as that found with Fedora
13157 (or later) and Red Hat Enterprise Linux version 5 (or later), it is possible,
1316and, indeed, preferable, to specify the bonding options in the ifcfg-bond0
1317file, e.g. a line of the format::
1318
1319 BONDING_OPTS="mode=active-backup arp_interval=60 arp_ip_target=192.168.1.254"
1320
1321will configure the bond with the specified options. The options
1322specified in BONDING_OPTS are identical to the bonding module parameters
1323except for the arp_ip_target field when using versions of initscripts older
1324than and 8.57 (Fedora 8) and 8.45.19 (Red Hat Enterprise Linux 5.2). When
1325using older versions each target should be included as a separate option and
1326should be preceded by a '+' to indicate it should be added to the list of
1327queried targets, e.g.,::
1328
1329 arp_ip_target=+192.168.1.1 arp_ip_target=+192.168.1.2
1330
1331is the proper syntax to specify multiple targets. When specifying
1332options via BONDING_OPTS, it is not necessary to edit
1333``/etc/modprobe.d/*.conf``.
1334
1335For even older versions of initscripts that do not support
1336BONDING_OPTS, it is necessary to edit /etc/modprobe.d/*.conf, depending upon
1337your distro) to load the bonding module with your desired options when the
1338bond0 interface is brought up. The following lines in /etc/modprobe.d/*.conf
1339will load the bonding module, and select its options:
1340
1341 alias bond0 bonding
1342 options bond0 mode=balance-alb miimon=100
1343
1344Replace the sample parameters with the appropriate set of
1345options for your configuration.
1346
1347Finally run "/etc/rc.d/init.d/network restart" as root. This
1348will restart the networking subsystem and your bond link should be now
1349up and running.
1350
13513.2.1 Using DHCP with Initscripts
1352---------------------------------
1353
1354Recent versions of initscripts (the versions supplied with Fedora
1355Core 3 and Red Hat Enterprise Linux 4, or later versions, are reported to
1356work) have support for assigning IP information to bonding devices via
1357DHCP.
1358
1359To configure bonding for DHCP, configure it as described
1360above, except replace the line "BOOTPROTO=none" with "BOOTPROTO=dhcp"
1361and add a line consisting of "TYPE=Bonding". Note that the TYPE value
1362is case sensitive.
1363
13643.2.2 Configuring Multiple Bonds with Initscripts
1365-------------------------------------------------
1366
1367Initscripts packages that are included with Fedora 7 and Red Hat
1368Enterprise Linux 5 support multiple bonding interfaces by simply
1369specifying the appropriate BONDING_OPTS= in ifcfg-bondX where X is the
1370number of the bond. This support requires sysfs support in the kernel,
1371and a bonding driver of version 3.0.0 or later. Other configurations may
1372not support this method for specifying multiple bonding interfaces; for
1373those instances, see the "Configuring Multiple Bonds Manually" section,
1374below.
1375
13763.3 Configuring Bonding Manually with iproute2
1377-----------------------------------------------
1378
1379This section applies to distros whose network initialization
1380scripts (the sysconfig or initscripts package) do not have specific
1381knowledge of bonding. One such distro is SuSE Linux Enterprise Server
1382version 8.
1383
1384The general method for these systems is to place the bonding
1385module parameters into a config file in /etc/modprobe.d/ (as
1386appropriate for the installed distro), then add modprobe and/or
1387`ip link` commands to the system's global init script. The name of
1388the global init script differs; for sysconfig, it is
1389/etc/init.d/boot.local and for initscripts it is /etc/rc.d/rc.local.
1390
1391For example, if you wanted to make a simple bond of two e100
1392devices (presumed to be eth0 and eth1), and have it persist across
1393reboots, edit the appropriate file (/etc/init.d/boot.local or
1394/etc/rc.d/rc.local), and add the following::
1395
1396 modprobe bonding mode=balance-alb miimon=100
1397 modprobe e100
1398 ifconfig bond0 192.168.1.1 netmask 255.255.255.0 up
1399 ip link set eth0 master bond0
1400 ip link set eth1 master bond0
1401
1402Replace the example bonding module parameters and bond0
1403network configuration (IP address, netmask, etc) with the appropriate
1404values for your configuration.
1405
1406Unfortunately, this method will not provide support for the
1407ifup and ifdown scripts on the bond devices. To reload the bonding
1408configuration, it is necessary to run the initialization script, e.g.,::
1409
1410 # /etc/init.d/boot.local
1411
1412or::
1413
1414 # /etc/rc.d/rc.local
1415
1416It may be desirable in such a case to create a separate script
1417which only initializes the bonding configuration, then call that
1418separate script from within boot.local. This allows for bonding to be
1419enabled without re-running the entire global init script.
1420
1421To shut down the bonding devices, it is necessary to first
1422mark the bonding device itself as being down, then remove the
1423appropriate device driver modules. For our example above, you can do
1424the following::
1425
1426 # ifconfig bond0 down
1427 # rmmod bonding
1428 # rmmod e100
1429
1430Again, for convenience, it may be desirable to create a script
1431with these commands.
1432
1433
14343.3.1 Configuring Multiple Bonds Manually
1435-----------------------------------------
1436
1437This section contains information on configuring multiple
1438bonding devices with differing options for those systems whose network
1439initialization scripts lack support for configuring multiple bonds.
1440
1441If you require multiple bonding devices, but all with the same
1442options, you may wish to use the "max_bonds" module parameter,
1443documented above.
1444
1445To create multiple bonding devices with differing options, it is
1446preferable to use bonding parameters exported by sysfs, documented in the
1447section below.
1448
1449For versions of bonding without sysfs support, the only means to
1450provide multiple instances of bonding with differing options is to load
1451the bonding driver multiple times. Note that current versions of the
1452sysconfig network initialization scripts handle this automatically; if
1453your distro uses these scripts, no special action is needed. See the
1454section Configuring Bonding Devices, above, if you're not sure about your
1455network initialization scripts.
1456
1457To load multiple instances of the module, it is necessary to
1458specify a different name for each instance (the module loading system
1459requires that every loaded module, even multiple instances of the same
1460module, have a unique name). This is accomplished by supplying multiple
1461sets of bonding options in ``/etc/modprobe.d/*.conf``, for example::
1462
1463 alias bond0 bonding
1464 options bond0 -o bond0 mode=balance-rr miimon=100
1465
1466 alias bond1 bonding
1467 options bond1 -o bond1 mode=balance-alb miimon=50
1468
1469will load the bonding module two times. The first instance is
1470named "bond0" and creates the bond0 device in balance-rr mode with an
1471miimon of 100. The second instance is named "bond1" and creates the
1472bond1 device in balance-alb mode with an miimon of 50.
1473
1474In some circumstances (typically with older distributions),
1475the above does not work, and the second bonding instance never sees
1476its options. In that case, the second options line can be substituted
1477as follows::
1478
1479 install bond1 /sbin/modprobe --ignore-install bonding -o bond1 \
1480 mode=balance-alb miimon=50
1481
1482This may be repeated any number of times, specifying a new and
1483unique name in place of bond1 for each subsequent instance.
1484
1485It has been observed that some Red Hat supplied kernels are unable
1486to rename modules at load time (the "-o bond1" part). Attempts to pass
1487that option to modprobe will produce an "Operation not permitted" error.
1488This has been reported on some Fedora Core kernels, and has been seen on
1489RHEL 4 as well. On kernels exhibiting this problem, it will be impossible
1490to configure multiple bonds with differing parameters (as they are older
1491kernels, and also lack sysfs support).
1492
14933.4 Configuring Bonding Manually via Sysfs
1494------------------------------------------
1495
1496Starting with version 3.0.0, Channel Bonding may be configured
1497via the sysfs interface. This interface allows dynamic configuration
1498of all bonds in the system without unloading the module. It also
1499allows for adding and removing bonds at runtime. Ifenslave is no
1500longer required, though it is still supported.
1501
1502Use of the sysfs interface allows you to use multiple bonds
1503with different configurations without having to reload the module.
1504It also allows you to use multiple, differently configured bonds when
1505bonding is compiled into the kernel.
1506
1507You must have the sysfs filesystem mounted to configure
1508bonding this way. The examples in this document assume that you
1509are using the standard mount point for sysfs, e.g. /sys. If your
1510sysfs filesystem is mounted elsewhere, you will need to adjust the
1511example paths accordingly.
1512
1513Creating and Destroying Bonds
1514-----------------------------
1515To add a new bond foo::
1516
1517 # echo +foo > /sys/class/net/bonding_masters
1518
1519To remove an existing bond bar::
1520
1521 # echo -bar > /sys/class/net/bonding_masters
1522
1523To show all existing bonds::
1524
1525 # cat /sys/class/net/bonding_masters
1526
1527.. note::
1528
1529 due to 4K size limitation of sysfs files, this list may be
1530 truncated if you have more than a few hundred bonds. This is unlikely
1531 to occur under normal operating conditions.
1532
1533Adding and Removing Slaves
1534--------------------------
1535Interfaces may be enslaved to a bond using the file
1536/sys/class/net/<bond>/bonding/slaves. The semantics for this file
1537are the same as for the bonding_masters file.
1538
1539To enslave interface eth0 to bond bond0::
1540
1541 # ifconfig bond0 up
1542 # echo +eth0 > /sys/class/net/bond0/bonding/slaves
1543
1544To free slave eth0 from bond bond0::
1545
1546 # echo -eth0 > /sys/class/net/bond0/bonding/slaves
1547
1548When an interface is enslaved to a bond, symlinks between the
1549two are created in the sysfs filesystem. In this case, you would get
1550/sys/class/net/bond0/slave_eth0 pointing to /sys/class/net/eth0, and
1551/sys/class/net/eth0/master pointing to /sys/class/net/bond0.
1552
1553This means that you can tell quickly whether or not an
1554interface is enslaved by looking for the master symlink. Thus:
1555# echo -eth0 > /sys/class/net/eth0/master/bonding/slaves
1556will free eth0 from whatever bond it is enslaved to, regardless of
1557the name of the bond interface.
1558
1559Changing a Bond's Configuration
1560-------------------------------
1561Each bond may be configured individually by manipulating the
1562files located in /sys/class/net/<bond name>/bonding
1563
1564The names of these files correspond directly with the command-
1565line parameters described elsewhere in this file, and, with the
1566exception of arp_ip_target, they accept the same values. To see the
1567current setting, simply cat the appropriate file.
1568
1569A few examples will be given here; for specific usage
1570guidelines for each parameter, see the appropriate section in this
1571document.
1572
1573To configure bond0 for balance-alb mode::
1574
1575 # ifconfig bond0 down
1576 # echo 6 > /sys/class/net/bond0/bonding/mode
1577 - or -
1578 # echo balance-alb > /sys/class/net/bond0/bonding/mode
1579
1580.. note::
1581
1582 The bond interface must be down before the mode can be changed.
1583
1584To enable MII monitoring on bond0 with a 1 second interval::
1585
1586 # echo 1000 > /sys/class/net/bond0/bonding/miimon
1587
1588.. note::
1589
1590 If ARP monitoring is enabled, it will disabled when MII
1591 monitoring is enabled, and vice-versa.
1592
1593To add ARP targets::
1594
1595 # echo +192.168.0.100 > /sys/class/net/bond0/bonding/arp_ip_target
1596 # echo +192.168.0.101 > /sys/class/net/bond0/bonding/arp_ip_target
1597
1598.. note::
1599
1600 up to 16 target addresses may be specified.
1601
1602To remove an ARP target::
1603
1604 # echo -192.168.0.100 > /sys/class/net/bond0/bonding/arp_ip_target
1605
1606To configure the interval between learning packet transmits::
1607
1608 # echo 12 > /sys/class/net/bond0/bonding/lp_interval
1609
1610.. note::
1611
1612 the lp_interval is the number of seconds between instances where
1613 the bonding driver sends learning packets to each slaves peer switch. The
1614 default interval is 1 second.
1615
1616Example Configuration
1617---------------------
1618We begin with the same example that is shown in section 3.3,
1619executed with sysfs, and without using ifenslave.
1620
1621To make a simple bond of two e100 devices (presumed to be eth0
1622and eth1), and have it persist across reboots, edit the appropriate
1623file (/etc/init.d/boot.local or /etc/rc.d/rc.local), and add the
1624following::
1625
1626 modprobe bonding
1627 modprobe e100
1628 echo balance-alb > /sys/class/net/bond0/bonding/mode
1629 ifconfig bond0 192.168.1.1 netmask 255.255.255.0 up
1630 echo 100 > /sys/class/net/bond0/bonding/miimon
1631 echo +eth0 > /sys/class/net/bond0/bonding/slaves
1632 echo +eth1 > /sys/class/net/bond0/bonding/slaves
1633
1634To add a second bond, with two e1000 interfaces in
1635active-backup mode, using ARP monitoring, add the following lines to
1636your init script::
1637
1638 modprobe e1000
1639 echo +bond1 > /sys/class/net/bonding_masters
1640 echo active-backup > /sys/class/net/bond1/bonding/mode
1641 ifconfig bond1 192.168.2.1 netmask 255.255.255.0 up
1642 echo +192.168.2.100 /sys/class/net/bond1/bonding/arp_ip_target
1643 echo 2000 > /sys/class/net/bond1/bonding/arp_interval
1644 echo +eth2 > /sys/class/net/bond1/bonding/slaves
1645 echo +eth3 > /sys/class/net/bond1/bonding/slaves
1646
16473.5 Configuration with Interfaces Support
1648-----------------------------------------
1649
1650This section applies to distros which use /etc/network/interfaces file
1651to describe network interface configuration, most notably Debian and its
1652derivatives.
1653
1654The ifup and ifdown commands on Debian don't support bonding out of
1655the box. The ifenslave-2.6 package should be installed to provide bonding
1656support. Once installed, this package will provide ``bond-*`` options
1657to be used into /etc/network/interfaces.
1658
1659Note that ifenslave-2.6 package will load the bonding module and use
1660the ifenslave command when appropriate.
1661
1662Example Configurations
1663----------------------
1664
1665In /etc/network/interfaces, the following stanza will configure bond0, in
1666active-backup mode, with eth0 and eth1 as slaves::
1667
1668 auto bond0
1669 iface bond0 inet dhcp
1670 bond-slaves eth0 eth1
1671 bond-mode active-backup
1672 bond-miimon 100
1673 bond-primary eth0 eth1
1674
1675If the above configuration doesn't work, you might have a system using
1676upstart for system startup. This is most notably true for recent
1677Ubuntu versions. The following stanza in /etc/network/interfaces will
1678produce the same result on those systems::
1679
1680 auto bond0
1681 iface bond0 inet dhcp
1682 bond-slaves none
1683 bond-mode active-backup
1684 bond-miimon 100
1685
1686 auto eth0
1687 iface eth0 inet manual
1688 bond-master bond0
1689 bond-primary eth0 eth1
1690
1691 auto eth1
1692 iface eth1 inet manual
1693 bond-master bond0
1694 bond-primary eth0 eth1
1695
1696For a full list of ``bond-*`` supported options in /etc/network/interfaces and
1697some more advanced examples tailored to you particular distros, see the files in
1698/usr/share/doc/ifenslave-2.6.
1699
17003.6 Overriding Configuration for Special Cases
1701----------------------------------------------
1702
1703When using the bonding driver, the physical port which transmits a frame is
1704typically selected by the bonding driver, and is not relevant to the user or
1705system administrator. The output port is simply selected using the policies of
1706the selected bonding mode. On occasion however, it is helpful to direct certain
1707classes of traffic to certain physical interfaces on output to implement
1708slightly more complex policies. For example, to reach a web server over a
1709bonded interface in which eth0 connects to a private network, while eth1
1710connects via a public network, it may be desirous to bias the bond to send said
1711traffic over eth0 first, using eth1 only as a fall back, while all other traffic
1712can safely be sent over either interface. Such configurations may be achieved
1713using the traffic control utilities inherent in linux.
1714
1715By default the bonding driver is multiqueue aware and 16 queues are created
1716when the driver initializes (see Documentation/networking/multiqueue.rst
1717for details). If more or less queues are desired the module parameter
1718tx_queues can be used to change this value. There is no sysfs parameter
1719available as the allocation is done at module init time.
1720
1721The output of the file /proc/net/bonding/bondX has changed so the output Queue
1722ID is now printed for each slave::
1723
1724 Bonding Mode: fault-tolerance (active-backup)
1725 Primary Slave: None
1726 Currently Active Slave: eth0
1727 MII Status: up
1728 MII Polling Interval (ms): 0
1729 Up Delay (ms): 0
1730 Down Delay (ms): 0
1731
1732 Slave Interface: eth0
1733 MII Status: up
1734 Link Failure Count: 0
1735 Permanent HW addr: 00:1a:a0:12:8f:cb
1736 Slave queue ID: 0
1737
1738 Slave Interface: eth1
1739 MII Status: up
1740 Link Failure Count: 0
1741 Permanent HW addr: 00:1a:a0:12:8f:cc
1742 Slave queue ID: 2
1743
1744The queue_id for a slave can be set using the command::
1745
1746 # echo "eth1:2" > /sys/class/net/bond0/bonding/queue_id
1747
1748Any interface that needs a queue_id set should set it with multiple calls
1749like the one above until proper priorities are set for all interfaces. On
1750distributions that allow configuration via initscripts, multiple 'queue_id'
1751arguments can be added to BONDING_OPTS to set all needed slave queues.
1752
1753These queue id's can be used in conjunction with the tc utility to configure
1754a multiqueue qdisc and filters to bias certain traffic to transmit on certain
1755slave devices. For instance, say we wanted, in the above configuration to
1756force all traffic bound to 192.168.1.100 to use eth1 in the bond as its output
1757device. The following commands would accomplish this::
1758
1759 # tc qdisc add dev bond0 handle 1 root multiq
1760
1761 # tc filter add dev bond0 protocol ip parent 1: prio 1 u32 match ip \
1762 dst 192.168.1.100 action skbedit queue_mapping 2
1763
1764These commands tell the kernel to attach a multiqueue queue discipline to the
1765bond0 interface and filter traffic enqueued to it, such that packets with a dst
1766ip of 192.168.1.100 have their output queue mapping value overwritten to 2.
1767This value is then passed into the driver, causing the normal output path
1768selection policy to be overridden, selecting instead qid 2, which maps to eth1.
1769
1770Note that qid values begin at 1. Qid 0 is reserved to initiate to the driver
1771that normal output policy selection should take place. One benefit to simply
1772leaving the qid for a slave to 0 is the multiqueue awareness in the bonding
1773driver that is now present. This awareness allows tc filters to be placed on
1774slave devices as well as bond devices and the bonding driver will simply act as
1775a pass-through for selecting output queues on the slave device rather than
1776output port selection.
1777
1778This feature first appeared in bonding driver version 3.7.0 and support for
1779output slave selection was limited to round-robin and active-backup modes.
1780
17813.7 Configuring LACP for 802.3ad mode in a more secure way
1782----------------------------------------------------------
1783
1784When using 802.3ad bonding mode, the Actor (host) and Partner (switch)
1785exchange LACPDUs. These LACPDUs cannot be sniffed, because they are
1786destined to link local mac addresses (which switches/bridges are not
1787supposed to forward). However, most of the values are easily predictable
1788or are simply the machine's MAC address (which is trivially known to all
1789other hosts in the same L2). This implies that other machines in the L2
1790domain can spoof LACPDU packets from other hosts to the switch and potentially
1791cause mayhem by joining (from the point of view of the switch) another
1792machine's aggregate, thus receiving a portion of that hosts incoming
1793traffic and / or spoofing traffic from that machine themselves (potentially
1794even successfully terminating some portion of flows). Though this is not
1795a likely scenario, one could avoid this possibility by simply configuring
1796few bonding parameters:
1797
1798 (a) ad_actor_system : You can set a random mac-address that can be used for
1799 these LACPDU exchanges. The value can not be either NULL or Multicast.
1800 Also it's preferable to set the local-admin bit. Following shell code
1801 generates a random mac-address as described above::
1802
1803 # sys_mac_addr=$(printf '%02x:%02x:%02x:%02x:%02x:%02x' \
1804 $(( (RANDOM & 0xFE) | 0x02 )) \
1805 $(( RANDOM & 0xFF )) \
1806 $(( RANDOM & 0xFF )) \
1807 $(( RANDOM & 0xFF )) \
1808 $(( RANDOM & 0xFF )) \
1809 $(( RANDOM & 0xFF )))
1810 # echo $sys_mac_addr > /sys/class/net/bond0/bonding/ad_actor_system
1811
1812 (b) ad_actor_sys_prio : Randomize the system priority. The default value
1813 is 65535, but system can take the value from 1 - 65535. Following shell
1814 code generates random priority and sets it::
1815
1816 # sys_prio=$(( 1 + RANDOM + RANDOM ))
1817 # echo $sys_prio > /sys/class/net/bond0/bonding/ad_actor_sys_prio
1818
1819 (c) ad_user_port_key : Use the user portion of the port-key. The default
1820 keeps this empty. These are the upper 10 bits of the port-key and value
1821 ranges from 0 - 1023. Following shell code generates these 10 bits and
1822 sets it::
1823
1824 # usr_port_key=$(( RANDOM & 0x3FF ))
1825 # echo $usr_port_key > /sys/class/net/bond0/bonding/ad_user_port_key
1826
1827
18284 Querying Bonding Configuration
1829=================================
1830
18314.1 Bonding Configuration
1832-------------------------
1833
1834Each bonding device has a read-only file residing in the
1835/proc/net/bonding directory. The file contents include information
1836about the bonding configuration, options and state of each slave.
1837
1838For example, the contents of /proc/net/bonding/bond0 after the
1839driver is loaded with parameters of mode=0 and miimon=1000 is
1840generally as follows::
1841
1842 Ethernet Channel Bonding Driver: 2.6.1 (October 29, 2004)
1843 Bonding Mode: load balancing (round-robin)
1844 Currently Active Slave: eth0
1845 MII Status: up
1846 MII Polling Interval (ms): 1000
1847 Up Delay (ms): 0
1848 Down Delay (ms): 0
1849
1850 Slave Interface: eth1
1851 MII Status: up
1852 Link Failure Count: 1
1853
1854 Slave Interface: eth0
1855 MII Status: up
1856 Link Failure Count: 1
1857
1858The precise format and contents will change depending upon the
1859bonding configuration, state, and version of the bonding driver.
1860
18614.2 Network configuration
1862-------------------------
1863
1864The network configuration can be inspected using the ifconfig
1865command. Bonding devices will have the MASTER flag set; Bonding slave
1866devices will have the SLAVE flag set. The ifconfig output does not
1867contain information on which slaves are associated with which masters.
1868
1869In the example below, the bond0 interface is the master
1870(MASTER) while eth0 and eth1 are slaves (SLAVE). Notice all slaves of
1871bond0 have the same MAC address (HWaddr) as bond0 for all modes except
1872TLB and ALB that require a unique MAC address for each slave::
1873
1874 # /sbin/ifconfig
1875 bond0 Link encap:Ethernet HWaddr 00:C0:F0:1F:37:B4
1876 inet addr:XXX.XXX.XXX.YYY Bcast:XXX.XXX.XXX.255 Mask:255.255.252.0
1877 UP BROADCAST RUNNING MASTER MULTICAST MTU:1500 Metric:1
1878 RX packets:7224794 errors:0 dropped:0 overruns:0 frame:0
1879 TX packets:3286647 errors:1 dropped:0 overruns:1 carrier:0
1880 collisions:0 txqueuelen:0
1881
1882 eth0 Link encap:Ethernet HWaddr 00:C0:F0:1F:37:B4
1883 UP BROADCAST RUNNING SLAVE MULTICAST MTU:1500 Metric:1
1884 RX packets:3573025 errors:0 dropped:0 overruns:0 frame:0
1885 TX packets:1643167 errors:1 dropped:0 overruns:1 carrier:0
1886 collisions:0 txqueuelen:100
1887 Interrupt:10 Base address:0x1080
1888
1889 eth1 Link encap:Ethernet HWaddr 00:C0:F0:1F:37:B4
1890 UP BROADCAST RUNNING SLAVE MULTICAST MTU:1500 Metric:1
1891 RX packets:3651769 errors:0 dropped:0 overruns:0 frame:0
1892 TX packets:1643480 errors:0 dropped:0 overruns:0 carrier:0
1893 collisions:0 txqueuelen:100
1894 Interrupt:9 Base address:0x1400
1895
18965. Switch Configuration
1897=======================
1898
1899For this section, "switch" refers to whatever system the
1900bonded devices are directly connected to (i.e., where the other end of
1901the cable plugs into). This may be an actual dedicated switch device,
1902or it may be another regular system (e.g., another computer running
1903Linux),
1904
1905The active-backup, balance-tlb and balance-alb modes do not
1906require any specific configuration of the switch.
1907
1908The 802.3ad mode requires that the switch have the appropriate
1909ports configured as an 802.3ad aggregation. The precise method used
1910to configure this varies from switch to switch, but, for example, a
1911Cisco 3550 series switch requires that the appropriate ports first be
1912grouped together in a single etherchannel instance, then that
1913etherchannel is set to mode "lacp" to enable 802.3ad (instead of
1914standard EtherChannel).
1915
1916The balance-rr, balance-xor and broadcast modes generally
1917require that the switch have the appropriate ports grouped together.
1918The nomenclature for such a group differs between switches, it may be
1919called an "etherchannel" (as in the Cisco example, above), a "trunk
1920group" or some other similar variation. For these modes, each switch
1921will also have its own configuration options for the switch's transmit
1922policy to the bond. Typical choices include XOR of either the MAC or
1923IP addresses. The transmit policy of the two peers does not need to
1924match. For these three modes, the bonding mode really selects a
1925transmit policy for an EtherChannel group; all three will interoperate
1926with another EtherChannel group.
1927
1928
19296. 802.1q VLAN Support
1930======================
1931
1932It is possible to configure VLAN devices over a bond interface
1933using the 8021q driver. However, only packets coming from the 8021q
1934driver and passing through bonding will be tagged by default. Self
1935generated packets, for example, bonding's learning packets or ARP
1936packets generated by either ALB mode or the ARP monitor mechanism, are
1937tagged internally by bonding itself. As a result, bonding must
1938"learn" the VLAN IDs configured above it, and use those IDs to tag
1939self generated packets.
1940
1941For reasons of simplicity, and to support the use of adapters
1942that can do VLAN hardware acceleration offloading, the bonding
1943interface declares itself as fully hardware offloading capable, it gets
1944the add_vid/kill_vid notifications to gather the necessary
1945information, and it propagates those actions to the slaves. In case
1946of mixed adapter types, hardware accelerated tagged packets that
1947should go through an adapter that is not offloading capable are
1948"un-accelerated" by the bonding driver so the VLAN tag sits in the
1949regular location.
1950
1951VLAN interfaces *must* be added on top of a bonding interface
1952only after enslaving at least one slave. The bonding interface has a
1953hardware address of 00:00:00:00:00:00 until the first slave is added.
1954If the VLAN interface is created prior to the first enslavement, it
1955would pick up the all-zeroes hardware address. Once the first slave
1956is attached to the bond, the bond device itself will pick up the
1957slave's hardware address, which is then available for the VLAN device.
1958
1959Also, be aware that a similar problem can occur if all slaves
1960are released from a bond that still has one or more VLAN interfaces on
1961top of it. When a new slave is added, the bonding interface will
1962obtain its hardware address from the first slave, which might not
1963match the hardware address of the VLAN interfaces (which was
1964ultimately copied from an earlier slave).
1965
1966There are two methods to insure that the VLAN device operates
1967with the correct hardware address if all slaves are removed from a
1968bond interface:
1969
19701. Remove all VLAN interfaces then recreate them
1971
19722. Set the bonding interface's hardware address so that it
1973matches the hardware address of the VLAN interfaces.
1974
1975Note that changing a VLAN interface's HW address would set the
1976underlying device -- i.e. the bonding interface -- to promiscuous
1977mode, which might not be what you want.
1978
1979
19807. Link Monitoring
1981==================
1982
1983The bonding driver at present supports two schemes for
1984monitoring a slave device's link state: the ARP monitor and the MII
1985monitor.
1986
1987At the present time, due to implementation restrictions in the
1988bonding driver itself, it is not possible to enable both ARP and MII
1989monitoring simultaneously.
1990
19917.1 ARP Monitor Operation
1992-------------------------
1993
1994The ARP monitor operates as its name suggests: it sends ARP
1995queries to one or more designated peer systems on the network, and
1996uses the response as an indication that the link is operating. This
1997gives some assurance that traffic is actually flowing to and from one
1998or more peers on the local network.
1999
20007.2 Configuring Multiple ARP Targets
2001------------------------------------
2002
2003While ARP monitoring can be done with just one target, it can
2004be useful in a High Availability setup to have several targets to
2005monitor. In the case of just one target, the target itself may go
2006down or have a problem making it unresponsive to ARP requests. Having
2007an additional target (or several) increases the reliability of the ARP
2008monitoring.
2009
2010Multiple ARP targets must be separated by commas as follows::
2011
2012 # example options for ARP monitoring with three targets
2013 alias bond0 bonding
2014 options bond0 arp_interval=60 arp_ip_target=192.168.0.1,192.168.0.3,192.168.0.9
2015
2016For just a single target the options would resemble::
2017
2018 # example options for ARP monitoring with one target
2019 alias bond0 bonding
2020 options bond0 arp_interval=60 arp_ip_target=192.168.0.100
2021
2022
20237.3 MII Monitor Operation
2024-------------------------
2025
2026The MII monitor monitors only the carrier state of the local
2027network interface. It accomplishes this in one of three ways: by
2028depending upon the device driver to maintain its carrier state, by
2029querying the device's MII registers, or by making an ethtool query to
2030the device.
2031
2032If the use_carrier module parameter is 1 (the default value),
2033then the MII monitor will rely on the driver for carrier state
2034information (via the netif_carrier subsystem). As explained in the
2035use_carrier parameter information, above, if the MII monitor fails to
2036detect carrier loss on the device (e.g., when the cable is physically
2037disconnected), it may be that the driver does not support
2038netif_carrier.
2039
2040If use_carrier is 0, then the MII monitor will first query the
2041device's (via ioctl) MII registers and check the link state. If that
2042request fails (not just that it returns carrier down), then the MII
2043monitor will make an ethtool ETHTOOL_GLINK request to attempt to obtain
2044the same information. If both methods fail (i.e., the driver either
2045does not support or had some error in processing both the MII register
2046and ethtool requests), then the MII monitor will assume the link is
2047up.
2048
20498. Potential Sources of Trouble
2050===============================
2051
20528.1 Adventures in Routing
2053-------------------------
2054
2055When bonding is configured, it is important that the slave
2056devices not have routes that supersede routes of the master (or,
2057generally, not have routes at all). For example, suppose the bonding
2058device bond0 has two slaves, eth0 and eth1, and the routing table is
2059as follows::
2060
2061 Kernel IP routing table
2062 Destination Gateway Genmask Flags MSS Window irtt Iface
2063 10.0.0.0 0.0.0.0 255.255.0.0 U 40 0 0 eth0
2064 10.0.0.0 0.0.0.0 255.255.0.0 U 40 0 0 eth1
2065 10.0.0.0 0.0.0.0 255.255.0.0 U 40 0 0 bond0
2066 127.0.0.0 0.0.0.0 255.0.0.0 U 40 0 0 lo
2067
2068This routing configuration will likely still update the
2069receive/transmit times in the driver (needed by the ARP monitor), but
2070may bypass the bonding driver (because outgoing traffic to, in this
2071case, another host on network 10 would use eth0 or eth1 before bond0).
2072
2073The ARP monitor (and ARP itself) may become confused by this
2074configuration, because ARP requests (generated by the ARP monitor)
2075will be sent on one interface (bond0), but the corresponding reply
2076will arrive on a different interface (eth0). This reply looks to ARP
2077as an unsolicited ARP reply (because ARP matches replies on an
2078interface basis), and is discarded. The MII monitor is not affected
2079by the state of the routing table.
2080
2081The solution here is simply to insure that slaves do not have
2082routes of their own, and if for some reason they must, those routes do
2083not supersede routes of their master. This should generally be the
2084case, but unusual configurations or errant manual or automatic static
2085route additions may cause trouble.
2086
20878.2 Ethernet Device Renaming
2088----------------------------
2089
2090On systems with network configuration scripts that do not
2091associate physical devices directly with network interface names (so
2092that the same physical device always has the same "ethX" name), it may
2093be necessary to add some special logic to config files in
2094/etc/modprobe.d/.
2095
2096For example, given a modules.conf containing the following::
2097
2098 alias bond0 bonding
2099 options bond0 mode=some-mode miimon=50
2100 alias eth0 tg3
2101 alias eth1 tg3
2102 alias eth2 e1000
2103 alias eth3 e1000
2104
2105If neither eth0 and eth1 are slaves to bond0, then when the
2106bond0 interface comes up, the devices may end up reordered. This
2107happens because bonding is loaded first, then its slave device's
2108drivers are loaded next. Since no other drivers have been loaded,
2109when the e1000 driver loads, it will receive eth0 and eth1 for its
2110devices, but the bonding configuration tries to enslave eth2 and eth3
2111(which may later be assigned to the tg3 devices).
2112
2113Adding the following::
2114
2115 add above bonding e1000 tg3
2116
2117causes modprobe to load e1000 then tg3, in that order, when
2118bonding is loaded. This command is fully documented in the
2119modules.conf manual page.
2120
2121On systems utilizing modprobe an equivalent problem can occur.
2122In this case, the following can be added to config files in
2123/etc/modprobe.d/ as::
2124
2125 softdep bonding pre: tg3 e1000
2126
2127This will load tg3 and e1000 modules before loading the bonding one.
2128Full documentation on this can be found in the modprobe.d and modprobe
2129manual pages.
2130
21318.3. Painfully Slow Or No Failed Link Detection By Miimon
2132---------------------------------------------------------
2133
2134By default, bonding enables the use_carrier option, which
2135instructs bonding to trust the driver to maintain carrier state.
2136
2137As discussed in the options section, above, some drivers do
2138not support the netif_carrier_on/_off link state tracking system.
2139With use_carrier enabled, bonding will always see these links as up,
2140regardless of their actual state.
2141
2142Additionally, other drivers do support netif_carrier, but do
2143not maintain it in real time, e.g., only polling the link state at
2144some fixed interval. In this case, miimon will detect failures, but
2145only after some long period of time has expired. If it appears that
2146miimon is very slow in detecting link failures, try specifying
2147use_carrier=0 to see if that improves the failure detection time. If
2148it does, then it may be that the driver checks the carrier state at a
2149fixed interval, but does not cache the MII register values (so the
2150use_carrier=0 method of querying the registers directly works). If
2151use_carrier=0 does not improve the failover, then the driver may cache
2152the registers, or the problem may be elsewhere.
2153
2154Also, remember that miimon only checks for the device's
2155carrier state. It has no way to determine the state of devices on or
2156beyond other ports of a switch, or if a switch is refusing to pass
2157traffic while still maintaining carrier on.
2158
21599. SNMP agents
2160===============
2161
2162If running SNMP agents, the bonding driver should be loaded
2163before any network drivers participating in a bond. This requirement
2164is due to the interface index (ipAdEntIfIndex) being associated to
2165the first interface found with a given IP address. That is, there is
2166only one ipAdEntIfIndex for each IP address. For example, if eth0 and
2167eth1 are slaves of bond0 and the driver for eth0 is loaded before the
2168bonding driver, the interface for the IP address will be associated
2169with the eth0 interface. This configuration is shown below, the IP
2170address 192.168.1.1 has an interface index of 2 which indexes to eth0
2171in the ifDescr table (ifDescr.2).
2172
2173::
2174
2175 interfaces.ifTable.ifEntry.ifDescr.1 = lo
2176 interfaces.ifTable.ifEntry.ifDescr.2 = eth0
2177 interfaces.ifTable.ifEntry.ifDescr.3 = eth1
2178 interfaces.ifTable.ifEntry.ifDescr.4 = eth2
2179 interfaces.ifTable.ifEntry.ifDescr.5 = eth3
2180 interfaces.ifTable.ifEntry.ifDescr.6 = bond0
2181 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.10.10.10 = 5
2182 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.192.168.1.1 = 2
2183 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.74.20.94 = 4
2184 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.127.0.0.1 = 1
2185
2186This problem is avoided by loading the bonding driver before
2187any network drivers participating in a bond. Below is an example of
2188loading the bonding driver first, the IP address 192.168.1.1 is
2189correctly associated with ifDescr.2.
2190
2191 interfaces.ifTable.ifEntry.ifDescr.1 = lo
2192 interfaces.ifTable.ifEntry.ifDescr.2 = bond0
2193 interfaces.ifTable.ifEntry.ifDescr.3 = eth0
2194 interfaces.ifTable.ifEntry.ifDescr.4 = eth1
2195 interfaces.ifTable.ifEntry.ifDescr.5 = eth2
2196 interfaces.ifTable.ifEntry.ifDescr.6 = eth3
2197 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.10.10.10 = 6
2198 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.192.168.1.1 = 2
2199 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.74.20.94 = 5
2200 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.127.0.0.1 = 1
2201
2202While some distributions may not report the interface name in
2203ifDescr, the association between the IP address and IfIndex remains
2204and SNMP functions such as Interface_Scan_Next will report that
2205association.
2206
220710. Promiscuous mode
2208====================
2209
2210When running network monitoring tools, e.g., tcpdump, it is
2211common to enable promiscuous mode on the device, so that all traffic
2212is seen (instead of seeing only traffic destined for the local host).
2213The bonding driver handles promiscuous mode changes to the bonding
2214master device (e.g., bond0), and propagates the setting to the slave
2215devices.
2216
2217For the balance-rr, balance-xor, broadcast, and 802.3ad modes,
2218the promiscuous mode setting is propagated to all slaves.
2219
2220For the active-backup, balance-tlb and balance-alb modes, the
2221promiscuous mode setting is propagated only to the active slave.
2222
2223For balance-tlb mode, the active slave is the slave currently
2224receiving inbound traffic.
2225
2226For balance-alb mode, the active slave is the slave used as a
2227"primary." This slave is used for mode-specific control traffic, for
2228sending to peers that are unassigned or if the load is unbalanced.
2229
2230For the active-backup, balance-tlb and balance-alb modes, when
2231the active slave changes (e.g., due to a link failure), the
2232promiscuous setting will be propagated to the new active slave.
2233
223411. Configuring Bonding for High Availability
2235=============================================
2236
2237High Availability refers to configurations that provide
2238maximum network availability by having redundant or backup devices,
2239links or switches between the host and the rest of the world. The
2240goal is to provide the maximum availability of network connectivity
2241(i.e., the network always works), even though other configurations
2242could provide higher throughput.
2243
224411.1 High Availability in a Single Switch Topology
2245--------------------------------------------------
2246
2247If two hosts (or a host and a single switch) are directly
2248connected via multiple physical links, then there is no availability
2249penalty to optimizing for maximum bandwidth. In this case, there is
2250only one switch (or peer), so if it fails, there is no alternative
2251access to fail over to. Additionally, the bonding load balance modes
2252support link monitoring of their members, so if individual links fail,
2253the load will be rebalanced across the remaining devices.
2254
2255See Section 12, "Configuring Bonding for Maximum Throughput"
2256for information on configuring bonding with one peer device.
2257
225811.2 High Availability in a Multiple Switch Topology
2259----------------------------------------------------
2260
2261With multiple switches, the configuration of bonding and the
2262network changes dramatically. In multiple switch topologies, there is
2263a trade off between network availability and usable bandwidth.
2264
2265Below is a sample network, configured to maximize the
2266availability of the network::
2267
2268 | |
2269 |port3 port3|
2270 +-----+----+ +-----+----+
2271 | |port2 ISL port2| |
2272 | switch A +--------------------------+ switch B |
2273 | | | |
2274 +-----+----+ +-----++---+
2275 |port1 port1|
2276 | +-------+ |
2277 +-------------+ host1 +---------------+
2278 eth0 +-------+ eth1
2279
2280In this configuration, there is a link between the two
2281switches (ISL, or inter switch link), and multiple ports connecting to
2282the outside world ("port3" on each switch). There is no technical
2283reason that this could not be extended to a third switch.
2284
228511.2.1 HA Bonding Mode Selection for Multiple Switch Topology
2286-------------------------------------------------------------
2287
2288In a topology such as the example above, the active-backup and
2289broadcast modes are the only useful bonding modes when optimizing for
2290availability; the other modes require all links to terminate on the
2291same peer for them to behave rationally.
2292
2293active-backup:
2294 This is generally the preferred mode, particularly if
2295 the switches have an ISL and play together well. If the
2296 network configuration is such that one switch is specifically
2297 a backup switch (e.g., has lower capacity, higher cost, etc),
2298 then the primary option can be used to insure that the
2299 preferred link is always used when it is available.
2300
2301broadcast:
2302 This mode is really a special purpose mode, and is suitable
2303 only for very specific needs. For example, if the two
2304 switches are not connected (no ISL), and the networks beyond
2305 them are totally independent. In this case, if it is
2306 necessary for some specific one-way traffic to reach both
2307 independent networks, then the broadcast mode may be suitable.
2308
230911.2.2 HA Link Monitoring Selection for Multiple Switch Topology
2310----------------------------------------------------------------
2311
2312The choice of link monitoring ultimately depends upon your
2313switch. If the switch can reliably fail ports in response to other
2314failures, then either the MII or ARP monitors should work. For
2315example, in the above example, if the "port3" link fails at the remote
2316end, the MII monitor has no direct means to detect this. The ARP
2317monitor could be configured with a target at the remote end of port3,
2318thus detecting that failure without switch support.
2319
2320In general, however, in a multiple switch topology, the ARP
2321monitor can provide a higher level of reliability in detecting end to
2322end connectivity failures (which may be caused by the failure of any
2323individual component to pass traffic for any reason). Additionally,
2324the ARP monitor should be configured with multiple targets (at least
2325one for each switch in the network). This will insure that,
2326regardless of which switch is active, the ARP monitor has a suitable
2327target to query.
2328
2329Note, also, that of late many switches now support a functionality
2330generally referred to as "trunk failover." This is a feature of the
2331switch that causes the link state of a particular switch port to be set
2332down (or up) when the state of another switch port goes down (or up).
2333Its purpose is to propagate link failures from logically "exterior" ports
2334to the logically "interior" ports that bonding is able to monitor via
2335miimon. Availability and configuration for trunk failover varies by
2336switch, but this can be a viable alternative to the ARP monitor when using
2337suitable switches.
2338
233912. Configuring Bonding for Maximum Throughput
2340==============================================
2341
234212.1 Maximizing Throughput in a Single Switch Topology
2343------------------------------------------------------
2344
2345In a single switch configuration, the best method to maximize
2346throughput depends upon the application and network environment. The
2347various load balancing modes each have strengths and weaknesses in
2348different environments, as detailed below.
2349
2350For this discussion, we will break down the topologies into
2351two categories. Depending upon the destination of most traffic, we
2352categorize them into either "gatewayed" or "local" configurations.
2353
2354In a gatewayed configuration, the "switch" is acting primarily
2355as a router, and the majority of traffic passes through this router to
2356other networks. An example would be the following::
2357
2358
2359 +----------+ +----------+
2360 | |eth0 port1| | to other networks
2361 | Host A +---------------------+ router +------------------->
2362 | +---------------------+ | Hosts B and C are out
2363 | |eth1 port2| | here somewhere
2364 +----------+ +----------+
2365
2366The router may be a dedicated router device, or another host
2367acting as a gateway. For our discussion, the important point is that
2368the majority of traffic from Host A will pass through the router to
2369some other network before reaching its final destination.
2370
2371In a gatewayed network configuration, although Host A may
2372communicate with many other systems, all of its traffic will be sent
2373and received via one other peer on the local network, the router.
2374
2375Note that the case of two systems connected directly via
2376multiple physical links is, for purposes of configuring bonding, the
2377same as a gatewayed configuration. In that case, it happens that all
2378traffic is destined for the "gateway" itself, not some other network
2379beyond the gateway.
2380
2381In a local configuration, the "switch" is acting primarily as
2382a switch, and the majority of traffic passes through this switch to
2383reach other stations on the same network. An example would be the
2384following::
2385
2386 +----------+ +----------+ +--------+
2387 | |eth0 port1| +-------+ Host B |
2388 | Host A +------------+ switch |port3 +--------+
2389 | +------------+ | +--------+
2390 | |eth1 port2| +------------------+ Host C |
2391 +----------+ +----------+port4 +--------+
2392
2393
2394Again, the switch may be a dedicated switch device, or another
2395host acting as a gateway. For our discussion, the important point is
2396that the majority of traffic from Host A is destined for other hosts
2397on the same local network (Hosts B and C in the above example).
2398
2399In summary, in a gatewayed configuration, traffic to and from
2400the bonded device will be to the same MAC level peer on the network
2401(the gateway itself, i.e., the router), regardless of its final
2402destination. In a local configuration, traffic flows directly to and
2403from the final destinations, thus, each destination (Host B, Host C)
2404will be addressed directly by their individual MAC addresses.
2405
2406This distinction between a gatewayed and a local network
2407configuration is important because many of the load balancing modes
2408available use the MAC addresses of the local network source and
2409destination to make load balancing decisions. The behavior of each
2410mode is described below.
2411
2412
241312.1.1 MT Bonding Mode Selection for Single Switch Topology
2414-----------------------------------------------------------
2415
2416This configuration is the easiest to set up and to understand,
2417although you will have to decide which bonding mode best suits your
2418needs. The trade offs for each mode are detailed below:
2419
2420balance-rr:
2421 This mode is the only mode that will permit a single
2422 TCP/IP connection to stripe traffic across multiple
2423 interfaces. It is therefore the only mode that will allow a
2424 single TCP/IP stream to utilize more than one interface's
2425 worth of throughput. This comes at a cost, however: the
2426 striping generally results in peer systems receiving packets out
2427 of order, causing TCP/IP's congestion control system to kick
2428 in, often by retransmitting segments.
2429
2430 It is possible to adjust TCP/IP's congestion limits by
2431 altering the net.ipv4.tcp_reordering sysctl parameter. The
2432 usual default value is 3. But keep in mind TCP stack is able
2433 to automatically increase this when it detects reorders.
2434
2435 Note that the fraction of packets that will be delivered out of
2436 order is highly variable, and is unlikely to be zero. The level
2437 of reordering depends upon a variety of factors, including the
2438 networking interfaces, the switch, and the topology of the
2439 configuration. Speaking in general terms, higher speed network
2440 cards produce more reordering (due to factors such as packet
2441 coalescing), and a "many to many" topology will reorder at a
2442 higher rate than a "many slow to one fast" configuration.
2443
2444 Many switches do not support any modes that stripe traffic
2445 (instead choosing a port based upon IP or MAC level addresses);
2446 for those devices, traffic for a particular connection flowing
2447 through the switch to a balance-rr bond will not utilize greater
2448 than one interface's worth of bandwidth.
2449
2450 If you are utilizing protocols other than TCP/IP, UDP for
2451 example, and your application can tolerate out of order
2452 delivery, then this mode can allow for single stream datagram
2453 performance that scales near linearly as interfaces are added
2454 to the bond.
2455
2456 This mode requires the switch to have the appropriate ports
2457 configured for "etherchannel" or "trunking."
2458
2459active-backup:
2460 There is not much advantage in this network topology to
2461 the active-backup mode, as the inactive backup devices are all
2462 connected to the same peer as the primary. In this case, a
2463 load balancing mode (with link monitoring) will provide the
2464 same level of network availability, but with increased
2465 available bandwidth. On the plus side, active-backup mode
2466 does not require any configuration of the switch, so it may
2467 have value if the hardware available does not support any of
2468 the load balance modes.
2469
2470balance-xor:
2471 This mode will limit traffic such that packets destined
2472 for specific peers will always be sent over the same
2473 interface. Since the destination is determined by the MAC
2474 addresses involved, this mode works best in a "local" network
2475 configuration (as described above), with destinations all on
2476 the same local network. This mode is likely to be suboptimal
2477 if all your traffic is passed through a single router (i.e., a
2478 "gatewayed" network configuration, as described above).
2479
2480 As with balance-rr, the switch ports need to be configured for
2481 "etherchannel" or "trunking."
2482
2483broadcast:
2484 Like active-backup, there is not much advantage to this
2485 mode in this type of network topology.
2486
2487802.3ad:
2488 This mode can be a good choice for this type of network
2489 topology. The 802.3ad mode is an IEEE standard, so all peers
2490 that implement 802.3ad should interoperate well. The 802.3ad
2491 protocol includes automatic configuration of the aggregates,
2492 so minimal manual configuration of the switch is needed
2493 (typically only to designate that some set of devices is
2494 available for 802.3ad). The 802.3ad standard also mandates
2495 that frames be delivered in order (within certain limits), so
2496 in general single connections will not see misordering of
2497 packets. The 802.3ad mode does have some drawbacks: the
2498 standard mandates that all devices in the aggregate operate at
2499 the same speed and duplex. Also, as with all bonding load
2500 balance modes other than balance-rr, no single connection will
2501 be able to utilize more than a single interface's worth of
2502 bandwidth.
2503
2504 Additionally, the linux bonding 802.3ad implementation
2505 distributes traffic by peer (using an XOR of MAC addresses
2506 and packet type ID), so in a "gatewayed" configuration, all
2507 outgoing traffic will generally use the same device. Incoming
2508 traffic may also end up on a single device, but that is
2509 dependent upon the balancing policy of the peer's 802.3ad
2510 implementation. In a "local" configuration, traffic will be
2511 distributed across the devices in the bond.
2512
2513 Finally, the 802.3ad mode mandates the use of the MII monitor,
2514 therefore, the ARP monitor is not available in this mode.
2515
2516balance-tlb:
2517 The balance-tlb mode balances outgoing traffic by peer.
2518 Since the balancing is done according to MAC address, in a
2519 "gatewayed" configuration (as described above), this mode will
2520 send all traffic across a single device. However, in a
2521 "local" network configuration, this mode balances multiple
2522 local network peers across devices in a vaguely intelligent
2523 manner (not a simple XOR as in balance-xor or 802.3ad mode),
2524 so that mathematically unlucky MAC addresses (i.e., ones that
2525 XOR to the same value) will not all "bunch up" on a single
2526 interface.
2527
2528 Unlike 802.3ad, interfaces may be of differing speeds, and no
2529 special switch configuration is required. On the down side,
2530 in this mode all incoming traffic arrives over a single
2531 interface, this mode requires certain ethtool support in the
2532 network device driver of the slave interfaces, and the ARP
2533 monitor is not available.
2534
2535balance-alb:
2536 This mode is everything that balance-tlb is, and more.
2537 It has all of the features (and restrictions) of balance-tlb,
2538 and will also balance incoming traffic from local network
2539 peers (as described in the Bonding Module Options section,
2540 above).
2541
2542 The only additional down side to this mode is that the network
2543 device driver must support changing the hardware address while
2544 the device is open.
2545
254612.1.2 MT Link Monitoring for Single Switch Topology
2547----------------------------------------------------
2548
2549The choice of link monitoring may largely depend upon which
2550mode you choose to use. The more advanced load balancing modes do not
2551support the use of the ARP monitor, and are thus restricted to using
2552the MII monitor (which does not provide as high a level of end to end
2553assurance as the ARP monitor).
2554
255512.2 Maximum Throughput in a Multiple Switch Topology
2556-----------------------------------------------------
2557
2558Multiple switches may be utilized to optimize for throughput
2559when they are configured in parallel as part of an isolated network
2560between two or more systems, for example::
2561
2562 +-----------+
2563 | Host A |
2564 +-+---+---+-+
2565 | | |
2566 +--------+ | +---------+
2567 | | |
2568 +------+---+ +-----+----+ +-----+----+
2569 | Switch A | | Switch B | | Switch C |
2570 +------+---+ +-----+----+ +-----+----+
2571 | | |
2572 +--------+ | +---------+
2573 | | |
2574 +-+---+---+-+
2575 | Host B |
2576 +-----------+
2577
2578In this configuration, the switches are isolated from one
2579another. One reason to employ a topology such as this is for an
2580isolated network with many hosts (a cluster configured for high
2581performance, for example), using multiple smaller switches can be more
2582cost effective than a single larger switch, e.g., on a network with 24
2583hosts, three 24 port switches can be significantly less expensive than
2584a single 72 port switch.
2585
2586If access beyond the network is required, an individual host
2587can be equipped with an additional network device connected to an
2588external network; this host then additionally acts as a gateway.
2589
259012.2.1 MT Bonding Mode Selection for Multiple Switch Topology
2591-------------------------------------------------------------
2592
2593In actual practice, the bonding mode typically employed in
2594configurations of this type is balance-rr. Historically, in this
2595network configuration, the usual caveats about out of order packet
2596delivery are mitigated by the use of network adapters that do not do
2597any kind of packet coalescing (via the use of NAPI, or because the
2598device itself does not generate interrupts until some number of
2599packets has arrived). When employed in this fashion, the balance-rr
2600mode allows individual connections between two hosts to effectively
2601utilize greater than one interface's bandwidth.
2602
260312.2.2 MT Link Monitoring for Multiple Switch Topology
2604------------------------------------------------------
2605
2606Again, in actual practice, the MII monitor is most often used
2607in this configuration, as performance is given preference over
2608availability. The ARP monitor will function in this topology, but its
2609advantages over the MII monitor are mitigated by the volume of probes
2610needed as the number of systems involved grows (remember that each
2611host in the network is configured with bonding).
2612
261313. Switch Behavior Issues
2614==========================
2615
261613.1 Link Establishment and Failover Delays
2617-------------------------------------------
2618
2619Some switches exhibit undesirable behavior with regard to the
2620timing of link up and down reporting by the switch.
2621
2622First, when a link comes up, some switches may indicate that
2623the link is up (carrier available), but not pass traffic over the
2624interface for some period of time. This delay is typically due to
2625some type of autonegotiation or routing protocol, but may also occur
2626during switch initialization (e.g., during recovery after a switch
2627failure). If you find this to be a problem, specify an appropriate
2628value to the updelay bonding module option to delay the use of the
2629relevant interface(s).
2630
2631Second, some switches may "bounce" the link state one or more
2632times while a link is changing state. This occurs most commonly while
2633the switch is initializing. Again, an appropriate updelay value may
2634help.
2635
2636Note that when a bonding interface has no active links, the
2637driver will immediately reuse the first link that goes up, even if the
2638updelay parameter has been specified (the updelay is ignored in this
2639case). If there are slave interfaces waiting for the updelay timeout
2640to expire, the interface that first went into that state will be
2641immediately reused. This reduces down time of the network if the
2642value of updelay has been overestimated, and since this occurs only in
2643cases with no connectivity, there is no additional penalty for
2644ignoring the updelay.
2645
2646In addition to the concerns about switch timings, if your
2647switches take a long time to go into backup mode, it may be desirable
2648to not activate a backup interface immediately after a link goes down.
2649Failover may be delayed via the downdelay bonding module option.
2650
265113.2 Duplicated Incoming Packets
2652--------------------------------
2653
2654NOTE: Starting with version 3.0.2, the bonding driver has logic to
2655suppress duplicate packets, which should largely eliminate this problem.
2656The following description is kept for reference.
2657
2658It is not uncommon to observe a short burst of duplicated
2659traffic when the bonding device is first used, or after it has been
2660idle for some period of time. This is most easily observed by issuing
2661a "ping" to some other host on the network, and noticing that the
2662output from ping flags duplicates (typically one per slave).
2663
2664For example, on a bond in active-backup mode with five slaves
2665all connected to one switch, the output may appear as follows::
2666
2667 # ping -n 10.0.4.2
2668 PING 10.0.4.2 (10.0.4.2) from 10.0.3.10 : 56(84) bytes of data.
2669 64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.7 ms
2670 64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
2671 64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
2672 64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
2673 64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
2674 64 bytes from 10.0.4.2: icmp_seq=2 ttl=64 time=0.216 ms
2675 64 bytes from 10.0.4.2: icmp_seq=3 ttl=64 time=0.267 ms
2676 64 bytes from 10.0.4.2: icmp_seq=4 ttl=64 time=0.222 ms
2677
2678This is not due to an error in the bonding driver, rather, it
2679is a side effect of how many switches update their MAC forwarding
2680tables. Initially, the switch does not associate the MAC address in
2681the packet with a particular switch port, and so it may send the
2682traffic to all ports until its MAC forwarding table is updated. Since
2683the interfaces attached to the bond may occupy multiple ports on a
2684single switch, when the switch (temporarily) floods the traffic to all
2685ports, the bond device receives multiple copies of the same packet
2686(one per slave device).
2687
2688The duplicated packet behavior is switch dependent, some
2689switches exhibit this, and some do not. On switches that display this
2690behavior, it can be induced by clearing the MAC forwarding table (on
2691most Cisco switches, the privileged command "clear mac address-table
2692dynamic" will accomplish this).
2693
269414. Hardware Specific Considerations
2695====================================
2696
2697This section contains additional information for configuring
2698bonding on specific hardware platforms, or for interfacing bonding
2699with particular switches or other devices.
2700
270114.1 IBM BladeCenter
2702--------------------
2703
2704This applies to the JS20 and similar systems.
2705
2706On the JS20 blades, the bonding driver supports only
2707balance-rr, active-backup, balance-tlb and balance-alb modes. This is
2708largely due to the network topology inside the BladeCenter, detailed
2709below.
2710
2711JS20 network adapter information
2712--------------------------------
2713
2714All JS20s come with two Broadcom Gigabit Ethernet ports
2715integrated on the planar (that's "motherboard" in IBM-speak). In the
2716BladeCenter chassis, the eth0 port of all JS20 blades is hard wired to
2717I/O Module #1; similarly, all eth1 ports are wired to I/O Module #2.
2718An add-on Broadcom daughter card can be installed on a JS20 to provide
2719two more Gigabit Ethernet ports. These ports, eth2 and eth3, are
2720wired to I/O Modules 3 and 4, respectively.
2721
2722Each I/O Module may contain either a switch or a passthrough
2723module (which allows ports to be directly connected to an external
2724switch). Some bonding modes require a specific BladeCenter internal
2725network topology in order to function; these are detailed below.
2726
2727Additional BladeCenter-specific networking information can be
2728found in two IBM Redbooks (www.ibm.com/redbooks):
2729
2730- "IBM eServer BladeCenter Networking Options"
2731- "IBM eServer BladeCenter Layer 2-7 Network Switching"
2732
2733BladeCenter networking configuration
2734------------------------------------
2735
2736Because a BladeCenter can be configured in a very large number
2737of ways, this discussion will be confined to describing basic
2738configurations.
2739
2740Normally, Ethernet Switch Modules (ESMs) are used in I/O
2741modules 1 and 2. In this configuration, the eth0 and eth1 ports of a
2742JS20 will be connected to different internal switches (in the
2743respective I/O modules).
2744
2745A passthrough module (OPM or CPM, optical or copper,
2746passthrough module) connects the I/O module directly to an external
2747switch. By using PMs in I/O module #1 and #2, the eth0 and eth1
2748interfaces of a JS20 can be redirected to the outside world and
2749connected to a common external switch.
2750
2751Depending upon the mix of ESMs and PMs, the network will
2752appear to bonding as either a single switch topology (all PMs) or as a
2753multiple switch topology (one or more ESMs, zero or more PMs). It is
2754also possible to connect ESMs together, resulting in a configuration
2755much like the example in "High Availability in a Multiple Switch
2756Topology," above.
2757
2758Requirements for specific modes
2759-------------------------------
2760
2761The balance-rr mode requires the use of passthrough modules
2762for devices in the bond, all connected to an common external switch.
2763That switch must be configured for "etherchannel" or "trunking" on the
2764appropriate ports, as is usual for balance-rr.
2765
2766The balance-alb and balance-tlb modes will function with
2767either switch modules or passthrough modules (or a mix). The only
2768specific requirement for these modes is that all network interfaces
2769must be able to reach all destinations for traffic sent over the
2770bonding device (i.e., the network must converge at some point outside
2771the BladeCenter).
2772
2773The active-backup mode has no additional requirements.
2774
2775Link monitoring issues
2776----------------------
2777
2778When an Ethernet Switch Module is in place, only the ARP
2779monitor will reliably detect link loss to an external switch. This is
2780nothing unusual, but examination of the BladeCenter cabinet would
2781suggest that the "external" network ports are the ethernet ports for
2782the system, when it fact there is a switch between these "external"
2783ports and the devices on the JS20 system itself. The MII monitor is
2784only able to detect link failures between the ESM and the JS20 system.
2785
2786When a passthrough module is in place, the MII monitor does
2787detect failures to the "external" port, which is then directly
2788connected to the JS20 system.
2789
2790Other concerns
2791--------------
2792
2793The Serial Over LAN (SoL) link is established over the primary
2794ethernet (eth0) only, therefore, any loss of link to eth0 will result
2795in losing your SoL connection. It will not fail over with other
2796network traffic, as the SoL system is beyond the control of the
2797bonding driver.
2798
2799It may be desirable to disable spanning tree on the switch
2800(either the internal Ethernet Switch Module, or an external switch) to
2801avoid fail-over delay issues when using bonding.
2802
2803
280415. Frequently Asked Questions
2805==============================
2806
28071. Is it SMP safe?
2808-------------------
2809
2810Yes. The old 2.0.xx channel bonding patch was not SMP safe.
2811The new driver was designed to be SMP safe from the start.
2812
28132. What type of cards will work with it?
2814-----------------------------------------
2815
2816Any Ethernet type cards (you can even mix cards - a Intel
2817EtherExpress PRO/100 and a 3com 3c905b, for example). For most modes,
2818devices need not be of the same speed.
2819
2820Starting with version 3.2.1, bonding also supports Infiniband
2821slaves in active-backup mode.
2822
28233. How many bonding devices can I have?
2824----------------------------------------
2825
2826There is no limit.
2827
28284. How many slaves can a bonding device have?
2829----------------------------------------------
2830
2831This is limited only by the number of network interfaces Linux
2832supports and/or the number of network cards you can place in your
2833system.
2834
28355. What happens when a slave link dies?
2836----------------------------------------
2837
2838If link monitoring is enabled, then the failing device will be
2839disabled. The active-backup mode will fail over to a backup link, and
2840other modes will ignore the failed link. The link will continue to be
2841monitored, and should it recover, it will rejoin the bond (in whatever
2842manner is appropriate for the mode). See the sections on High
2843Availability and the documentation for each mode for additional
2844information.
2845
2846Link monitoring can be enabled via either the miimon or
2847arp_interval parameters (described in the module parameters section,
2848above). In general, miimon monitors the carrier state as sensed by
2849the underlying network device, and the arp monitor (arp_interval)
2850monitors connectivity to another host on the local network.
2851
2852If no link monitoring is configured, the bonding driver will
2853be unable to detect link failures, and will assume that all links are
2854always available. This will likely result in lost packets, and a
2855resulting degradation of performance. The precise performance loss
2856depends upon the bonding mode and network configuration.
2857
28586. Can bonding be used for High Availability?
2859----------------------------------------------
2860
2861Yes. See the section on High Availability for details.
2862
28637. Which switches/systems does it work with?
2864---------------------------------------------
2865
2866The full answer to this depends upon the desired mode.
2867
2868In the basic balance modes (balance-rr and balance-xor), it
2869works with any system that supports etherchannel (also called
2870trunking). Most managed switches currently available have such
2871support, and many unmanaged switches as well.
2872
2873The advanced balance modes (balance-tlb and balance-alb) do
2874not have special switch requirements, but do need device drivers that
2875support specific features (described in the appropriate section under
2876module parameters, above).
2877
2878In 802.3ad mode, it works with systems that support IEEE
2879802.3ad Dynamic Link Aggregation. Most managed and many unmanaged
2880switches currently available support 802.3ad.
2881
2882The active-backup mode should work with any Layer-II switch.
2883
28848. Where does a bonding device get its MAC address from?
2885---------------------------------------------------------
2886
2887When using slave devices that have fixed MAC addresses, or when
2888the fail_over_mac option is enabled, the bonding device's MAC address is
2889the MAC address of the active slave.
2890
2891For other configurations, if not explicitly configured (with
2892ifconfig or ip link), the MAC address of the bonding device is taken from
2893its first slave device. This MAC address is then passed to all following
2894slaves and remains persistent (even if the first slave is removed) until
2895the bonding device is brought down or reconfigured.
2896
2897If you wish to change the MAC address, you can set it with
2898ifconfig or ip link::
2899
2900 # ifconfig bond0 hw ether 00:11:22:33:44:55
2901
2902 # ip link set bond0 address 66:77:88:99:aa:bb
2903
2904The MAC address can be also changed by bringing down/up the
2905device and then changing its slaves (or their order)::
2906
2907 # ifconfig bond0 down ; modprobe -r bonding
2908 # ifconfig bond0 .... up
2909 # ifenslave bond0 eth...
2910
2911This method will automatically take the address from the next
2912slave that is added.
2913
2914To restore your slaves' MAC addresses, you need to detach them
2915from the bond (``ifenslave -d bond0 eth0``). The bonding driver will
2916then restore the MAC addresses that the slaves had before they were
2917enslaved.
2918
29199. What bonding modes support native XDP?
2920------------------------------------------
2921
2922 * balance-rr (0)
2923 * active-backup (1)
2924 * balance-xor (2)
2925 * 802.3ad (4)
2926
2927Note that the vlan+srcmac hash policy does not support native XDP.
2928For other bonding modes, the XDP program must be loaded with generic mode.
2929
293016. Resources and Links
2931=======================
2932
2933The latest version of the bonding driver can be found in the latest
2934version of the linux kernel, found on http://kernel.org
2935
2936The latest version of this document can be found in the latest kernel
2937source (named Documentation/networking/bonding.rst).
2938
2939Discussions regarding the development of the bonding driver take place
2940on the main Linux network mailing list, hosted at vger.kernel.org. The list
2941address is:
2942
2943netdev@vger.kernel.org
2944
2945The administrative interface (to subscribe or unsubscribe) can
2946be found at:
2947
2948http://vger.kernel.org/vger-lists.html#netdev
1.. SPDX-License-Identifier: GPL-2.0
2
3===================================
4Linux Ethernet Bonding Driver HOWTO
5===================================
6
7Latest update: 27 April 2011
8
9Initial release: Thomas Davis <tadavis at lbl.gov>
10
11Corrections, HA extensions: 2000/10/03-15:
12
13 - Willy Tarreau <willy at meta-x.org>
14 - Constantine Gavrilov <const-g at xpert.com>
15 - Chad N. Tindel <ctindel at ieee dot org>
16 - Janice Girouard <girouard at us dot ibm dot com>
17 - Jay Vosburgh <fubar at us dot ibm dot com>
18
19Reorganized and updated Feb 2005 by Jay Vosburgh
20Added Sysfs information: 2006/04/24
21
22 - Mitch Williams <mitch.a.williams at intel.com>
23
24Introduction
25============
26
27The Linux bonding driver provides a method for aggregating
28multiple network interfaces into a single logical "bonded" interface.
29The behavior of the bonded interfaces depends upon the mode; generally
30speaking, modes provide either hot standby or load balancing services.
31Additionally, link integrity monitoring may be performed.
32
33The bonding driver originally came from Donald Becker's
34beowulf patches for kernel 2.0. It has changed quite a bit since, and
35the original tools from extreme-linux and beowulf sites will not work
36with this version of the driver.
37
38For new versions of the driver, updated userspace tools, and
39who to ask for help, please follow the links at the end of this file.
40
41.. Table of Contents
42
43 1. Bonding Driver Installation
44
45 2. Bonding Driver Options
46
47 3. Configuring Bonding Devices
48 3.1 Configuration with Sysconfig Support
49 3.1.1 Using DHCP with Sysconfig
50 3.1.2 Configuring Multiple Bonds with Sysconfig
51 3.2 Configuration with Initscripts Support
52 3.2.1 Using DHCP with Initscripts
53 3.2.2 Configuring Multiple Bonds with Initscripts
54 3.3 Configuring Bonding Manually with Ifenslave
55 3.3.1 Configuring Multiple Bonds Manually
56 3.4 Configuring Bonding Manually via Sysfs
57 3.5 Configuration with Interfaces Support
58 3.6 Overriding Configuration for Special Cases
59 3.7 Configuring LACP for 802.3ad mode in a more secure way
60
61 4. Querying Bonding Configuration
62 4.1 Bonding Configuration
63 4.2 Network Configuration
64
65 5. Switch Configuration
66
67 6. 802.1q VLAN Support
68
69 7. Link Monitoring
70 7.1 ARP Monitor Operation
71 7.2 Configuring Multiple ARP Targets
72 7.3 MII Monitor Operation
73
74 8. Potential Trouble Sources
75 8.1 Adventures in Routing
76 8.2 Ethernet Device Renaming
77 8.3 Painfully Slow Or No Failed Link Detection By Miimon
78
79 9. SNMP agents
80
81 10. Promiscuous mode
82
83 11. Configuring Bonding for High Availability
84 11.1 High Availability in a Single Switch Topology
85 11.2 High Availability in a Multiple Switch Topology
86 11.2.1 HA Bonding Mode Selection for Multiple Switch Topology
87 11.2.2 HA Link Monitoring for Multiple Switch Topology
88
89 12. Configuring Bonding for Maximum Throughput
90 12.1 Maximum Throughput in a Single Switch Topology
91 12.1.1 MT Bonding Mode Selection for Single Switch Topology
92 12.1.2 MT Link Monitoring for Single Switch Topology
93 12.2 Maximum Throughput in a Multiple Switch Topology
94 12.2.1 MT Bonding Mode Selection for Multiple Switch Topology
95 12.2.2 MT Link Monitoring for Multiple Switch Topology
96
97 13. Switch Behavior Issues
98 13.1 Link Establishment and Failover Delays
99 13.2 Duplicated Incoming Packets
100
101 14. Hardware Specific Considerations
102 14.1 IBM BladeCenter
103
104 15. Frequently Asked Questions
105
106 16. Resources and Links
107
108
1091. Bonding Driver Installation
110==============================
111
112Most popular distro kernels ship with the bonding driver
113already available as a module. If your distro does not, or you
114have need to compile bonding from source (e.g., configuring and
115installing a mainline kernel from kernel.org), you'll need to perform
116the following steps:
117
1181.1 Configure and build the kernel with bonding
119-----------------------------------------------
120
121The current version of the bonding driver is available in the
122drivers/net/bonding subdirectory of the most recent kernel source
123(which is available on http://kernel.org). Most users "rolling their
124own" will want to use the most recent kernel from kernel.org.
125
126Configure kernel with "make menuconfig" (or "make xconfig" or
127"make config"), then select "Bonding driver support" in the "Network
128device support" section. It is recommended that you configure the
129driver as module since it is currently the only way to pass parameters
130to the driver or configure more than one bonding device.
131
132Build and install the new kernel and modules.
133
1341.2 Bonding Control Utility
135---------------------------
136
137It is recommended to configure bonding via iproute2 (netlink)
138or sysfs, the old ifenslave control utility is obsolete.
139
1402. Bonding Driver Options
141=========================
142
143Options for the bonding driver are supplied as parameters to the
144bonding module at load time, or are specified via sysfs.
145
146Module options may be given as command line arguments to the
147insmod or modprobe command, but are usually specified in either the
148``/etc/modprobe.d/*.conf`` configuration files, or in a distro-specific
149configuration file (some of which are detailed in the next section).
150
151Details on bonding support for sysfs is provided in the
152"Configuring Bonding Manually via Sysfs" section, below.
153
154The available bonding driver parameters are listed below. If a
155parameter is not specified the default value is used. When initially
156configuring a bond, it is recommended "tail -f /var/log/messages" be
157run in a separate window to watch for bonding driver error messages.
158
159It is critical that either the miimon or arp_interval and
160arp_ip_target parameters be specified, otherwise serious network
161degradation will occur during link failures. Very few devices do not
162support at least miimon, so there is really no reason not to use it.
163
164Options with textual values will accept either the text name
165or, for backwards compatibility, the option value. E.g.,
166"mode=802.3ad" and "mode=4" set the same mode.
167
168The parameters are as follows:
169
170active_slave
171
172 Specifies the new active slave for modes that support it
173 (active-backup, balance-alb and balance-tlb). Possible values
174 are the name of any currently enslaved interface, or an empty
175 string. If a name is given, the slave and its link must be up in order
176 to be selected as the new active slave. If an empty string is
177 specified, the current active slave is cleared, and a new active
178 slave is selected automatically.
179
180 Note that this is only available through the sysfs interface. No module
181 parameter by this name exists.
182
183 The normal value of this option is the name of the currently
184 active slave, or the empty string if there is no active slave or
185 the current mode does not use an active slave.
186
187ad_actor_sys_prio
188
189 In an AD system, this specifies the system priority. The allowed range
190 is 1 - 65535. If the value is not specified, it takes 65535 as the
191 default value.
192
193 This parameter has effect only in 802.3ad mode and is available through
194 SysFs interface.
195
196ad_actor_system
197
198 In an AD system, this specifies the mac-address for the actor in
199 protocol packet exchanges (LACPDUs). The value cannot be a multicast
200 address. If the all-zeroes MAC is specified, bonding will internally
201 use the MAC of the bond itself. It is preferred to have the
202 local-admin bit set for this mac but driver does not enforce it. If
203 the value is not given then system defaults to using the masters'
204 mac address as actors' system address.
205
206 This parameter has effect only in 802.3ad mode and is available through
207 SysFs interface.
208
209ad_select
210
211 Specifies the 802.3ad aggregation selection logic to use. The
212 possible values and their effects are:
213
214 stable or 0
215
216 The active aggregator is chosen by largest aggregate
217 bandwidth.
218
219 Reselection of the active aggregator occurs only when all
220 slaves of the active aggregator are down or the active
221 aggregator has no slaves.
222
223 This is the default value.
224
225 bandwidth or 1
226
227 The active aggregator is chosen by largest aggregate
228 bandwidth. Reselection occurs if:
229
230 - A slave is added to or removed from the bond
231
232 - Any slave's link state changes
233
234 - Any slave's 802.3ad association state changes
235
236 - The bond's administrative state changes to up
237
238 count or 2
239
240 The active aggregator is chosen by the largest number of
241 ports (slaves). Reselection occurs as described under the
242 "bandwidth" setting, above.
243
244 The bandwidth and count selection policies permit failover of
245 802.3ad aggregations when partial failure of the active aggregator
246 occurs. This keeps the aggregator with the highest availability
247 (either in bandwidth or in number of ports) active at all times.
248
249 This option was added in bonding version 3.4.0.
250
251ad_user_port_key
252
253 In an AD system, the port-key has three parts as shown below -
254
255 ===== ============
256 Bits Use
257 ===== ============
258 00 Duplex
259 01-05 Speed
260 06-15 User-defined
261 ===== ============
262
263 This defines the upper 10 bits of the port key. The values can be
264 from 0 - 1023. If not given, the system defaults to 0.
265
266 This parameter has effect only in 802.3ad mode and is available through
267 SysFs interface.
268
269all_slaves_active
270
271 Specifies that duplicate frames (received on inactive ports) should be
272 dropped (0) or delivered (1).
273
274 Normally, bonding will drop duplicate frames (received on inactive
275 ports), which is desirable for most users. But there are some times
276 it is nice to allow duplicate frames to be delivered.
277
278 The default value is 0 (drop duplicate frames received on inactive
279 ports).
280
281arp_interval
282
283 Specifies the ARP link monitoring frequency in milliseconds.
284
285 The ARP monitor works by periodically checking the slave
286 devices to determine whether they have sent or received
287 traffic recently (the precise criteria depends upon the
288 bonding mode, and the state of the slave). Regular traffic is
289 generated via ARP probes issued for the addresses specified by
290 the arp_ip_target option.
291
292 This behavior can be modified by the arp_validate option,
293 below.
294
295 If ARP monitoring is used in an etherchannel compatible mode
296 (modes 0 and 2), the switch should be configured in a mode
297 that evenly distributes packets across all links. If the
298 switch is configured to distribute the packets in an XOR
299 fashion, all replies from the ARP targets will be received on
300 the same link which could cause the other team members to
301 fail. ARP monitoring should not be used in conjunction with
302 miimon. A value of 0 disables ARP monitoring. The default
303 value is 0.
304
305arp_ip_target
306
307 Specifies the IP addresses to use as ARP monitoring peers when
308 arp_interval is > 0. These are the targets of the ARP request
309 sent to determine the health of the link to the targets.
310 Specify these values in ddd.ddd.ddd.ddd format. Multiple IP
311 addresses must be separated by a comma. At least one IP
312 address must be given for ARP monitoring to function. The
313 maximum number of targets that can be specified is 16. The
314 default value is no IP addresses.
315
316ns_ip6_target
317
318 Specifies the IPv6 addresses to use as IPv6 monitoring peers when
319 arp_interval is > 0. These are the targets of the NS request
320 sent to determine the health of the link to the targets.
321 Specify these values in ffff:ffff::ffff:ffff format. Multiple IPv6
322 addresses must be separated by a comma. At least one IPv6
323 address must be given for NS/NA monitoring to function. The
324 maximum number of targets that can be specified is 16. The
325 default value is no IPv6 addresses.
326
327arp_validate
328
329 Specifies whether or not ARP probes and replies should be
330 validated in any mode that supports arp monitoring, or whether
331 non-ARP traffic should be filtered (disregarded) for link
332 monitoring purposes.
333
334 Possible values are:
335
336 none or 0
337
338 No validation or filtering is performed.
339
340 active or 1
341
342 Validation is performed only for the active slave.
343
344 backup or 2
345
346 Validation is performed only for backup slaves.
347
348 all or 3
349
350 Validation is performed for all slaves.
351
352 filter or 4
353
354 Filtering is applied to all slaves. No validation is
355 performed.
356
357 filter_active or 5
358
359 Filtering is applied to all slaves, validation is performed
360 only for the active slave.
361
362 filter_backup or 6
363
364 Filtering is applied to all slaves, validation is performed
365 only for backup slaves.
366
367 Validation:
368
369 Enabling validation causes the ARP monitor to examine the incoming
370 ARP requests and replies, and only consider a slave to be up if it
371 is receiving the appropriate ARP traffic.
372
373 For an active slave, the validation checks ARP replies to confirm
374 that they were generated by an arp_ip_target. Since backup slaves
375 do not typically receive these replies, the validation performed
376 for backup slaves is on the broadcast ARP request sent out via the
377 active slave. It is possible that some switch or network
378 configurations may result in situations wherein the backup slaves
379 do not receive the ARP requests; in such a situation, validation
380 of backup slaves must be disabled.
381
382 The validation of ARP requests on backup slaves is mainly helping
383 bonding to decide which slaves are more likely to work in case of
384 the active slave failure, it doesn't really guarantee that the
385 backup slave will work if it's selected as the next active slave.
386
387 Validation is useful in network configurations in which multiple
388 bonding hosts are concurrently issuing ARPs to one or more targets
389 beyond a common switch. Should the link between the switch and
390 target fail (but not the switch itself), the probe traffic
391 generated by the multiple bonding instances will fool the standard
392 ARP monitor into considering the links as still up. Use of
393 validation can resolve this, as the ARP monitor will only consider
394 ARP requests and replies associated with its own instance of
395 bonding.
396
397 Filtering:
398
399 Enabling filtering causes the ARP monitor to only use incoming ARP
400 packets for link availability purposes. Arriving packets that are
401 not ARPs are delivered normally, but do not count when determining
402 if a slave is available.
403
404 Filtering operates by only considering the reception of ARP
405 packets (any ARP packet, regardless of source or destination) when
406 determining if a slave has received traffic for link availability
407 purposes.
408
409 Filtering is useful in network configurations in which significant
410 levels of third party broadcast traffic would fool the standard
411 ARP monitor into considering the links as still up. Use of
412 filtering can resolve this, as only ARP traffic is considered for
413 link availability purposes.
414
415 This option was added in bonding version 3.1.0.
416
417arp_all_targets
418
419 Specifies the quantity of arp_ip_targets that must be reachable
420 in order for the ARP monitor to consider a slave as being up.
421 This option affects only active-backup mode for slaves with
422 arp_validation enabled.
423
424 Possible values are:
425
426 any or 0
427
428 consider the slave up only when any of the arp_ip_targets
429 is reachable
430
431 all or 1
432
433 consider the slave up only when all of the arp_ip_targets
434 are reachable
435
436arp_missed_max
437
438 Specifies the number of arp_interval monitor checks that must
439 fail in order for an interface to be marked down by the ARP monitor.
440
441 In order to provide orderly failover semantics, backup interfaces
442 are permitted an extra monitor check (i.e., they must fail
443 arp_missed_max + 1 times before being marked down).
444
445 The default value is 2, and the allowable range is 1 - 255.
446
447downdelay
448
449 Specifies the time, in milliseconds, to wait before disabling
450 a slave after a link failure has been detected. This option
451 is only valid for the miimon link monitor. The downdelay
452 value should be a multiple of the miimon value; if not, it
453 will be rounded down to the nearest multiple. The default
454 value is 0.
455
456fail_over_mac
457
458 Specifies whether active-backup mode should set all slaves to
459 the same MAC address at enslavement (the traditional
460 behavior), or, when enabled, perform special handling of the
461 bond's MAC address in accordance with the selected policy.
462
463 Possible values are:
464
465 none or 0
466
467 This setting disables fail_over_mac, and causes
468 bonding to set all slaves of an active-backup bond to
469 the same MAC address at enslavement time. This is the
470 default.
471
472 active or 1
473
474 The "active" fail_over_mac policy indicates that the
475 MAC address of the bond should always be the MAC
476 address of the currently active slave. The MAC
477 address of the slaves is not changed; instead, the MAC
478 address of the bond changes during a failover.
479
480 This policy is useful for devices that cannot ever
481 alter their MAC address, or for devices that refuse
482 incoming broadcasts with their own source MAC (which
483 interferes with the ARP monitor).
484
485 The down side of this policy is that every device on
486 the network must be updated via gratuitous ARP,
487 vs. just updating a switch or set of switches (which
488 often takes place for any traffic, not just ARP
489 traffic, if the switch snoops incoming traffic to
490 update its tables) for the traditional method. If the
491 gratuitous ARP is lost, communication may be
492 disrupted.
493
494 When this policy is used in conjunction with the mii
495 monitor, devices which assert link up prior to being
496 able to actually transmit and receive are particularly
497 susceptible to loss of the gratuitous ARP, and an
498 appropriate updelay setting may be required.
499
500 follow or 2
501
502 The "follow" fail_over_mac policy causes the MAC
503 address of the bond to be selected normally (normally
504 the MAC address of the first slave added to the bond).
505 However, the second and subsequent slaves are not set
506 to this MAC address while they are in a backup role; a
507 slave is programmed with the bond's MAC address at
508 failover time (and the formerly active slave receives
509 the newly active slave's MAC address).
510
511 This policy is useful for multiport devices that
512 either become confused or incur a performance penalty
513 when multiple ports are programmed with the same MAC
514 address.
515
516
517 The default policy is none, unless the first slave cannot
518 change its MAC address, in which case the active policy is
519 selected by default.
520
521 This option may be modified via sysfs only when no slaves are
522 present in the bond.
523
524 This option was added in bonding version 3.2.0. The "follow"
525 policy was added in bonding version 3.3.0.
526
527lacp_active
528 Option specifying whether to send LACPDU frames periodically.
529
530 off or 0
531 LACPDU frames acts as "speak when spoken to".
532
533 on or 1
534 LACPDU frames are sent along the configured links
535 periodically. See lacp_rate for more details.
536
537 The default is on.
538
539lacp_rate
540
541 Option specifying the rate in which we'll ask our link partner
542 to transmit LACPDU packets in 802.3ad mode. Possible values
543 are:
544
545 slow or 0
546 Request partner to transmit LACPDUs every 30 seconds
547
548 fast or 1
549 Request partner to transmit LACPDUs every 1 second
550
551 The default is slow.
552
553max_bonds
554
555 Specifies the number of bonding devices to create for this
556 instance of the bonding driver. E.g., if max_bonds is 3, and
557 the bonding driver is not already loaded, then bond0, bond1
558 and bond2 will be created. The default value is 1. Specifying
559 a value of 0 will load bonding, but will not create any devices.
560
561miimon
562
563 Specifies the MII link monitoring frequency in milliseconds.
564 This determines how often the link state of each slave is
565 inspected for link failures. A value of zero disables MII
566 link monitoring. A value of 100 is a good starting point.
567 The use_carrier option, below, affects how the link state is
568 determined. See the High Availability section for additional
569 information. The default value is 100 if arp_interval is not
570 set.
571
572min_links
573
574 Specifies the minimum number of links that must be active before
575 asserting carrier. It is similar to the Cisco EtherChannel min-links
576 feature. This allows setting the minimum number of member ports that
577 must be up (link-up state) before marking the bond device as up
578 (carrier on). This is useful for situations where higher level services
579 such as clustering want to ensure a minimum number of low bandwidth
580 links are active before switchover. This option only affect 802.3ad
581 mode.
582
583 The default value is 0. This will cause carrier to be asserted (for
584 802.3ad mode) whenever there is an active aggregator, regardless of the
585 number of available links in that aggregator. Note that, because an
586 aggregator cannot be active without at least one available link,
587 setting this option to 0 or to 1 has the exact same effect.
588
589mode
590
591 Specifies one of the bonding policies. The default is
592 balance-rr (round robin). Possible values are:
593
594 balance-rr or 0
595
596 Round-robin policy: Transmit packets in sequential
597 order from the first available slave through the
598 last. This mode provides load balancing and fault
599 tolerance.
600
601 active-backup or 1
602
603 Active-backup policy: Only one slave in the bond is
604 active. A different slave becomes active if, and only
605 if, the active slave fails. The bond's MAC address is
606 externally visible on only one port (network adapter)
607 to avoid confusing the switch.
608
609 In bonding version 2.6.2 or later, when a failover
610 occurs in active-backup mode, bonding will issue one
611 or more gratuitous ARPs on the newly active slave.
612 One gratuitous ARP is issued for the bonding master
613 interface and each VLAN interfaces configured above
614 it, provided that the interface has at least one IP
615 address configured. Gratuitous ARPs issued for VLAN
616 interfaces are tagged with the appropriate VLAN id.
617
618 This mode provides fault tolerance. The primary
619 option, documented below, affects the behavior of this
620 mode.
621
622 balance-xor or 2
623
624 XOR policy: Transmit based on the selected transmit
625 hash policy. The default policy is a simple [(source
626 MAC address XOR'd with destination MAC address XOR
627 packet type ID) modulo slave count]. Alternate transmit
628 policies may be selected via the xmit_hash_policy option,
629 described below.
630
631 This mode provides load balancing and fault tolerance.
632
633 broadcast or 3
634
635 Broadcast policy: transmits everything on all slave
636 interfaces. This mode provides fault tolerance.
637
638 802.3ad or 4
639
640 IEEE 802.3ad Dynamic link aggregation. Creates
641 aggregation groups that share the same speed and
642 duplex settings. Utilizes all slaves in the active
643 aggregator according to the 802.3ad specification.
644
645 Slave selection for outgoing traffic is done according
646 to the transmit hash policy, which may be changed from
647 the default simple XOR policy via the xmit_hash_policy
648 option, documented below. Note that not all transmit
649 policies may be 802.3ad compliant, particularly in
650 regards to the packet mis-ordering requirements of
651 section 43.2.4 of the 802.3ad standard. Differing
652 peer implementations will have varying tolerances for
653 noncompliance.
654
655 Prerequisites:
656
657 1. Ethtool support in the base drivers for retrieving
658 the speed and duplex of each slave.
659
660 2. A switch that supports IEEE 802.3ad Dynamic link
661 aggregation.
662
663 Most switches will require some type of configuration
664 to enable 802.3ad mode.
665
666 balance-tlb or 5
667
668 Adaptive transmit load balancing: channel bonding that
669 does not require any special switch support.
670
671 In tlb_dynamic_lb=1 mode; the outgoing traffic is
672 distributed according to the current load (computed
673 relative to the speed) on each slave.
674
675 In tlb_dynamic_lb=0 mode; the load balancing based on
676 current load is disabled and the load is distributed
677 only using the hash distribution.
678
679 Incoming traffic is received by the current slave.
680 If the receiving slave fails, another slave takes over
681 the MAC address of the failed receiving slave.
682
683 Prerequisite:
684
685 Ethtool support in the base drivers for retrieving the
686 speed of each slave.
687
688 balance-alb or 6
689
690 Adaptive load balancing: includes balance-tlb plus
691 receive load balancing (rlb) for IPV4 traffic, and
692 does not require any special switch support. The
693 receive load balancing is achieved by ARP negotiation.
694 The bonding driver intercepts the ARP Replies sent by
695 the local system on their way out and overwrites the
696 source hardware address with the unique hardware
697 address of one of the slaves in the bond such that
698 different peers use different hardware addresses for
699 the server.
700
701 Receive traffic from connections created by the server
702 is also balanced. When the local system sends an ARP
703 Request the bonding driver copies and saves the peer's
704 IP information from the ARP packet. When the ARP
705 Reply arrives from the peer, its hardware address is
706 retrieved and the bonding driver initiates an ARP
707 reply to this peer assigning it to one of the slaves
708 in the bond. A problematic outcome of using ARP
709 negotiation for balancing is that each time that an
710 ARP request is broadcast it uses the hardware address
711 of the bond. Hence, peers learn the hardware address
712 of the bond and the balancing of receive traffic
713 collapses to the current slave. This is handled by
714 sending updates (ARP Replies) to all the peers with
715 their individually assigned hardware address such that
716 the traffic is redistributed. Receive traffic is also
717 redistributed when a new slave is added to the bond
718 and when an inactive slave is re-activated. The
719 receive load is distributed sequentially (round robin)
720 among the group of highest speed slaves in the bond.
721
722 When a link is reconnected or a new slave joins the
723 bond the receive traffic is redistributed among all
724 active slaves in the bond by initiating ARP Replies
725 with the selected MAC address to each of the
726 clients. The updelay parameter (detailed below) must
727 be set to a value equal or greater than the switch's
728 forwarding delay so that the ARP Replies sent to the
729 peers will not be blocked by the switch.
730
731 Prerequisites:
732
733 1. Ethtool support in the base drivers for retrieving
734 the speed of each slave.
735
736 2. Base driver support for setting the hardware
737 address of a device while it is open. This is
738 required so that there will always be one slave in the
739 team using the bond hardware address (the
740 curr_active_slave) while having a unique hardware
741 address for each slave in the bond. If the
742 curr_active_slave fails its hardware address is
743 swapped with the new curr_active_slave that was
744 chosen.
745
746num_grat_arp,
747num_unsol_na
748
749 Specify the number of peer notifications (gratuitous ARPs and
750 unsolicited IPv6 Neighbor Advertisements) to be issued after a
751 failover event. As soon as the link is up on the new slave
752 (possibly immediately) a peer notification is sent on the
753 bonding device and each VLAN sub-device. This is repeated at
754 the rate specified by peer_notif_delay if the number is
755 greater than 1.
756
757 The valid range is 0 - 255; the default value is 1. These options
758 affect only the active-backup mode. These options were added for
759 bonding versions 3.3.0 and 3.4.0 respectively.
760
761 From Linux 3.0 and bonding version 3.7.1, these notifications
762 are generated by the ipv4 and ipv6 code and the numbers of
763 repetitions cannot be set independently.
764
765packets_per_slave
766
767 Specify the number of packets to transmit through a slave before
768 moving to the next one. When set to 0 then a slave is chosen at
769 random.
770
771 The valid range is 0 - 65535; the default value is 1. This option
772 has effect only in balance-rr mode.
773
774peer_notif_delay
775
776 Specify the delay, in milliseconds, between each peer
777 notification (gratuitous ARP and unsolicited IPv6 Neighbor
778 Advertisement) when they are issued after a failover event.
779 This delay should be a multiple of the link monitor interval
780 (arp_interval or miimon, whichever is active). The default
781 value is 0 which means to match the value of the link monitor
782 interval.
783
784prio
785 Slave priority. A higher number means higher priority.
786 The primary slave has the highest priority. This option also
787 follows the primary_reselect rules.
788
789 This option could only be configured via netlink, and is only valid
790 for active-backup(1), balance-tlb (5) and balance-alb (6) mode.
791 The valid value range is a signed 32 bit integer.
792
793 The default value is 0.
794
795primary
796
797 A string (eth0, eth2, etc) specifying which slave is the
798 primary device. The specified device will always be the
799 active slave while it is available. Only when the primary is
800 off-line will alternate devices be used. This is useful when
801 one slave is preferred over another, e.g., when one slave has
802 higher throughput than another.
803
804 The primary option is only valid for active-backup(1),
805 balance-tlb (5) and balance-alb (6) mode.
806
807primary_reselect
808
809 Specifies the reselection policy for the primary slave. This
810 affects how the primary slave is chosen to become the active slave
811 when failure of the active slave or recovery of the primary slave
812 occurs. This option is designed to prevent flip-flopping between
813 the primary slave and other slaves. Possible values are:
814
815 always or 0 (default)
816
817 The primary slave becomes the active slave whenever it
818 comes back up.
819
820 better or 1
821
822 The primary slave becomes the active slave when it comes
823 back up, if the speed and duplex of the primary slave is
824 better than the speed and duplex of the current active
825 slave.
826
827 failure or 2
828
829 The primary slave becomes the active slave only if the
830 current active slave fails and the primary slave is up.
831
832 The primary_reselect setting is ignored in two cases:
833
834 If no slaves are active, the first slave to recover is
835 made the active slave.
836
837 When initially enslaved, the primary slave is always made
838 the active slave.
839
840 Changing the primary_reselect policy via sysfs will cause an
841 immediate selection of the best active slave according to the new
842 policy. This may or may not result in a change of the active
843 slave, depending upon the circumstances.
844
845 This option was added for bonding version 3.6.0.
846
847tlb_dynamic_lb
848
849 Specifies if dynamic shuffling of flows is enabled in tlb
850 or alb mode. The value has no effect on any other modes.
851
852 The default behavior of tlb mode is to shuffle active flows across
853 slaves based on the load in that interval. This gives nice lb
854 characteristics but can cause packet reordering. If re-ordering is
855 a concern use this variable to disable flow shuffling and rely on
856 load balancing provided solely by the hash distribution.
857 xmit-hash-policy can be used to select the appropriate hashing for
858 the setup.
859
860 The sysfs entry can be used to change the setting per bond device
861 and the initial value is derived from the module parameter. The
862 sysfs entry is allowed to be changed only if the bond device is
863 down.
864
865 The default value is "1" that enables flow shuffling while value "0"
866 disables it. This option was added in bonding driver 3.7.1
867
868
869updelay
870
871 Specifies the time, in milliseconds, to wait before enabling a
872 slave after a link recovery has been detected. This option is
873 only valid for the miimon link monitor. The updelay value
874 should be a multiple of the miimon value; if not, it will be
875 rounded down to the nearest multiple. The default value is 0.
876
877use_carrier
878
879 Specifies whether or not miimon should use MII or ETHTOOL
880 ioctls vs. netif_carrier_ok() to determine the link
881 status. The MII or ETHTOOL ioctls are less efficient and
882 utilize a deprecated calling sequence within the kernel. The
883 netif_carrier_ok() relies on the device driver to maintain its
884 state with netif_carrier_on/off; at this writing, most, but
885 not all, device drivers support this facility.
886
887 If bonding insists that the link is up when it should not be,
888 it may be that your network device driver does not support
889 netif_carrier_on/off. The default state for netif_carrier is
890 "carrier on," so if a driver does not support netif_carrier,
891 it will appear as if the link is always up. In this case,
892 setting use_carrier to 0 will cause bonding to revert to the
893 MII / ETHTOOL ioctl method to determine the link state.
894
895 A value of 1 enables the use of netif_carrier_ok(), a value of
896 0 will use the deprecated MII / ETHTOOL ioctls. The default
897 value is 1.
898
899xmit_hash_policy
900
901 Selects the transmit hash policy to use for slave selection in
902 balance-xor, 802.3ad, and tlb modes. Possible values are:
903
904 layer2
905
906 Uses XOR of hardware MAC addresses and packet type ID
907 field to generate the hash. The formula is
908
909 hash = source MAC[5] XOR destination MAC[5] XOR packet type ID
910 slave number = hash modulo slave count
911
912 This algorithm will place all traffic to a particular
913 network peer on the same slave.
914
915 This algorithm is 802.3ad compliant.
916
917 layer2+3
918
919 This policy uses a combination of layer2 and layer3
920 protocol information to generate the hash.
921
922 Uses XOR of hardware MAC addresses and IP addresses to
923 generate the hash. The formula is
924
925 hash = source MAC[5] XOR destination MAC[5] XOR packet type ID
926 hash = hash XOR source IP XOR destination IP
927 hash = hash XOR (hash RSHIFT 16)
928 hash = hash XOR (hash RSHIFT 8)
929 And then hash is reduced modulo slave count.
930
931 If the protocol is IPv6 then the source and destination
932 addresses are first hashed using ipv6_addr_hash.
933
934 This algorithm will place all traffic to a particular
935 network peer on the same slave. For non-IP traffic,
936 the formula is the same as for the layer2 transmit
937 hash policy.
938
939 This policy is intended to provide a more balanced
940 distribution of traffic than layer2 alone, especially
941 in environments where a layer3 gateway device is
942 required to reach most destinations.
943
944 This algorithm is 802.3ad compliant.
945
946 layer3+4
947
948 This policy uses upper layer protocol information,
949 when available, to generate the hash. This allows for
950 traffic to a particular network peer to span multiple
951 slaves, although a single connection will not span
952 multiple slaves.
953
954 The formula for unfragmented TCP and UDP packets is
955
956 hash = source port, destination port (as in the header)
957 hash = hash XOR source IP XOR destination IP
958 hash = hash XOR (hash RSHIFT 16)
959 hash = hash XOR (hash RSHIFT 8)
960 hash = hash RSHIFT 1
961 And then hash is reduced modulo slave count.
962
963 If the protocol is IPv6 then the source and destination
964 addresses are first hashed using ipv6_addr_hash.
965
966 For fragmented TCP or UDP packets and all other IPv4 and
967 IPv6 protocol traffic, the source and destination port
968 information is omitted. For non-IP traffic, the
969 formula is the same as for the layer2 transmit hash
970 policy.
971
972 This algorithm is not fully 802.3ad compliant. A
973 single TCP or UDP conversation containing both
974 fragmented and unfragmented packets will see packets
975 striped across two interfaces. This may result in out
976 of order delivery. Most traffic types will not meet
977 this criteria, as TCP rarely fragments traffic, and
978 most UDP traffic is not involved in extended
979 conversations. Other implementations of 802.3ad may
980 or may not tolerate this noncompliance.
981
982 encap2+3
983
984 This policy uses the same formula as layer2+3 but it
985 relies on skb_flow_dissect to obtain the header fields
986 which might result in the use of inner headers if an
987 encapsulation protocol is used. For example this will
988 improve the performance for tunnel users because the
989 packets will be distributed according to the encapsulated
990 flows.
991
992 encap3+4
993
994 This policy uses the same formula as layer3+4 but it
995 relies on skb_flow_dissect to obtain the header fields
996 which might result in the use of inner headers if an
997 encapsulation protocol is used. For example this will
998 improve the performance for tunnel users because the
999 packets will be distributed according to the encapsulated
1000 flows.
1001
1002 vlan+srcmac
1003
1004 This policy uses a very rudimentary vlan ID and source mac
1005 hash to load-balance traffic per-vlan, with failover
1006 should one leg fail. The intended use case is for a bond
1007 shared by multiple virtual machines, all configured to
1008 use their own vlan, to give lacp-like functionality
1009 without requiring lacp-capable switching hardware.
1010
1011 The formula for the hash is simply
1012
1013 hash = (vlan ID) XOR (source MAC vendor) XOR (source MAC dev)
1014
1015 The default value is layer2. This option was added in bonding
1016 version 2.6.3. In earlier versions of bonding, this parameter
1017 does not exist, and the layer2 policy is the only policy. The
1018 layer2+3 value was added for bonding version 3.2.2.
1019
1020resend_igmp
1021
1022 Specifies the number of IGMP membership reports to be issued after
1023 a failover event. One membership report is issued immediately after
1024 the failover, subsequent packets are sent in each 200ms interval.
1025
1026 The valid range is 0 - 255; the default value is 1. A value of 0
1027 prevents the IGMP membership report from being issued in response
1028 to the failover event.
1029
1030 This option is useful for bonding modes balance-rr (0), active-backup
1031 (1), balance-tlb (5) and balance-alb (6), in which a failover can
1032 switch the IGMP traffic from one slave to another. Therefore a fresh
1033 IGMP report must be issued to cause the switch to forward the incoming
1034 IGMP traffic over the newly selected slave.
1035
1036 This option was added for bonding version 3.7.0.
1037
1038lp_interval
1039
1040 Specifies the number of seconds between instances where the bonding
1041 driver sends learning packets to each slaves peer switch.
1042
1043 The valid range is 1 - 0x7fffffff; the default value is 1. This Option
1044 has effect only in balance-tlb and balance-alb modes.
1045
10463. Configuring Bonding Devices
1047==============================
1048
1049You can configure bonding using either your distro's network
1050initialization scripts, or manually using either iproute2 or the
1051sysfs interface. Distros generally use one of three packages for the
1052network initialization scripts: initscripts, sysconfig or interfaces.
1053Recent versions of these packages have support for bonding, while older
1054versions do not.
1055
1056We will first describe the options for configuring bonding for
1057distros using versions of initscripts, sysconfig and interfaces with full
1058or partial support for bonding, then provide information on enabling
1059bonding without support from the network initialization scripts (i.e.,
1060older versions of initscripts or sysconfig).
1061
1062If you're unsure whether your distro uses sysconfig,
1063initscripts or interfaces, or don't know if it's new enough, have no fear.
1064Determining this is fairly straightforward.
1065
1066First, look for a file called interfaces in /etc/network directory.
1067If this file is present in your system, then your system use interfaces. See
1068Configuration with Interfaces Support.
1069
1070Else, issue the command::
1071
1072 $ rpm -qf /sbin/ifup
1073
1074It will respond with a line of text starting with either
1075"initscripts" or "sysconfig," followed by some numbers. This is the
1076package that provides your network initialization scripts.
1077
1078Next, to determine if your installation supports bonding,
1079issue the command::
1080
1081 $ grep ifenslave /sbin/ifup
1082
1083If this returns any matches, then your initscripts or
1084sysconfig has support for bonding.
1085
10863.1 Configuration with Sysconfig Support
1087----------------------------------------
1088
1089This section applies to distros using a version of sysconfig
1090with bonding support, for example, SuSE Linux Enterprise Server 9.
1091
1092SuSE SLES 9's networking configuration system does support
1093bonding, however, at this writing, the YaST system configuration
1094front end does not provide any means to work with bonding devices.
1095Bonding devices can be managed by hand, however, as follows.
1096
1097First, if they have not already been configured, configure the
1098slave devices. On SLES 9, this is most easily done by running the
1099yast2 sysconfig configuration utility. The goal is for to create an
1100ifcfg-id file for each slave device. The simplest way to accomplish
1101this is to configure the devices for DHCP (this is only to get the
1102file ifcfg-id file created; see below for some issues with DHCP). The
1103name of the configuration file for each device will be of the form::
1104
1105 ifcfg-id-xx:xx:xx:xx:xx:xx
1106
1107Where the "xx" portion will be replaced with the digits from
1108the device's permanent MAC address.
1109
1110Once the set of ifcfg-id-xx:xx:xx:xx:xx:xx files has been
1111created, it is necessary to edit the configuration files for the slave
1112devices (the MAC addresses correspond to those of the slave devices).
1113Before editing, the file will contain multiple lines, and will look
1114something like this::
1115
1116 BOOTPROTO='dhcp'
1117 STARTMODE='on'
1118 USERCTL='no'
1119 UNIQUE='XNzu.WeZGOGF+4wE'
1120 _nm_name='bus-pci-0001:61:01.0'
1121
1122Change the BOOTPROTO and STARTMODE lines to the following::
1123
1124 BOOTPROTO='none'
1125 STARTMODE='off'
1126
1127Do not alter the UNIQUE or _nm_name lines. Remove any other
1128lines (USERCTL, etc).
1129
1130Once the ifcfg-id-xx:xx:xx:xx:xx:xx files have been modified,
1131it's time to create the configuration file for the bonding device
1132itself. This file is named ifcfg-bondX, where X is the number of the
1133bonding device to create, starting at 0. The first such file is
1134ifcfg-bond0, the second is ifcfg-bond1, and so on. The sysconfig
1135network configuration system will correctly start multiple instances
1136of bonding.
1137
1138The contents of the ifcfg-bondX file is as follows::
1139
1140 BOOTPROTO="static"
1141 BROADCAST="10.0.2.255"
1142 IPADDR="10.0.2.10"
1143 NETMASK="255.255.0.0"
1144 NETWORK="10.0.2.0"
1145 REMOTE_IPADDR=""
1146 STARTMODE="onboot"
1147 BONDING_MASTER="yes"
1148 BONDING_MODULE_OPTS="mode=active-backup miimon=100"
1149 BONDING_SLAVE0="eth0"
1150 BONDING_SLAVE1="bus-pci-0000:06:08.1"
1151
1152Replace the sample BROADCAST, IPADDR, NETMASK and NETWORK
1153values with the appropriate values for your network.
1154
1155The STARTMODE specifies when the device is brought online.
1156The possible values are:
1157
1158 ======== ======================================================
1159 onboot The device is started at boot time. If you're not
1160 sure, this is probably what you want.
1161
1162 manual The device is started only when ifup is called
1163 manually. Bonding devices may be configured this
1164 way if you do not wish them to start automatically
1165 at boot for some reason.
1166
1167 hotplug The device is started by a hotplug event. This is not
1168 a valid choice for a bonding device.
1169
1170 off or The device configuration is ignored.
1171 ignore
1172 ======== ======================================================
1173
1174The line BONDING_MASTER='yes' indicates that the device is a
1175bonding master device. The only useful value is "yes."
1176
1177The contents of BONDING_MODULE_OPTS are supplied to the
1178instance of the bonding module for this device. Specify the options
1179for the bonding mode, link monitoring, and so on here. Do not include
1180the max_bonds bonding parameter; this will confuse the configuration
1181system if you have multiple bonding devices.
1182
1183Finally, supply one BONDING_SLAVEn="slave device" for each
1184slave. where "n" is an increasing value, one for each slave. The
1185"slave device" is either an interface name, e.g., "eth0", or a device
1186specifier for the network device. The interface name is easier to
1187find, but the ethN names are subject to change at boot time if, e.g.,
1188a device early in the sequence has failed. The device specifiers
1189(bus-pci-0000:06:08.1 in the example above) specify the physical
1190network device, and will not change unless the device's bus location
1191changes (for example, it is moved from one PCI slot to another). The
1192example above uses one of each type for demonstration purposes; most
1193configurations will choose one or the other for all slave devices.
1194
1195When all configuration files have been modified or created,
1196networking must be restarted for the configuration changes to take
1197effect. This can be accomplished via the following::
1198
1199 # /etc/init.d/network restart
1200
1201Note that the network control script (/sbin/ifdown) will
1202remove the bonding module as part of the network shutdown processing,
1203so it is not necessary to remove the module by hand if, e.g., the
1204module parameters have changed.
1205
1206Also, at this writing, YaST/YaST2 will not manage bonding
1207devices (they do not show bonding interfaces on its list of network
1208devices). It is necessary to edit the configuration file by hand to
1209change the bonding configuration.
1210
1211Additional general options and details of the ifcfg file
1212format can be found in an example ifcfg template file::
1213
1214 /etc/sysconfig/network/ifcfg.template
1215
1216Note that the template does not document the various ``BONDING_*``
1217settings described above, but does describe many of the other options.
1218
12193.1.1 Using DHCP with Sysconfig
1220-------------------------------
1221
1222Under sysconfig, configuring a device with BOOTPROTO='dhcp'
1223will cause it to query DHCP for its IP address information. At this
1224writing, this does not function for bonding devices; the scripts
1225attempt to obtain the device address from DHCP prior to adding any of
1226the slave devices. Without active slaves, the DHCP requests are not
1227sent to the network.
1228
12293.1.2 Configuring Multiple Bonds with Sysconfig
1230-----------------------------------------------
1231
1232The sysconfig network initialization system is capable of
1233handling multiple bonding devices. All that is necessary is for each
1234bonding instance to have an appropriately configured ifcfg-bondX file
1235(as described above). Do not specify the "max_bonds" parameter to any
1236instance of bonding, as this will confuse sysconfig. If you require
1237multiple bonding devices with identical parameters, create multiple
1238ifcfg-bondX files.
1239
1240Because the sysconfig scripts supply the bonding module
1241options in the ifcfg-bondX file, it is not necessary to add them to
1242the system ``/etc/modules.d/*.conf`` configuration files.
1243
12443.2 Configuration with Initscripts Support
1245------------------------------------------
1246
1247This section applies to distros using a recent version of
1248initscripts with bonding support, for example, Red Hat Enterprise Linux
1249version 3 or later, Fedora, etc. On these systems, the network
1250initialization scripts have knowledge of bonding, and can be configured to
1251control bonding devices. Note that older versions of the initscripts
1252package have lower levels of support for bonding; this will be noted where
1253applicable.
1254
1255These distros will not automatically load the network adapter
1256driver unless the ethX device is configured with an IP address.
1257Because of this constraint, users must manually configure a
1258network-script file for all physical adapters that will be members of
1259a bondX link. Network script files are located in the directory:
1260
1261/etc/sysconfig/network-scripts
1262
1263The file name must be prefixed with "ifcfg-eth" and suffixed
1264with the adapter's physical adapter number. For example, the script
1265for eth0 would be named /etc/sysconfig/network-scripts/ifcfg-eth0.
1266Place the following text in the file::
1267
1268 DEVICE=eth0
1269 USERCTL=no
1270 ONBOOT=yes
1271 MASTER=bond0
1272 SLAVE=yes
1273 BOOTPROTO=none
1274
1275The DEVICE= line will be different for every ethX device and
1276must correspond with the name of the file, i.e., ifcfg-eth1 must have
1277a device line of DEVICE=eth1. The setting of the MASTER= line will
1278also depend on the final bonding interface name chosen for your bond.
1279As with other network devices, these typically start at 0, and go up
1280one for each device, i.e., the first bonding instance is bond0, the
1281second is bond1, and so on.
1282
1283Next, create a bond network script. The file name for this
1284script will be /etc/sysconfig/network-scripts/ifcfg-bondX where X is
1285the number of the bond. For bond0 the file is named "ifcfg-bond0",
1286for bond1 it is named "ifcfg-bond1", and so on. Within that file,
1287place the following text::
1288
1289 DEVICE=bond0
1290 IPADDR=192.168.1.1
1291 NETMASK=255.255.255.0
1292 NETWORK=192.168.1.0
1293 BROADCAST=192.168.1.255
1294 ONBOOT=yes
1295 BOOTPROTO=none
1296 USERCTL=no
1297
1298Be sure to change the networking specific lines (IPADDR,
1299NETMASK, NETWORK and BROADCAST) to match your network configuration.
1300
1301For later versions of initscripts, such as that found with Fedora
13027 (or later) and Red Hat Enterprise Linux version 5 (or later), it is possible,
1303and, indeed, preferable, to specify the bonding options in the ifcfg-bond0
1304file, e.g. a line of the format::
1305
1306 BONDING_OPTS="mode=active-backup arp_interval=60 arp_ip_target=192.168.1.254"
1307
1308will configure the bond with the specified options. The options
1309specified in BONDING_OPTS are identical to the bonding module parameters
1310except for the arp_ip_target field when using versions of initscripts older
1311than and 8.57 (Fedora 8) and 8.45.19 (Red Hat Enterprise Linux 5.2). When
1312using older versions each target should be included as a separate option and
1313should be preceded by a '+' to indicate it should be added to the list of
1314queried targets, e.g.,::
1315
1316 arp_ip_target=+192.168.1.1 arp_ip_target=+192.168.1.2
1317
1318is the proper syntax to specify multiple targets. When specifying
1319options via BONDING_OPTS, it is not necessary to edit
1320``/etc/modprobe.d/*.conf``.
1321
1322For even older versions of initscripts that do not support
1323BONDING_OPTS, it is necessary to edit /etc/modprobe.d/*.conf, depending upon
1324your distro) to load the bonding module with your desired options when the
1325bond0 interface is brought up. The following lines in /etc/modprobe.d/*.conf
1326will load the bonding module, and select its options:
1327
1328 alias bond0 bonding
1329 options bond0 mode=balance-alb miimon=100
1330
1331Replace the sample parameters with the appropriate set of
1332options for your configuration.
1333
1334Finally run "/etc/rc.d/init.d/network restart" as root. This
1335will restart the networking subsystem and your bond link should be now
1336up and running.
1337
13383.2.1 Using DHCP with Initscripts
1339---------------------------------
1340
1341Recent versions of initscripts (the versions supplied with Fedora
1342Core 3 and Red Hat Enterprise Linux 4, or later versions, are reported to
1343work) have support for assigning IP information to bonding devices via
1344DHCP.
1345
1346To configure bonding for DHCP, configure it as described
1347above, except replace the line "BOOTPROTO=none" with "BOOTPROTO=dhcp"
1348and add a line consisting of "TYPE=Bonding". Note that the TYPE value
1349is case sensitive.
1350
13513.2.2 Configuring Multiple Bonds with Initscripts
1352-------------------------------------------------
1353
1354Initscripts packages that are included with Fedora 7 and Red Hat
1355Enterprise Linux 5 support multiple bonding interfaces by simply
1356specifying the appropriate BONDING_OPTS= in ifcfg-bondX where X is the
1357number of the bond. This support requires sysfs support in the kernel,
1358and a bonding driver of version 3.0.0 or later. Other configurations may
1359not support this method for specifying multiple bonding interfaces; for
1360those instances, see the "Configuring Multiple Bonds Manually" section,
1361below.
1362
13633.3 Configuring Bonding Manually with iproute2
1364-----------------------------------------------
1365
1366This section applies to distros whose network initialization
1367scripts (the sysconfig or initscripts package) do not have specific
1368knowledge of bonding. One such distro is SuSE Linux Enterprise Server
1369version 8.
1370
1371The general method for these systems is to place the bonding
1372module parameters into a config file in /etc/modprobe.d/ (as
1373appropriate for the installed distro), then add modprobe and/or
1374`ip link` commands to the system's global init script. The name of
1375the global init script differs; for sysconfig, it is
1376/etc/init.d/boot.local and for initscripts it is /etc/rc.d/rc.local.
1377
1378For example, if you wanted to make a simple bond of two e100
1379devices (presumed to be eth0 and eth1), and have it persist across
1380reboots, edit the appropriate file (/etc/init.d/boot.local or
1381/etc/rc.d/rc.local), and add the following::
1382
1383 modprobe bonding mode=balance-alb miimon=100
1384 modprobe e100
1385 ifconfig bond0 192.168.1.1 netmask 255.255.255.0 up
1386 ip link set eth0 master bond0
1387 ip link set eth1 master bond0
1388
1389Replace the example bonding module parameters and bond0
1390network configuration (IP address, netmask, etc) with the appropriate
1391values for your configuration.
1392
1393Unfortunately, this method will not provide support for the
1394ifup and ifdown scripts on the bond devices. To reload the bonding
1395configuration, it is necessary to run the initialization script, e.g.,::
1396
1397 # /etc/init.d/boot.local
1398
1399or::
1400
1401 # /etc/rc.d/rc.local
1402
1403It may be desirable in such a case to create a separate script
1404which only initializes the bonding configuration, then call that
1405separate script from within boot.local. This allows for bonding to be
1406enabled without re-running the entire global init script.
1407
1408To shut down the bonding devices, it is necessary to first
1409mark the bonding device itself as being down, then remove the
1410appropriate device driver modules. For our example above, you can do
1411the following::
1412
1413 # ifconfig bond0 down
1414 # rmmod bonding
1415 # rmmod e100
1416
1417Again, for convenience, it may be desirable to create a script
1418with these commands.
1419
1420
14213.3.1 Configuring Multiple Bonds Manually
1422-----------------------------------------
1423
1424This section contains information on configuring multiple
1425bonding devices with differing options for those systems whose network
1426initialization scripts lack support for configuring multiple bonds.
1427
1428If you require multiple bonding devices, but all with the same
1429options, you may wish to use the "max_bonds" module parameter,
1430documented above.
1431
1432To create multiple bonding devices with differing options, it is
1433preferable to use bonding parameters exported by sysfs, documented in the
1434section below.
1435
1436For versions of bonding without sysfs support, the only means to
1437provide multiple instances of bonding with differing options is to load
1438the bonding driver multiple times. Note that current versions of the
1439sysconfig network initialization scripts handle this automatically; if
1440your distro uses these scripts, no special action is needed. See the
1441section Configuring Bonding Devices, above, if you're not sure about your
1442network initialization scripts.
1443
1444To load multiple instances of the module, it is necessary to
1445specify a different name for each instance (the module loading system
1446requires that every loaded module, even multiple instances of the same
1447module, have a unique name). This is accomplished by supplying multiple
1448sets of bonding options in ``/etc/modprobe.d/*.conf``, for example::
1449
1450 alias bond0 bonding
1451 options bond0 -o bond0 mode=balance-rr miimon=100
1452
1453 alias bond1 bonding
1454 options bond1 -o bond1 mode=balance-alb miimon=50
1455
1456will load the bonding module two times. The first instance is
1457named "bond0" and creates the bond0 device in balance-rr mode with an
1458miimon of 100. The second instance is named "bond1" and creates the
1459bond1 device in balance-alb mode with an miimon of 50.
1460
1461In some circumstances (typically with older distributions),
1462the above does not work, and the second bonding instance never sees
1463its options. In that case, the second options line can be substituted
1464as follows::
1465
1466 install bond1 /sbin/modprobe --ignore-install bonding -o bond1 \
1467 mode=balance-alb miimon=50
1468
1469This may be repeated any number of times, specifying a new and
1470unique name in place of bond1 for each subsequent instance.
1471
1472It has been observed that some Red Hat supplied kernels are unable
1473to rename modules at load time (the "-o bond1" part). Attempts to pass
1474that option to modprobe will produce an "Operation not permitted" error.
1475This has been reported on some Fedora Core kernels, and has been seen on
1476RHEL 4 as well. On kernels exhibiting this problem, it will be impossible
1477to configure multiple bonds with differing parameters (as they are older
1478kernels, and also lack sysfs support).
1479
14803.4 Configuring Bonding Manually via Sysfs
1481------------------------------------------
1482
1483Starting with version 3.0.0, Channel Bonding may be configured
1484via the sysfs interface. This interface allows dynamic configuration
1485of all bonds in the system without unloading the module. It also
1486allows for adding and removing bonds at runtime. Ifenslave is no
1487longer required, though it is still supported.
1488
1489Use of the sysfs interface allows you to use multiple bonds
1490with different configurations without having to reload the module.
1491It also allows you to use multiple, differently configured bonds when
1492bonding is compiled into the kernel.
1493
1494You must have the sysfs filesystem mounted to configure
1495bonding this way. The examples in this document assume that you
1496are using the standard mount point for sysfs, e.g. /sys. If your
1497sysfs filesystem is mounted elsewhere, you will need to adjust the
1498example paths accordingly.
1499
1500Creating and Destroying Bonds
1501-----------------------------
1502To add a new bond foo::
1503
1504 # echo +foo > /sys/class/net/bonding_masters
1505
1506To remove an existing bond bar::
1507
1508 # echo -bar > /sys/class/net/bonding_masters
1509
1510To show all existing bonds::
1511
1512 # cat /sys/class/net/bonding_masters
1513
1514.. note::
1515
1516 due to 4K size limitation of sysfs files, this list may be
1517 truncated if you have more than a few hundred bonds. This is unlikely
1518 to occur under normal operating conditions.
1519
1520Adding and Removing Slaves
1521--------------------------
1522Interfaces may be enslaved to a bond using the file
1523/sys/class/net/<bond>/bonding/slaves. The semantics for this file
1524are the same as for the bonding_masters file.
1525
1526To enslave interface eth0 to bond bond0::
1527
1528 # ifconfig bond0 up
1529 # echo +eth0 > /sys/class/net/bond0/bonding/slaves
1530
1531To free slave eth0 from bond bond0::
1532
1533 # echo -eth0 > /sys/class/net/bond0/bonding/slaves
1534
1535When an interface is enslaved to a bond, symlinks between the
1536two are created in the sysfs filesystem. In this case, you would get
1537/sys/class/net/bond0/slave_eth0 pointing to /sys/class/net/eth0, and
1538/sys/class/net/eth0/master pointing to /sys/class/net/bond0.
1539
1540This means that you can tell quickly whether or not an
1541interface is enslaved by looking for the master symlink. Thus:
1542# echo -eth0 > /sys/class/net/eth0/master/bonding/slaves
1543will free eth0 from whatever bond it is enslaved to, regardless of
1544the name of the bond interface.
1545
1546Changing a Bond's Configuration
1547-------------------------------
1548Each bond may be configured individually by manipulating the
1549files located in /sys/class/net/<bond name>/bonding
1550
1551The names of these files correspond directly with the command-
1552line parameters described elsewhere in this file, and, with the
1553exception of arp_ip_target, they accept the same values. To see the
1554current setting, simply cat the appropriate file.
1555
1556A few examples will be given here; for specific usage
1557guidelines for each parameter, see the appropriate section in this
1558document.
1559
1560To configure bond0 for balance-alb mode::
1561
1562 # ifconfig bond0 down
1563 # echo 6 > /sys/class/net/bond0/bonding/mode
1564 - or -
1565 # echo balance-alb > /sys/class/net/bond0/bonding/mode
1566
1567.. note::
1568
1569 The bond interface must be down before the mode can be changed.
1570
1571To enable MII monitoring on bond0 with a 1 second interval::
1572
1573 # echo 1000 > /sys/class/net/bond0/bonding/miimon
1574
1575.. note::
1576
1577 If ARP monitoring is enabled, it will disabled when MII
1578 monitoring is enabled, and vice-versa.
1579
1580To add ARP targets::
1581
1582 # echo +192.168.0.100 > /sys/class/net/bond0/bonding/arp_ip_target
1583 # echo +192.168.0.101 > /sys/class/net/bond0/bonding/arp_ip_target
1584
1585.. note::
1586
1587 up to 16 target addresses may be specified.
1588
1589To remove an ARP target::
1590
1591 # echo -192.168.0.100 > /sys/class/net/bond0/bonding/arp_ip_target
1592
1593To configure the interval between learning packet transmits::
1594
1595 # echo 12 > /sys/class/net/bond0/bonding/lp_interval
1596
1597.. note::
1598
1599 the lp_interval is the number of seconds between instances where
1600 the bonding driver sends learning packets to each slaves peer switch. The
1601 default interval is 1 second.
1602
1603Example Configuration
1604---------------------
1605We begin with the same example that is shown in section 3.3,
1606executed with sysfs, and without using ifenslave.
1607
1608To make a simple bond of two e100 devices (presumed to be eth0
1609and eth1), and have it persist across reboots, edit the appropriate
1610file (/etc/init.d/boot.local or /etc/rc.d/rc.local), and add the
1611following::
1612
1613 modprobe bonding
1614 modprobe e100
1615 echo balance-alb > /sys/class/net/bond0/bonding/mode
1616 ifconfig bond0 192.168.1.1 netmask 255.255.255.0 up
1617 echo 100 > /sys/class/net/bond0/bonding/miimon
1618 echo +eth0 > /sys/class/net/bond0/bonding/slaves
1619 echo +eth1 > /sys/class/net/bond0/bonding/slaves
1620
1621To add a second bond, with two e1000 interfaces in
1622active-backup mode, using ARP monitoring, add the following lines to
1623your init script::
1624
1625 modprobe e1000
1626 echo +bond1 > /sys/class/net/bonding_masters
1627 echo active-backup > /sys/class/net/bond1/bonding/mode
1628 ifconfig bond1 192.168.2.1 netmask 255.255.255.0 up
1629 echo +192.168.2.100 /sys/class/net/bond1/bonding/arp_ip_target
1630 echo 2000 > /sys/class/net/bond1/bonding/arp_interval
1631 echo +eth2 > /sys/class/net/bond1/bonding/slaves
1632 echo +eth3 > /sys/class/net/bond1/bonding/slaves
1633
16343.5 Configuration with Interfaces Support
1635-----------------------------------------
1636
1637This section applies to distros which use /etc/network/interfaces file
1638to describe network interface configuration, most notably Debian and it's
1639derivatives.
1640
1641The ifup and ifdown commands on Debian don't support bonding out of
1642the box. The ifenslave-2.6 package should be installed to provide bonding
1643support. Once installed, this package will provide ``bond-*`` options
1644to be used into /etc/network/interfaces.
1645
1646Note that ifenslave-2.6 package will load the bonding module and use
1647the ifenslave command when appropriate.
1648
1649Example Configurations
1650----------------------
1651
1652In /etc/network/interfaces, the following stanza will configure bond0, in
1653active-backup mode, with eth0 and eth1 as slaves::
1654
1655 auto bond0
1656 iface bond0 inet dhcp
1657 bond-slaves eth0 eth1
1658 bond-mode active-backup
1659 bond-miimon 100
1660 bond-primary eth0 eth1
1661
1662If the above configuration doesn't work, you might have a system using
1663upstart for system startup. This is most notably true for recent
1664Ubuntu versions. The following stanza in /etc/network/interfaces will
1665produce the same result on those systems::
1666
1667 auto bond0
1668 iface bond0 inet dhcp
1669 bond-slaves none
1670 bond-mode active-backup
1671 bond-miimon 100
1672
1673 auto eth0
1674 iface eth0 inet manual
1675 bond-master bond0
1676 bond-primary eth0 eth1
1677
1678 auto eth1
1679 iface eth1 inet manual
1680 bond-master bond0
1681 bond-primary eth0 eth1
1682
1683For a full list of ``bond-*`` supported options in /etc/network/interfaces and
1684some more advanced examples tailored to you particular distros, see the files in
1685/usr/share/doc/ifenslave-2.6.
1686
16873.6 Overriding Configuration for Special Cases
1688----------------------------------------------
1689
1690When using the bonding driver, the physical port which transmits a frame is
1691typically selected by the bonding driver, and is not relevant to the user or
1692system administrator. The output port is simply selected using the policies of
1693the selected bonding mode. On occasion however, it is helpful to direct certain
1694classes of traffic to certain physical interfaces on output to implement
1695slightly more complex policies. For example, to reach a web server over a
1696bonded interface in which eth0 connects to a private network, while eth1
1697connects via a public network, it may be desirous to bias the bond to send said
1698traffic over eth0 first, using eth1 only as a fall back, while all other traffic
1699can safely be sent over either interface. Such configurations may be achieved
1700using the traffic control utilities inherent in linux.
1701
1702By default the bonding driver is multiqueue aware and 16 queues are created
1703when the driver initializes (see Documentation/networking/multiqueue.rst
1704for details). If more or less queues are desired the module parameter
1705tx_queues can be used to change this value. There is no sysfs parameter
1706available as the allocation is done at module init time.
1707
1708The output of the file /proc/net/bonding/bondX has changed so the output Queue
1709ID is now printed for each slave::
1710
1711 Bonding Mode: fault-tolerance (active-backup)
1712 Primary Slave: None
1713 Currently Active Slave: eth0
1714 MII Status: up
1715 MII Polling Interval (ms): 0
1716 Up Delay (ms): 0
1717 Down Delay (ms): 0
1718
1719 Slave Interface: eth0
1720 MII Status: up
1721 Link Failure Count: 0
1722 Permanent HW addr: 00:1a:a0:12:8f:cb
1723 Slave queue ID: 0
1724
1725 Slave Interface: eth1
1726 MII Status: up
1727 Link Failure Count: 0
1728 Permanent HW addr: 00:1a:a0:12:8f:cc
1729 Slave queue ID: 2
1730
1731The queue_id for a slave can be set using the command::
1732
1733 # echo "eth1:2" > /sys/class/net/bond0/bonding/queue_id
1734
1735Any interface that needs a queue_id set should set it with multiple calls
1736like the one above until proper priorities are set for all interfaces. On
1737distributions that allow configuration via initscripts, multiple 'queue_id'
1738arguments can be added to BONDING_OPTS to set all needed slave queues.
1739
1740These queue id's can be used in conjunction with the tc utility to configure
1741a multiqueue qdisc and filters to bias certain traffic to transmit on certain
1742slave devices. For instance, say we wanted, in the above configuration to
1743force all traffic bound to 192.168.1.100 to use eth1 in the bond as its output
1744device. The following commands would accomplish this::
1745
1746 # tc qdisc add dev bond0 handle 1 root multiq
1747
1748 # tc filter add dev bond0 protocol ip parent 1: prio 1 u32 match ip \
1749 dst 192.168.1.100 action skbedit queue_mapping 2
1750
1751These commands tell the kernel to attach a multiqueue queue discipline to the
1752bond0 interface and filter traffic enqueued to it, such that packets with a dst
1753ip of 192.168.1.100 have their output queue mapping value overwritten to 2.
1754This value is then passed into the driver, causing the normal output path
1755selection policy to be overridden, selecting instead qid 2, which maps to eth1.
1756
1757Note that qid values begin at 1. Qid 0 is reserved to initiate to the driver
1758that normal output policy selection should take place. One benefit to simply
1759leaving the qid for a slave to 0 is the multiqueue awareness in the bonding
1760driver that is now present. This awareness allows tc filters to be placed on
1761slave devices as well as bond devices and the bonding driver will simply act as
1762a pass-through for selecting output queues on the slave device rather than
1763output port selection.
1764
1765This feature first appeared in bonding driver version 3.7.0 and support for
1766output slave selection was limited to round-robin and active-backup modes.
1767
17683.7 Configuring LACP for 802.3ad mode in a more secure way
1769----------------------------------------------------------
1770
1771When using 802.3ad bonding mode, the Actor (host) and Partner (switch)
1772exchange LACPDUs. These LACPDUs cannot be sniffed, because they are
1773destined to link local mac addresses (which switches/bridges are not
1774supposed to forward). However, most of the values are easily predictable
1775or are simply the machine's MAC address (which is trivially known to all
1776other hosts in the same L2). This implies that other machines in the L2
1777domain can spoof LACPDU packets from other hosts to the switch and potentially
1778cause mayhem by joining (from the point of view of the switch) another
1779machine's aggregate, thus receiving a portion of that hosts incoming
1780traffic and / or spoofing traffic from that machine themselves (potentially
1781even successfully terminating some portion of flows). Though this is not
1782a likely scenario, one could avoid this possibility by simply configuring
1783few bonding parameters:
1784
1785 (a) ad_actor_system : You can set a random mac-address that can be used for
1786 these LACPDU exchanges. The value can not be either NULL or Multicast.
1787 Also it's preferable to set the local-admin bit. Following shell code
1788 generates a random mac-address as described above::
1789
1790 # sys_mac_addr=$(printf '%02x:%02x:%02x:%02x:%02x:%02x' \
1791 $(( (RANDOM & 0xFE) | 0x02 )) \
1792 $(( RANDOM & 0xFF )) \
1793 $(( RANDOM & 0xFF )) \
1794 $(( RANDOM & 0xFF )) \
1795 $(( RANDOM & 0xFF )) \
1796 $(( RANDOM & 0xFF )))
1797 # echo $sys_mac_addr > /sys/class/net/bond0/bonding/ad_actor_system
1798
1799 (b) ad_actor_sys_prio : Randomize the system priority. The default value
1800 is 65535, but system can take the value from 1 - 65535. Following shell
1801 code generates random priority and sets it::
1802
1803 # sys_prio=$(( 1 + RANDOM + RANDOM ))
1804 # echo $sys_prio > /sys/class/net/bond0/bonding/ad_actor_sys_prio
1805
1806 (c) ad_user_port_key : Use the user portion of the port-key. The default
1807 keeps this empty. These are the upper 10 bits of the port-key and value
1808 ranges from 0 - 1023. Following shell code generates these 10 bits and
1809 sets it::
1810
1811 # usr_port_key=$(( RANDOM & 0x3FF ))
1812 # echo $usr_port_key > /sys/class/net/bond0/bonding/ad_user_port_key
1813
1814
18154 Querying Bonding Configuration
1816=================================
1817
18184.1 Bonding Configuration
1819-------------------------
1820
1821Each bonding device has a read-only file residing in the
1822/proc/net/bonding directory. The file contents include information
1823about the bonding configuration, options and state of each slave.
1824
1825For example, the contents of /proc/net/bonding/bond0 after the
1826driver is loaded with parameters of mode=0 and miimon=1000 is
1827generally as follows::
1828
1829 Ethernet Channel Bonding Driver: 2.6.1 (October 29, 2004)
1830 Bonding Mode: load balancing (round-robin)
1831 Currently Active Slave: eth0
1832 MII Status: up
1833 MII Polling Interval (ms): 1000
1834 Up Delay (ms): 0
1835 Down Delay (ms): 0
1836
1837 Slave Interface: eth1
1838 MII Status: up
1839 Link Failure Count: 1
1840
1841 Slave Interface: eth0
1842 MII Status: up
1843 Link Failure Count: 1
1844
1845The precise format and contents will change depending upon the
1846bonding configuration, state, and version of the bonding driver.
1847
18484.2 Network configuration
1849-------------------------
1850
1851The network configuration can be inspected using the ifconfig
1852command. Bonding devices will have the MASTER flag set; Bonding slave
1853devices will have the SLAVE flag set. The ifconfig output does not
1854contain information on which slaves are associated with which masters.
1855
1856In the example below, the bond0 interface is the master
1857(MASTER) while eth0 and eth1 are slaves (SLAVE). Notice all slaves of
1858bond0 have the same MAC address (HWaddr) as bond0 for all modes except
1859TLB and ALB that require a unique MAC address for each slave::
1860
1861 # /sbin/ifconfig
1862 bond0 Link encap:Ethernet HWaddr 00:C0:F0:1F:37:B4
1863 inet addr:XXX.XXX.XXX.YYY Bcast:XXX.XXX.XXX.255 Mask:255.255.252.0
1864 UP BROADCAST RUNNING MASTER MULTICAST MTU:1500 Metric:1
1865 RX packets:7224794 errors:0 dropped:0 overruns:0 frame:0
1866 TX packets:3286647 errors:1 dropped:0 overruns:1 carrier:0
1867 collisions:0 txqueuelen:0
1868
1869 eth0 Link encap:Ethernet HWaddr 00:C0:F0:1F:37:B4
1870 UP BROADCAST RUNNING SLAVE MULTICAST MTU:1500 Metric:1
1871 RX packets:3573025 errors:0 dropped:0 overruns:0 frame:0
1872 TX packets:1643167 errors:1 dropped:0 overruns:1 carrier:0
1873 collisions:0 txqueuelen:100
1874 Interrupt:10 Base address:0x1080
1875
1876 eth1 Link encap:Ethernet HWaddr 00:C0:F0:1F:37:B4
1877 UP BROADCAST RUNNING SLAVE MULTICAST MTU:1500 Metric:1
1878 RX packets:3651769 errors:0 dropped:0 overruns:0 frame:0
1879 TX packets:1643480 errors:0 dropped:0 overruns:0 carrier:0
1880 collisions:0 txqueuelen:100
1881 Interrupt:9 Base address:0x1400
1882
18835. Switch Configuration
1884=======================
1885
1886For this section, "switch" refers to whatever system the
1887bonded devices are directly connected to (i.e., where the other end of
1888the cable plugs into). This may be an actual dedicated switch device,
1889or it may be another regular system (e.g., another computer running
1890Linux),
1891
1892The active-backup, balance-tlb and balance-alb modes do not
1893require any specific configuration of the switch.
1894
1895The 802.3ad mode requires that the switch have the appropriate
1896ports configured as an 802.3ad aggregation. The precise method used
1897to configure this varies from switch to switch, but, for example, a
1898Cisco 3550 series switch requires that the appropriate ports first be
1899grouped together in a single etherchannel instance, then that
1900etherchannel is set to mode "lacp" to enable 802.3ad (instead of
1901standard EtherChannel).
1902
1903The balance-rr, balance-xor and broadcast modes generally
1904require that the switch have the appropriate ports grouped together.
1905The nomenclature for such a group differs between switches, it may be
1906called an "etherchannel" (as in the Cisco example, above), a "trunk
1907group" or some other similar variation. For these modes, each switch
1908will also have its own configuration options for the switch's transmit
1909policy to the bond. Typical choices include XOR of either the MAC or
1910IP addresses. The transmit policy of the two peers does not need to
1911match. For these three modes, the bonding mode really selects a
1912transmit policy for an EtherChannel group; all three will interoperate
1913with another EtherChannel group.
1914
1915
19166. 802.1q VLAN Support
1917======================
1918
1919It is possible to configure VLAN devices over a bond interface
1920using the 8021q driver. However, only packets coming from the 8021q
1921driver and passing through bonding will be tagged by default. Self
1922generated packets, for example, bonding's learning packets or ARP
1923packets generated by either ALB mode or the ARP monitor mechanism, are
1924tagged internally by bonding itself. As a result, bonding must
1925"learn" the VLAN IDs configured above it, and use those IDs to tag
1926self generated packets.
1927
1928For reasons of simplicity, and to support the use of adapters
1929that can do VLAN hardware acceleration offloading, the bonding
1930interface declares itself as fully hardware offloading capable, it gets
1931the add_vid/kill_vid notifications to gather the necessary
1932information, and it propagates those actions to the slaves. In case
1933of mixed adapter types, hardware accelerated tagged packets that
1934should go through an adapter that is not offloading capable are
1935"un-accelerated" by the bonding driver so the VLAN tag sits in the
1936regular location.
1937
1938VLAN interfaces *must* be added on top of a bonding interface
1939only after enslaving at least one slave. The bonding interface has a
1940hardware address of 00:00:00:00:00:00 until the first slave is added.
1941If the VLAN interface is created prior to the first enslavement, it
1942would pick up the all-zeroes hardware address. Once the first slave
1943is attached to the bond, the bond device itself will pick up the
1944slave's hardware address, which is then available for the VLAN device.
1945
1946Also, be aware that a similar problem can occur if all slaves
1947are released from a bond that still has one or more VLAN interfaces on
1948top of it. When a new slave is added, the bonding interface will
1949obtain its hardware address from the first slave, which might not
1950match the hardware address of the VLAN interfaces (which was
1951ultimately copied from an earlier slave).
1952
1953There are two methods to insure that the VLAN device operates
1954with the correct hardware address if all slaves are removed from a
1955bond interface:
1956
19571. Remove all VLAN interfaces then recreate them
1958
19592. Set the bonding interface's hardware address so that it
1960matches the hardware address of the VLAN interfaces.
1961
1962Note that changing a VLAN interface's HW address would set the
1963underlying device -- i.e. the bonding interface -- to promiscuous
1964mode, which might not be what you want.
1965
1966
19677. Link Monitoring
1968==================
1969
1970The bonding driver at present supports two schemes for
1971monitoring a slave device's link state: the ARP monitor and the MII
1972monitor.
1973
1974At the present time, due to implementation restrictions in the
1975bonding driver itself, it is not possible to enable both ARP and MII
1976monitoring simultaneously.
1977
19787.1 ARP Monitor Operation
1979-------------------------
1980
1981The ARP monitor operates as its name suggests: it sends ARP
1982queries to one or more designated peer systems on the network, and
1983uses the response as an indication that the link is operating. This
1984gives some assurance that traffic is actually flowing to and from one
1985or more peers on the local network.
1986
19877.2 Configuring Multiple ARP Targets
1988------------------------------------
1989
1990While ARP monitoring can be done with just one target, it can
1991be useful in a High Availability setup to have several targets to
1992monitor. In the case of just one target, the target itself may go
1993down or have a problem making it unresponsive to ARP requests. Having
1994an additional target (or several) increases the reliability of the ARP
1995monitoring.
1996
1997Multiple ARP targets must be separated by commas as follows::
1998
1999 # example options for ARP monitoring with three targets
2000 alias bond0 bonding
2001 options bond0 arp_interval=60 arp_ip_target=192.168.0.1,192.168.0.3,192.168.0.9
2002
2003For just a single target the options would resemble::
2004
2005 # example options for ARP monitoring with one target
2006 alias bond0 bonding
2007 options bond0 arp_interval=60 arp_ip_target=192.168.0.100
2008
2009
20107.3 MII Monitor Operation
2011-------------------------
2012
2013The MII monitor monitors only the carrier state of the local
2014network interface. It accomplishes this in one of three ways: by
2015depending upon the device driver to maintain its carrier state, by
2016querying the device's MII registers, or by making an ethtool query to
2017the device.
2018
2019If the use_carrier module parameter is 1 (the default value),
2020then the MII monitor will rely on the driver for carrier state
2021information (via the netif_carrier subsystem). As explained in the
2022use_carrier parameter information, above, if the MII monitor fails to
2023detect carrier loss on the device (e.g., when the cable is physically
2024disconnected), it may be that the driver does not support
2025netif_carrier.
2026
2027If use_carrier is 0, then the MII monitor will first query the
2028device's (via ioctl) MII registers and check the link state. If that
2029request fails (not just that it returns carrier down), then the MII
2030monitor will make an ethtool ETHTOOL_GLINK request to attempt to obtain
2031the same information. If both methods fail (i.e., the driver either
2032does not support or had some error in processing both the MII register
2033and ethtool requests), then the MII monitor will assume the link is
2034up.
2035
20368. Potential Sources of Trouble
2037===============================
2038
20398.1 Adventures in Routing
2040-------------------------
2041
2042When bonding is configured, it is important that the slave
2043devices not have routes that supersede routes of the master (or,
2044generally, not have routes at all). For example, suppose the bonding
2045device bond0 has two slaves, eth0 and eth1, and the routing table is
2046as follows::
2047
2048 Kernel IP routing table
2049 Destination Gateway Genmask Flags MSS Window irtt Iface
2050 10.0.0.0 0.0.0.0 255.255.0.0 U 40 0 0 eth0
2051 10.0.0.0 0.0.0.0 255.255.0.0 U 40 0 0 eth1
2052 10.0.0.0 0.0.0.0 255.255.0.0 U 40 0 0 bond0
2053 127.0.0.0 0.0.0.0 255.0.0.0 U 40 0 0 lo
2054
2055This routing configuration will likely still update the
2056receive/transmit times in the driver (needed by the ARP monitor), but
2057may bypass the bonding driver (because outgoing traffic to, in this
2058case, another host on network 10 would use eth0 or eth1 before bond0).
2059
2060The ARP monitor (and ARP itself) may become confused by this
2061configuration, because ARP requests (generated by the ARP monitor)
2062will be sent on one interface (bond0), but the corresponding reply
2063will arrive on a different interface (eth0). This reply looks to ARP
2064as an unsolicited ARP reply (because ARP matches replies on an
2065interface basis), and is discarded. The MII monitor is not affected
2066by the state of the routing table.
2067
2068The solution here is simply to insure that slaves do not have
2069routes of their own, and if for some reason they must, those routes do
2070not supersede routes of their master. This should generally be the
2071case, but unusual configurations or errant manual or automatic static
2072route additions may cause trouble.
2073
20748.2 Ethernet Device Renaming
2075----------------------------
2076
2077On systems with network configuration scripts that do not
2078associate physical devices directly with network interface names (so
2079that the same physical device always has the same "ethX" name), it may
2080be necessary to add some special logic to config files in
2081/etc/modprobe.d/.
2082
2083For example, given a modules.conf containing the following::
2084
2085 alias bond0 bonding
2086 options bond0 mode=some-mode miimon=50
2087 alias eth0 tg3
2088 alias eth1 tg3
2089 alias eth2 e1000
2090 alias eth3 e1000
2091
2092If neither eth0 and eth1 are slaves to bond0, then when the
2093bond0 interface comes up, the devices may end up reordered. This
2094happens because bonding is loaded first, then its slave device's
2095drivers are loaded next. Since no other drivers have been loaded,
2096when the e1000 driver loads, it will receive eth0 and eth1 for its
2097devices, but the bonding configuration tries to enslave eth2 and eth3
2098(which may later be assigned to the tg3 devices).
2099
2100Adding the following::
2101
2102 add above bonding e1000 tg3
2103
2104causes modprobe to load e1000 then tg3, in that order, when
2105bonding is loaded. This command is fully documented in the
2106modules.conf manual page.
2107
2108On systems utilizing modprobe an equivalent problem can occur.
2109In this case, the following can be added to config files in
2110/etc/modprobe.d/ as::
2111
2112 softdep bonding pre: tg3 e1000
2113
2114This will load tg3 and e1000 modules before loading the bonding one.
2115Full documentation on this can be found in the modprobe.d and modprobe
2116manual pages.
2117
21188.3. Painfully Slow Or No Failed Link Detection By Miimon
2119---------------------------------------------------------
2120
2121By default, bonding enables the use_carrier option, which
2122instructs bonding to trust the driver to maintain carrier state.
2123
2124As discussed in the options section, above, some drivers do
2125not support the netif_carrier_on/_off link state tracking system.
2126With use_carrier enabled, bonding will always see these links as up,
2127regardless of their actual state.
2128
2129Additionally, other drivers do support netif_carrier, but do
2130not maintain it in real time, e.g., only polling the link state at
2131some fixed interval. In this case, miimon will detect failures, but
2132only after some long period of time has expired. If it appears that
2133miimon is very slow in detecting link failures, try specifying
2134use_carrier=0 to see if that improves the failure detection time. If
2135it does, then it may be that the driver checks the carrier state at a
2136fixed interval, but does not cache the MII register values (so the
2137use_carrier=0 method of querying the registers directly works). If
2138use_carrier=0 does not improve the failover, then the driver may cache
2139the registers, or the problem may be elsewhere.
2140
2141Also, remember that miimon only checks for the device's
2142carrier state. It has no way to determine the state of devices on or
2143beyond other ports of a switch, or if a switch is refusing to pass
2144traffic while still maintaining carrier on.
2145
21469. SNMP agents
2147===============
2148
2149If running SNMP agents, the bonding driver should be loaded
2150before any network drivers participating in a bond. This requirement
2151is due to the interface index (ipAdEntIfIndex) being associated to
2152the first interface found with a given IP address. That is, there is
2153only one ipAdEntIfIndex for each IP address. For example, if eth0 and
2154eth1 are slaves of bond0 and the driver for eth0 is loaded before the
2155bonding driver, the interface for the IP address will be associated
2156with the eth0 interface. This configuration is shown below, the IP
2157address 192.168.1.1 has an interface index of 2 which indexes to eth0
2158in the ifDescr table (ifDescr.2).
2159
2160::
2161
2162 interfaces.ifTable.ifEntry.ifDescr.1 = lo
2163 interfaces.ifTable.ifEntry.ifDescr.2 = eth0
2164 interfaces.ifTable.ifEntry.ifDescr.3 = eth1
2165 interfaces.ifTable.ifEntry.ifDescr.4 = eth2
2166 interfaces.ifTable.ifEntry.ifDescr.5 = eth3
2167 interfaces.ifTable.ifEntry.ifDescr.6 = bond0
2168 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.10.10.10 = 5
2169 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.192.168.1.1 = 2
2170 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.74.20.94 = 4
2171 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.127.0.0.1 = 1
2172
2173This problem is avoided by loading the bonding driver before
2174any network drivers participating in a bond. Below is an example of
2175loading the bonding driver first, the IP address 192.168.1.1 is
2176correctly associated with ifDescr.2.
2177
2178 interfaces.ifTable.ifEntry.ifDescr.1 = lo
2179 interfaces.ifTable.ifEntry.ifDescr.2 = bond0
2180 interfaces.ifTable.ifEntry.ifDescr.3 = eth0
2181 interfaces.ifTable.ifEntry.ifDescr.4 = eth1
2182 interfaces.ifTable.ifEntry.ifDescr.5 = eth2
2183 interfaces.ifTable.ifEntry.ifDescr.6 = eth3
2184 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.10.10.10 = 6
2185 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.192.168.1.1 = 2
2186 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.74.20.94 = 5
2187 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.127.0.0.1 = 1
2188
2189While some distributions may not report the interface name in
2190ifDescr, the association between the IP address and IfIndex remains
2191and SNMP functions such as Interface_Scan_Next will report that
2192association.
2193
219410. Promiscuous mode
2195====================
2196
2197When running network monitoring tools, e.g., tcpdump, it is
2198common to enable promiscuous mode on the device, so that all traffic
2199is seen (instead of seeing only traffic destined for the local host).
2200The bonding driver handles promiscuous mode changes to the bonding
2201master device (e.g., bond0), and propagates the setting to the slave
2202devices.
2203
2204For the balance-rr, balance-xor, broadcast, and 802.3ad modes,
2205the promiscuous mode setting is propagated to all slaves.
2206
2207For the active-backup, balance-tlb and balance-alb modes, the
2208promiscuous mode setting is propagated only to the active slave.
2209
2210For balance-tlb mode, the active slave is the slave currently
2211receiving inbound traffic.
2212
2213For balance-alb mode, the active slave is the slave used as a
2214"primary." This slave is used for mode-specific control traffic, for
2215sending to peers that are unassigned or if the load is unbalanced.
2216
2217For the active-backup, balance-tlb and balance-alb modes, when
2218the active slave changes (e.g., due to a link failure), the
2219promiscuous setting will be propagated to the new active slave.
2220
222111. Configuring Bonding for High Availability
2222=============================================
2223
2224High Availability refers to configurations that provide
2225maximum network availability by having redundant or backup devices,
2226links or switches between the host and the rest of the world. The
2227goal is to provide the maximum availability of network connectivity
2228(i.e., the network always works), even though other configurations
2229could provide higher throughput.
2230
223111.1 High Availability in a Single Switch Topology
2232--------------------------------------------------
2233
2234If two hosts (or a host and a single switch) are directly
2235connected via multiple physical links, then there is no availability
2236penalty to optimizing for maximum bandwidth. In this case, there is
2237only one switch (or peer), so if it fails, there is no alternative
2238access to fail over to. Additionally, the bonding load balance modes
2239support link monitoring of their members, so if individual links fail,
2240the load will be rebalanced across the remaining devices.
2241
2242See Section 12, "Configuring Bonding for Maximum Throughput"
2243for information on configuring bonding with one peer device.
2244
224511.2 High Availability in a Multiple Switch Topology
2246----------------------------------------------------
2247
2248With multiple switches, the configuration of bonding and the
2249network changes dramatically. In multiple switch topologies, there is
2250a trade off between network availability and usable bandwidth.
2251
2252Below is a sample network, configured to maximize the
2253availability of the network::
2254
2255 | |
2256 |port3 port3|
2257 +-----+----+ +-----+----+
2258 | |port2 ISL port2| |
2259 | switch A +--------------------------+ switch B |
2260 | | | |
2261 +-----+----+ +-----++---+
2262 |port1 port1|
2263 | +-------+ |
2264 +-------------+ host1 +---------------+
2265 eth0 +-------+ eth1
2266
2267In this configuration, there is a link between the two
2268switches (ISL, or inter switch link), and multiple ports connecting to
2269the outside world ("port3" on each switch). There is no technical
2270reason that this could not be extended to a third switch.
2271
227211.2.1 HA Bonding Mode Selection for Multiple Switch Topology
2273-------------------------------------------------------------
2274
2275In a topology such as the example above, the active-backup and
2276broadcast modes are the only useful bonding modes when optimizing for
2277availability; the other modes require all links to terminate on the
2278same peer for them to behave rationally.
2279
2280active-backup:
2281 This is generally the preferred mode, particularly if
2282 the switches have an ISL and play together well. If the
2283 network configuration is such that one switch is specifically
2284 a backup switch (e.g., has lower capacity, higher cost, etc),
2285 then the primary option can be used to insure that the
2286 preferred link is always used when it is available.
2287
2288broadcast:
2289 This mode is really a special purpose mode, and is suitable
2290 only for very specific needs. For example, if the two
2291 switches are not connected (no ISL), and the networks beyond
2292 them are totally independent. In this case, if it is
2293 necessary for some specific one-way traffic to reach both
2294 independent networks, then the broadcast mode may be suitable.
2295
229611.2.2 HA Link Monitoring Selection for Multiple Switch Topology
2297----------------------------------------------------------------
2298
2299The choice of link monitoring ultimately depends upon your
2300switch. If the switch can reliably fail ports in response to other
2301failures, then either the MII or ARP monitors should work. For
2302example, in the above example, if the "port3" link fails at the remote
2303end, the MII monitor has no direct means to detect this. The ARP
2304monitor could be configured with a target at the remote end of port3,
2305thus detecting that failure without switch support.
2306
2307In general, however, in a multiple switch topology, the ARP
2308monitor can provide a higher level of reliability in detecting end to
2309end connectivity failures (which may be caused by the failure of any
2310individual component to pass traffic for any reason). Additionally,
2311the ARP monitor should be configured with multiple targets (at least
2312one for each switch in the network). This will insure that,
2313regardless of which switch is active, the ARP monitor has a suitable
2314target to query.
2315
2316Note, also, that of late many switches now support a functionality
2317generally referred to as "trunk failover." This is a feature of the
2318switch that causes the link state of a particular switch port to be set
2319down (or up) when the state of another switch port goes down (or up).
2320Its purpose is to propagate link failures from logically "exterior" ports
2321to the logically "interior" ports that bonding is able to monitor via
2322miimon. Availability and configuration for trunk failover varies by
2323switch, but this can be a viable alternative to the ARP monitor when using
2324suitable switches.
2325
232612. Configuring Bonding for Maximum Throughput
2327==============================================
2328
232912.1 Maximizing Throughput in a Single Switch Topology
2330------------------------------------------------------
2331
2332In a single switch configuration, the best method to maximize
2333throughput depends upon the application and network environment. The
2334various load balancing modes each have strengths and weaknesses in
2335different environments, as detailed below.
2336
2337For this discussion, we will break down the topologies into
2338two categories. Depending upon the destination of most traffic, we
2339categorize them into either "gatewayed" or "local" configurations.
2340
2341In a gatewayed configuration, the "switch" is acting primarily
2342as a router, and the majority of traffic passes through this router to
2343other networks. An example would be the following::
2344
2345
2346 +----------+ +----------+
2347 | |eth0 port1| | to other networks
2348 | Host A +---------------------+ router +------------------->
2349 | +---------------------+ | Hosts B and C are out
2350 | |eth1 port2| | here somewhere
2351 +----------+ +----------+
2352
2353The router may be a dedicated router device, or another host
2354acting as a gateway. For our discussion, the important point is that
2355the majority of traffic from Host A will pass through the router to
2356some other network before reaching its final destination.
2357
2358In a gatewayed network configuration, although Host A may
2359communicate with many other systems, all of its traffic will be sent
2360and received via one other peer on the local network, the router.
2361
2362Note that the case of two systems connected directly via
2363multiple physical links is, for purposes of configuring bonding, the
2364same as a gatewayed configuration. In that case, it happens that all
2365traffic is destined for the "gateway" itself, not some other network
2366beyond the gateway.
2367
2368In a local configuration, the "switch" is acting primarily as
2369a switch, and the majority of traffic passes through this switch to
2370reach other stations on the same network. An example would be the
2371following::
2372
2373 +----------+ +----------+ +--------+
2374 | |eth0 port1| +-------+ Host B |
2375 | Host A +------------+ switch |port3 +--------+
2376 | +------------+ | +--------+
2377 | |eth1 port2| +------------------+ Host C |
2378 +----------+ +----------+port4 +--------+
2379
2380
2381Again, the switch may be a dedicated switch device, or another
2382host acting as a gateway. For our discussion, the important point is
2383that the majority of traffic from Host A is destined for other hosts
2384on the same local network (Hosts B and C in the above example).
2385
2386In summary, in a gatewayed configuration, traffic to and from
2387the bonded device will be to the same MAC level peer on the network
2388(the gateway itself, i.e., the router), regardless of its final
2389destination. In a local configuration, traffic flows directly to and
2390from the final destinations, thus, each destination (Host B, Host C)
2391will be addressed directly by their individual MAC addresses.
2392
2393This distinction between a gatewayed and a local network
2394configuration is important because many of the load balancing modes
2395available use the MAC addresses of the local network source and
2396destination to make load balancing decisions. The behavior of each
2397mode is described below.
2398
2399
240012.1.1 MT Bonding Mode Selection for Single Switch Topology
2401-----------------------------------------------------------
2402
2403This configuration is the easiest to set up and to understand,
2404although you will have to decide which bonding mode best suits your
2405needs. The trade offs for each mode are detailed below:
2406
2407balance-rr:
2408 This mode is the only mode that will permit a single
2409 TCP/IP connection to stripe traffic across multiple
2410 interfaces. It is therefore the only mode that will allow a
2411 single TCP/IP stream to utilize more than one interface's
2412 worth of throughput. This comes at a cost, however: the
2413 striping generally results in peer systems receiving packets out
2414 of order, causing TCP/IP's congestion control system to kick
2415 in, often by retransmitting segments.
2416
2417 It is possible to adjust TCP/IP's congestion limits by
2418 altering the net.ipv4.tcp_reordering sysctl parameter. The
2419 usual default value is 3. But keep in mind TCP stack is able
2420 to automatically increase this when it detects reorders.
2421
2422 Note that the fraction of packets that will be delivered out of
2423 order is highly variable, and is unlikely to be zero. The level
2424 of reordering depends upon a variety of factors, including the
2425 networking interfaces, the switch, and the topology of the
2426 configuration. Speaking in general terms, higher speed network
2427 cards produce more reordering (due to factors such as packet
2428 coalescing), and a "many to many" topology will reorder at a
2429 higher rate than a "many slow to one fast" configuration.
2430
2431 Many switches do not support any modes that stripe traffic
2432 (instead choosing a port based upon IP or MAC level addresses);
2433 for those devices, traffic for a particular connection flowing
2434 through the switch to a balance-rr bond will not utilize greater
2435 than one interface's worth of bandwidth.
2436
2437 If you are utilizing protocols other than TCP/IP, UDP for
2438 example, and your application can tolerate out of order
2439 delivery, then this mode can allow for single stream datagram
2440 performance that scales near linearly as interfaces are added
2441 to the bond.
2442
2443 This mode requires the switch to have the appropriate ports
2444 configured for "etherchannel" or "trunking."
2445
2446active-backup:
2447 There is not much advantage in this network topology to
2448 the active-backup mode, as the inactive backup devices are all
2449 connected to the same peer as the primary. In this case, a
2450 load balancing mode (with link monitoring) will provide the
2451 same level of network availability, but with increased
2452 available bandwidth. On the plus side, active-backup mode
2453 does not require any configuration of the switch, so it may
2454 have value if the hardware available does not support any of
2455 the load balance modes.
2456
2457balance-xor:
2458 This mode will limit traffic such that packets destined
2459 for specific peers will always be sent over the same
2460 interface. Since the destination is determined by the MAC
2461 addresses involved, this mode works best in a "local" network
2462 configuration (as described above), with destinations all on
2463 the same local network. This mode is likely to be suboptimal
2464 if all your traffic is passed through a single router (i.e., a
2465 "gatewayed" network configuration, as described above).
2466
2467 As with balance-rr, the switch ports need to be configured for
2468 "etherchannel" or "trunking."
2469
2470broadcast:
2471 Like active-backup, there is not much advantage to this
2472 mode in this type of network topology.
2473
2474802.3ad:
2475 This mode can be a good choice for this type of network
2476 topology. The 802.3ad mode is an IEEE standard, so all peers
2477 that implement 802.3ad should interoperate well. The 802.3ad
2478 protocol includes automatic configuration of the aggregates,
2479 so minimal manual configuration of the switch is needed
2480 (typically only to designate that some set of devices is
2481 available for 802.3ad). The 802.3ad standard also mandates
2482 that frames be delivered in order (within certain limits), so
2483 in general single connections will not see misordering of
2484 packets. The 802.3ad mode does have some drawbacks: the
2485 standard mandates that all devices in the aggregate operate at
2486 the same speed and duplex. Also, as with all bonding load
2487 balance modes other than balance-rr, no single connection will
2488 be able to utilize more than a single interface's worth of
2489 bandwidth.
2490
2491 Additionally, the linux bonding 802.3ad implementation
2492 distributes traffic by peer (using an XOR of MAC addresses
2493 and packet type ID), so in a "gatewayed" configuration, all
2494 outgoing traffic will generally use the same device. Incoming
2495 traffic may also end up on a single device, but that is
2496 dependent upon the balancing policy of the peer's 802.3ad
2497 implementation. In a "local" configuration, traffic will be
2498 distributed across the devices in the bond.
2499
2500 Finally, the 802.3ad mode mandates the use of the MII monitor,
2501 therefore, the ARP monitor is not available in this mode.
2502
2503balance-tlb:
2504 The balance-tlb mode balances outgoing traffic by peer.
2505 Since the balancing is done according to MAC address, in a
2506 "gatewayed" configuration (as described above), this mode will
2507 send all traffic across a single device. However, in a
2508 "local" network configuration, this mode balances multiple
2509 local network peers across devices in a vaguely intelligent
2510 manner (not a simple XOR as in balance-xor or 802.3ad mode),
2511 so that mathematically unlucky MAC addresses (i.e., ones that
2512 XOR to the same value) will not all "bunch up" on a single
2513 interface.
2514
2515 Unlike 802.3ad, interfaces may be of differing speeds, and no
2516 special switch configuration is required. On the down side,
2517 in this mode all incoming traffic arrives over a single
2518 interface, this mode requires certain ethtool support in the
2519 network device driver of the slave interfaces, and the ARP
2520 monitor is not available.
2521
2522balance-alb:
2523 This mode is everything that balance-tlb is, and more.
2524 It has all of the features (and restrictions) of balance-tlb,
2525 and will also balance incoming traffic from local network
2526 peers (as described in the Bonding Module Options section,
2527 above).
2528
2529 The only additional down side to this mode is that the network
2530 device driver must support changing the hardware address while
2531 the device is open.
2532
253312.1.2 MT Link Monitoring for Single Switch Topology
2534----------------------------------------------------
2535
2536The choice of link monitoring may largely depend upon which
2537mode you choose to use. The more advanced load balancing modes do not
2538support the use of the ARP monitor, and are thus restricted to using
2539the MII monitor (which does not provide as high a level of end to end
2540assurance as the ARP monitor).
2541
254212.2 Maximum Throughput in a Multiple Switch Topology
2543-----------------------------------------------------
2544
2545Multiple switches may be utilized to optimize for throughput
2546when they are configured in parallel as part of an isolated network
2547between two or more systems, for example::
2548
2549 +-----------+
2550 | Host A |
2551 +-+---+---+-+
2552 | | |
2553 +--------+ | +---------+
2554 | | |
2555 +------+---+ +-----+----+ +-----+----+
2556 | Switch A | | Switch B | | Switch C |
2557 +------+---+ +-----+----+ +-----+----+
2558 | | |
2559 +--------+ | +---------+
2560 | | |
2561 +-+---+---+-+
2562 | Host B |
2563 +-----------+
2564
2565In this configuration, the switches are isolated from one
2566another. One reason to employ a topology such as this is for an
2567isolated network with many hosts (a cluster configured for high
2568performance, for example), using multiple smaller switches can be more
2569cost effective than a single larger switch, e.g., on a network with 24
2570hosts, three 24 port switches can be significantly less expensive than
2571a single 72 port switch.
2572
2573If access beyond the network is required, an individual host
2574can be equipped with an additional network device connected to an
2575external network; this host then additionally acts as a gateway.
2576
257712.2.1 MT Bonding Mode Selection for Multiple Switch Topology
2578-------------------------------------------------------------
2579
2580In actual practice, the bonding mode typically employed in
2581configurations of this type is balance-rr. Historically, in this
2582network configuration, the usual caveats about out of order packet
2583delivery are mitigated by the use of network adapters that do not do
2584any kind of packet coalescing (via the use of NAPI, or because the
2585device itself does not generate interrupts until some number of
2586packets has arrived). When employed in this fashion, the balance-rr
2587mode allows individual connections between two hosts to effectively
2588utilize greater than one interface's bandwidth.
2589
259012.2.2 MT Link Monitoring for Multiple Switch Topology
2591------------------------------------------------------
2592
2593Again, in actual practice, the MII monitor is most often used
2594in this configuration, as performance is given preference over
2595availability. The ARP monitor will function in this topology, but its
2596advantages over the MII monitor are mitigated by the volume of probes
2597needed as the number of systems involved grows (remember that each
2598host in the network is configured with bonding).
2599
260013. Switch Behavior Issues
2601==========================
2602
260313.1 Link Establishment and Failover Delays
2604-------------------------------------------
2605
2606Some switches exhibit undesirable behavior with regard to the
2607timing of link up and down reporting by the switch.
2608
2609First, when a link comes up, some switches may indicate that
2610the link is up (carrier available), but not pass traffic over the
2611interface for some period of time. This delay is typically due to
2612some type of autonegotiation or routing protocol, but may also occur
2613during switch initialization (e.g., during recovery after a switch
2614failure). If you find this to be a problem, specify an appropriate
2615value to the updelay bonding module option to delay the use of the
2616relevant interface(s).
2617
2618Second, some switches may "bounce" the link state one or more
2619times while a link is changing state. This occurs most commonly while
2620the switch is initializing. Again, an appropriate updelay value may
2621help.
2622
2623Note that when a bonding interface has no active links, the
2624driver will immediately reuse the first link that goes up, even if the
2625updelay parameter has been specified (the updelay is ignored in this
2626case). If there are slave interfaces waiting for the updelay timeout
2627to expire, the interface that first went into that state will be
2628immediately reused. This reduces down time of the network if the
2629value of updelay has been overestimated, and since this occurs only in
2630cases with no connectivity, there is no additional penalty for
2631ignoring the updelay.
2632
2633In addition to the concerns about switch timings, if your
2634switches take a long time to go into backup mode, it may be desirable
2635to not activate a backup interface immediately after a link goes down.
2636Failover may be delayed via the downdelay bonding module option.
2637
263813.2 Duplicated Incoming Packets
2639--------------------------------
2640
2641NOTE: Starting with version 3.0.2, the bonding driver has logic to
2642suppress duplicate packets, which should largely eliminate this problem.
2643The following description is kept for reference.
2644
2645It is not uncommon to observe a short burst of duplicated
2646traffic when the bonding device is first used, or after it has been
2647idle for some period of time. This is most easily observed by issuing
2648a "ping" to some other host on the network, and noticing that the
2649output from ping flags duplicates (typically one per slave).
2650
2651For example, on a bond in active-backup mode with five slaves
2652all connected to one switch, the output may appear as follows::
2653
2654 # ping -n 10.0.4.2
2655 PING 10.0.4.2 (10.0.4.2) from 10.0.3.10 : 56(84) bytes of data.
2656 64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.7 ms
2657 64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
2658 64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
2659 64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
2660 64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
2661 64 bytes from 10.0.4.2: icmp_seq=2 ttl=64 time=0.216 ms
2662 64 bytes from 10.0.4.2: icmp_seq=3 ttl=64 time=0.267 ms
2663 64 bytes from 10.0.4.2: icmp_seq=4 ttl=64 time=0.222 ms
2664
2665This is not due to an error in the bonding driver, rather, it
2666is a side effect of how many switches update their MAC forwarding
2667tables. Initially, the switch does not associate the MAC address in
2668the packet with a particular switch port, and so it may send the
2669traffic to all ports until its MAC forwarding table is updated. Since
2670the interfaces attached to the bond may occupy multiple ports on a
2671single switch, when the switch (temporarily) floods the traffic to all
2672ports, the bond device receives multiple copies of the same packet
2673(one per slave device).
2674
2675The duplicated packet behavior is switch dependent, some
2676switches exhibit this, and some do not. On switches that display this
2677behavior, it can be induced by clearing the MAC forwarding table (on
2678most Cisco switches, the privileged command "clear mac address-table
2679dynamic" will accomplish this).
2680
268114. Hardware Specific Considerations
2682====================================
2683
2684This section contains additional information for configuring
2685bonding on specific hardware platforms, or for interfacing bonding
2686with particular switches or other devices.
2687
268814.1 IBM BladeCenter
2689--------------------
2690
2691This applies to the JS20 and similar systems.
2692
2693On the JS20 blades, the bonding driver supports only
2694balance-rr, active-backup, balance-tlb and balance-alb modes. This is
2695largely due to the network topology inside the BladeCenter, detailed
2696below.
2697
2698JS20 network adapter information
2699--------------------------------
2700
2701All JS20s come with two Broadcom Gigabit Ethernet ports
2702integrated on the planar (that's "motherboard" in IBM-speak). In the
2703BladeCenter chassis, the eth0 port of all JS20 blades is hard wired to
2704I/O Module #1; similarly, all eth1 ports are wired to I/O Module #2.
2705An add-on Broadcom daughter card can be installed on a JS20 to provide
2706two more Gigabit Ethernet ports. These ports, eth2 and eth3, are
2707wired to I/O Modules 3 and 4, respectively.
2708
2709Each I/O Module may contain either a switch or a passthrough
2710module (which allows ports to be directly connected to an external
2711switch). Some bonding modes require a specific BladeCenter internal
2712network topology in order to function; these are detailed below.
2713
2714Additional BladeCenter-specific networking information can be
2715found in two IBM Redbooks (www.ibm.com/redbooks):
2716
2717- "IBM eServer BladeCenter Networking Options"
2718- "IBM eServer BladeCenter Layer 2-7 Network Switching"
2719
2720BladeCenter networking configuration
2721------------------------------------
2722
2723Because a BladeCenter can be configured in a very large number
2724of ways, this discussion will be confined to describing basic
2725configurations.
2726
2727Normally, Ethernet Switch Modules (ESMs) are used in I/O
2728modules 1 and 2. In this configuration, the eth0 and eth1 ports of a
2729JS20 will be connected to different internal switches (in the
2730respective I/O modules).
2731
2732A passthrough module (OPM or CPM, optical or copper,
2733passthrough module) connects the I/O module directly to an external
2734switch. By using PMs in I/O module #1 and #2, the eth0 and eth1
2735interfaces of a JS20 can be redirected to the outside world and
2736connected to a common external switch.
2737
2738Depending upon the mix of ESMs and PMs, the network will
2739appear to bonding as either a single switch topology (all PMs) or as a
2740multiple switch topology (one or more ESMs, zero or more PMs). It is
2741also possible to connect ESMs together, resulting in a configuration
2742much like the example in "High Availability in a Multiple Switch
2743Topology," above.
2744
2745Requirements for specific modes
2746-------------------------------
2747
2748The balance-rr mode requires the use of passthrough modules
2749for devices in the bond, all connected to an common external switch.
2750That switch must be configured for "etherchannel" or "trunking" on the
2751appropriate ports, as is usual for balance-rr.
2752
2753The balance-alb and balance-tlb modes will function with
2754either switch modules or passthrough modules (or a mix). The only
2755specific requirement for these modes is that all network interfaces
2756must be able to reach all destinations for traffic sent over the
2757bonding device (i.e., the network must converge at some point outside
2758the BladeCenter).
2759
2760The active-backup mode has no additional requirements.
2761
2762Link monitoring issues
2763----------------------
2764
2765When an Ethernet Switch Module is in place, only the ARP
2766monitor will reliably detect link loss to an external switch. This is
2767nothing unusual, but examination of the BladeCenter cabinet would
2768suggest that the "external" network ports are the ethernet ports for
2769the system, when it fact there is a switch between these "external"
2770ports and the devices on the JS20 system itself. The MII monitor is
2771only able to detect link failures between the ESM and the JS20 system.
2772
2773When a passthrough module is in place, the MII monitor does
2774detect failures to the "external" port, which is then directly
2775connected to the JS20 system.
2776
2777Other concerns
2778--------------
2779
2780The Serial Over LAN (SoL) link is established over the primary
2781ethernet (eth0) only, therefore, any loss of link to eth0 will result
2782in losing your SoL connection. It will not fail over with other
2783network traffic, as the SoL system is beyond the control of the
2784bonding driver.
2785
2786It may be desirable to disable spanning tree on the switch
2787(either the internal Ethernet Switch Module, or an external switch) to
2788avoid fail-over delay issues when using bonding.
2789
2790
279115. Frequently Asked Questions
2792==============================
2793
27941. Is it SMP safe?
2795-------------------
2796
2797Yes. The old 2.0.xx channel bonding patch was not SMP safe.
2798The new driver was designed to be SMP safe from the start.
2799
28002. What type of cards will work with it?
2801-----------------------------------------
2802
2803Any Ethernet type cards (you can even mix cards - a Intel
2804EtherExpress PRO/100 and a 3com 3c905b, for example). For most modes,
2805devices need not be of the same speed.
2806
2807Starting with version 3.2.1, bonding also supports Infiniband
2808slaves in active-backup mode.
2809
28103. How many bonding devices can I have?
2811----------------------------------------
2812
2813There is no limit.
2814
28154. How many slaves can a bonding device have?
2816----------------------------------------------
2817
2818This is limited only by the number of network interfaces Linux
2819supports and/or the number of network cards you can place in your
2820system.
2821
28225. What happens when a slave link dies?
2823----------------------------------------
2824
2825If link monitoring is enabled, then the failing device will be
2826disabled. The active-backup mode will fail over to a backup link, and
2827other modes will ignore the failed link. The link will continue to be
2828monitored, and should it recover, it will rejoin the bond (in whatever
2829manner is appropriate for the mode). See the sections on High
2830Availability and the documentation for each mode for additional
2831information.
2832
2833Link monitoring can be enabled via either the miimon or
2834arp_interval parameters (described in the module parameters section,
2835above). In general, miimon monitors the carrier state as sensed by
2836the underlying network device, and the arp monitor (arp_interval)
2837monitors connectivity to another host on the local network.
2838
2839If no link monitoring is configured, the bonding driver will
2840be unable to detect link failures, and will assume that all links are
2841always available. This will likely result in lost packets, and a
2842resulting degradation of performance. The precise performance loss
2843depends upon the bonding mode and network configuration.
2844
28456. Can bonding be used for High Availability?
2846----------------------------------------------
2847
2848Yes. See the section on High Availability for details.
2849
28507. Which switches/systems does it work with?
2851---------------------------------------------
2852
2853The full answer to this depends upon the desired mode.
2854
2855In the basic balance modes (balance-rr and balance-xor), it
2856works with any system that supports etherchannel (also called
2857trunking). Most managed switches currently available have such
2858support, and many unmanaged switches as well.
2859
2860The advanced balance modes (balance-tlb and balance-alb) do
2861not have special switch requirements, but do need device drivers that
2862support specific features (described in the appropriate section under
2863module parameters, above).
2864
2865In 802.3ad mode, it works with systems that support IEEE
2866802.3ad Dynamic Link Aggregation. Most managed and many unmanaged
2867switches currently available support 802.3ad.
2868
2869The active-backup mode should work with any Layer-II switch.
2870
28718. Where does a bonding device get its MAC address from?
2872---------------------------------------------------------
2873
2874When using slave devices that have fixed MAC addresses, or when
2875the fail_over_mac option is enabled, the bonding device's MAC address is
2876the MAC address of the active slave.
2877
2878For other configurations, if not explicitly configured (with
2879ifconfig or ip link), the MAC address of the bonding device is taken from
2880its first slave device. This MAC address is then passed to all following
2881slaves and remains persistent (even if the first slave is removed) until
2882the bonding device is brought down or reconfigured.
2883
2884If you wish to change the MAC address, you can set it with
2885ifconfig or ip link::
2886
2887 # ifconfig bond0 hw ether 00:11:22:33:44:55
2888
2889 # ip link set bond0 address 66:77:88:99:aa:bb
2890
2891The MAC address can be also changed by bringing down/up the
2892device and then changing its slaves (or their order)::
2893
2894 # ifconfig bond0 down ; modprobe -r bonding
2895 # ifconfig bond0 .... up
2896 # ifenslave bond0 eth...
2897
2898This method will automatically take the address from the next
2899slave that is added.
2900
2901To restore your slaves' MAC addresses, you need to detach them
2902from the bond (``ifenslave -d bond0 eth0``). The bonding driver will
2903then restore the MAC addresses that the slaves had before they were
2904enslaved.
2905
290616. Resources and Links
2907=======================
2908
2909The latest version of the bonding driver can be found in the latest
2910version of the linux kernel, found on http://kernel.org
2911
2912The latest version of this document can be found in the latest kernel
2913source (named Documentation/networking/bonding.rst).
2914
2915Discussions regarding the development of the bonding driver take place
2916on the main Linux network mailing list, hosted at vger.kernel.org. The list
2917address is:
2918
2919netdev@vger.kernel.org
2920
2921The administrative interface (to subscribe or unsubscribe) can
2922be found at:
2923
2924http://vger.kernel.org/vger-lists.html#netdev