blob: e27202bb8d752dd72dead86953976762b78e34be [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001
Jay Vosburgh00354cf2005-07-21 12:18:02 -07002 Linux Ethernet Bonding Driver HOWTO
3
Jiri Pirkoa5499522009-09-25 03:28:09 +00004 Latest update: 23 September 2009
Linus Torvalds1da177e2005-04-16 15:20:36 -07005
6Initial release : Thomas Davis <tadavis at lbl.gov>
7Corrections, HA extensions : 2000/10/03-15 :
8 - Willy Tarreau <willy at meta-x.org>
9 - Constantine Gavrilov <const-g at xpert.com>
10 - Chad N. Tindel <ctindel at ieee dot org>
11 - Janice Girouard <girouard at us dot ibm dot com>
12 - Jay Vosburgh <fubar at us dot ibm dot com>
13
14Reorganized and updated Feb 2005 by Jay Vosburgh
Auke Kok6224e012006-06-08 11:15:35 -070015Added Sysfs information: 2006/04/24
16 - Mitch Williams <mitch.a.williams at intel.com>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017
Jay Vosburgh00354cf2005-07-21 12:18:02 -070018Introduction
19============
Linus Torvalds1da177e2005-04-16 15:20:36 -070020
Jay Vosburgh00354cf2005-07-21 12:18:02 -070021 The Linux bonding driver provides a method for aggregating
22multiple network interfaces into a single logical "bonded" interface.
23The behavior of the bonded interfaces depends upon the mode; generally
24speaking, modes provide either hot standby or load balancing services.
25Additionally, link integrity monitoring may be performed.
26
27 The bonding driver originally came from Donald Becker's
28beowulf patches for kernel 2.0. It has changed quite a bit since, and
29the original tools from extreme-linux and beowulf sites will not work
30with this version of the driver.
31
32 For new versions of the driver, updated userspace tools, and
33who to ask for help, please follow the links at the end of this file.
Linus Torvalds1da177e2005-04-16 15:20:36 -070034
35Table of Contents
36=================
37
381. Bonding Driver Installation
39
402. Bonding Driver Options
41
423. Configuring Bonding Devices
Auke Kok6224e012006-06-08 11:15:35 -0700433.1 Configuration with Sysconfig Support
443.1.1 Using DHCP with Sysconfig
453.1.2 Configuring Multiple Bonds with Sysconfig
463.2 Configuration with Initscripts Support
473.2.1 Using DHCP with Initscripts
483.2.2 Configuring Multiple Bonds with Initscripts
493.3 Configuring Bonding Manually with Ifenslave
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700503.3.1 Configuring Multiple Bonds Manually
Auke Kok6224e012006-06-08 11:15:35 -0700513.4 Configuring Bonding Manually via Sysfs
Nicolas de Pesloüande221bd2011-01-24 13:21:37 +0000523.5 Configuration with Interfaces Support
533.6 Overriding Configuration for Special Cases
Linus Torvalds1da177e2005-04-16 15:20:36 -070054
Auke Kok6224e012006-06-08 11:15:35 -0700554. Querying Bonding Configuration
564.1 Bonding Configuration
574.2 Network Configuration
Linus Torvalds1da177e2005-04-16 15:20:36 -070058
Auke Kok6224e012006-06-08 11:15:35 -0700595. Switch Configuration
Linus Torvalds1da177e2005-04-16 15:20:36 -070060
Auke Kok6224e012006-06-08 11:15:35 -0700616. 802.1q VLAN Support
Linus Torvalds1da177e2005-04-16 15:20:36 -070062
Auke Kok6224e012006-06-08 11:15:35 -0700637. Link Monitoring
647.1 ARP Monitor Operation
657.2 Configuring Multiple ARP Targets
667.3 MII Monitor Operation
Linus Torvalds1da177e2005-04-16 15:20:36 -070067
Auke Kok6224e012006-06-08 11:15:35 -0700688. Potential Trouble Sources
698.1 Adventures in Routing
708.2 Ethernet Device Renaming
718.3 Painfully Slow Or No Failed Link Detection By Miimon
Linus Torvalds1da177e2005-04-16 15:20:36 -070072
Auke Kok6224e012006-06-08 11:15:35 -0700739. SNMP agents
Linus Torvalds1da177e2005-04-16 15:20:36 -070074
Auke Kok6224e012006-06-08 11:15:35 -07007510. Promiscuous mode
Linus Torvalds1da177e2005-04-16 15:20:36 -070076
Auke Kok6224e012006-06-08 11:15:35 -07007711. Configuring Bonding for High Availability
7811.1 High Availability in a Single Switch Topology
7911.2 High Availability in a Multiple Switch Topology
8011.2.1 HA Bonding Mode Selection for Multiple Switch Topology
8111.2.2 HA Link Monitoring for Multiple Switch Topology
Linus Torvalds1da177e2005-04-16 15:20:36 -070082
Auke Kok6224e012006-06-08 11:15:35 -07008312. Configuring Bonding for Maximum Throughput
8412.1 Maximum Throughput in a Single Switch Topology
8512.1.1 MT Bonding Mode Selection for Single Switch Topology
8612.1.2 MT Link Monitoring for Single Switch Topology
8712.2 Maximum Throughput in a Multiple Switch Topology
8812.2.1 MT Bonding Mode Selection for Multiple Switch Topology
8912.2.2 MT Link Monitoring for Multiple Switch Topology
Linus Torvalds1da177e2005-04-16 15:20:36 -070090
Auke Kok6224e012006-06-08 11:15:35 -07009113. Switch Behavior Issues
9213.1 Link Establishment and Failover Delays
9313.2 Duplicated Incoming Packets
Linus Torvalds1da177e2005-04-16 15:20:36 -070094
Auke Kok6224e012006-06-08 11:15:35 -07009514. Hardware Specific Considerations
9614.1 IBM BladeCenter
Jay Vosburgh00354cf2005-07-21 12:18:02 -070097
Auke Kok6224e012006-06-08 11:15:35 -07009815. Frequently Asked Questions
Jay Vosburgh00354cf2005-07-21 12:18:02 -070099
Auke Kok6224e012006-06-08 11:15:35 -070010016. Resources and Links
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101
102
1031. Bonding Driver Installation
104==============================
105
106 Most popular distro kernels ship with the bonding driver
107already available as a module and the ifenslave user level control
108program installed and ready for use. If your distro does not, or you
109have need to compile bonding from source (e.g., configuring and
110installing a mainline kernel from kernel.org), you'll need to perform
111the following steps:
112
1131.1 Configure and build the kernel with bonding
114-----------------------------------------------
115
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700116 The current version of the bonding driver is available in the
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117drivers/net/bonding subdirectory of the most recent kernel source
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700118(which is available on http://kernel.org). Most users "rolling their
119own" will want to use the most recent kernel from kernel.org.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120
121 Configure kernel with "make menuconfig" (or "make xconfig" or
122"make config"), then select "Bonding driver support" in the "Network
123device support" section. It is recommended that you configure the
124driver as module since it is currently the only way to pass parameters
125to the driver or configure more than one bonding device.
126
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700127 Build and install the new kernel and modules, then continue
128below to install ifenslave.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129
1301.2 Install ifenslave Control Utility
131-------------------------------------
132
133 The ifenslave user level control program is included in the
134kernel source tree, in the file Documentation/networking/ifenslave.c.
135It is generally recommended that you use the ifenslave that
136corresponds to the kernel that you are using (either from the same
137source tree or supplied with the distro), however, ifenslave
138executables from older kernels should function (but features newer
139than the ifenslave release are not supported). Running an ifenslave
140that is newer than the kernel is not supported, and may or may not
141work.
142
143 To install ifenslave, do the following:
144
145# gcc -Wall -O -I/usr/src/linux/include ifenslave.c -o ifenslave
146# cp ifenslave /sbin/ifenslave
147
148 If your kernel source is not in "/usr/src/linux," then replace
149"/usr/src/linux/include" in the above with the location of your kernel
150source include directory.
151
152 You may wish to back up any existing /sbin/ifenslave, or, for
153testing or informal use, tag the ifenslave to the kernel version
154(e.g., name the ifenslave executable /sbin/ifenslave-2.6.10).
155
156IMPORTANT NOTE:
157
158 If you omit the "-I" or specify an incorrect directory, you
159may end up with an ifenslave that is incompatible with the kernel
160you're trying to build it for. Some distros (e.g., Red Hat from 7.1
161onwards) do not have /usr/include/linux symbolically linked to the
162default kernel source include directory.
163
Auke Kok6224e012006-06-08 11:15:35 -0700164SECOND IMPORTANT NOTE:
Nicolas de Pesloüande221bd2011-01-24 13:21:37 +0000165 If you plan to configure bonding using sysfs or using the
166/etc/network/interfaces file, you do not need to use ifenslave.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167
1682. Bonding Driver Options
169=========================
170
Jay Vosburgh9a6c6862007-11-13 20:25:48 -0800171 Options for the bonding driver are supplied as parameters to the
172bonding module at load time, or are specified via sysfs.
173
174 Module options may be given as command line arguments to the
175insmod or modprobe command, but are usually specified in either the
176/etc/modules.conf or /etc/modprobe.conf configuration file, or in a
177distro-specific configuration file (some of which are detailed in the next
178section).
179
180 Details on bonding support for sysfs is provided in the
181"Configuring Bonding Manually via Sysfs" section, below.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182
183 The available bonding driver parameters are listed below. If a
184parameter is not specified the default value is used. When initially
185configuring a bond, it is recommended "tail -f /var/log/messages" be
186run in a separate window to watch for bonding driver error messages.
187
188 It is critical that either the miimon or arp_interval and
189arp_ip_target parameters be specified, otherwise serious network
190degradation will occur during link failures. Very few devices do not
191support at least miimon, so there is really no reason not to use it.
192
193 Options with textual values will accept either the text name
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700194or, for backwards compatibility, the option value. E.g.,
195"mode=802.3ad" and "mode=4" set the same mode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196
197 The parameters are as follows:
198
Jay Vosburghfd989c82008-11-04 17:51:16 -0800199ad_select
200
201 Specifies the 802.3ad aggregation selection logic to use. The
202 possible values and their effects are:
203
204 stable or 0
205
206 The active aggregator is chosen by largest aggregate
207 bandwidth.
208
209 Reselection of the active aggregator occurs only when all
210 slaves of the active aggregator are down or the active
211 aggregator has no slaves.
212
213 This is the default value.
214
215 bandwidth or 1
216
217 The active aggregator is chosen by largest aggregate
218 bandwidth. Reselection occurs if:
219
220 - A slave is added to or removed from the bond
221
222 - Any slave's link state changes
223
224 - Any slave's 802.3ad association state changes
225
Matt LaPlante19f59462009-04-27 15:06:31 +0200226 - The bond's administrative state changes to up
Jay Vosburghfd989c82008-11-04 17:51:16 -0800227
228 count or 2
229
230 The active aggregator is chosen by the largest number of
231 ports (slaves). Reselection occurs as described under the
232 "bandwidth" setting, above.
233
234 The bandwidth and count selection policies permit failover of
235 802.3ad aggregations when partial failure of the active aggregator
236 occurs. This keeps the aggregator with the highest availability
237 (either in bandwidth or in number of ports) active at all times.
238
239 This option was added in bonding version 3.4.0.
240
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241arp_interval
242
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700243 Specifies the ARP link monitoring frequency in milliseconds.
Jay Vosburghf5b2b962006-09-22 21:54:53 -0700244
245 The ARP monitor works by periodically checking the slave
246 devices to determine whether they have sent or received
247 traffic recently (the precise criteria depends upon the
248 bonding mode, and the state of the slave). Regular traffic is
249 generated via ARP probes issued for the addresses specified by
250 the arp_ip_target option.
251
252 This behavior can be modified by the arp_validate option,
253 below.
254
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700255 If ARP monitoring is used in an etherchannel compatible mode
256 (modes 0 and 2), the switch should be configured in a mode
257 that evenly distributes packets across all links. If the
258 switch is configured to distribute the packets in an XOR
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 fashion, all replies from the ARP targets will be received on
260 the same link which could cause the other team members to
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700261 fail. ARP monitoring should not be used in conjunction with
262 miimon. A value of 0 disables ARP monitoring. The default
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 value is 0.
264
265arp_ip_target
266
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700267 Specifies the IP addresses to use as ARP monitoring peers when
268 arp_interval is > 0. These are the targets of the ARP request
269 sent to determine the health of the link to the targets.
270 Specify these values in ddd.ddd.ddd.ddd format. Multiple IP
271 addresses must be separated by a comma. At least one IP
272 address must be given for ARP monitoring to function. The
273 maximum number of targets that can be specified is 16. The
274 default value is no IP addresses.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275
Jay Vosburghf5b2b962006-09-22 21:54:53 -0700276arp_validate
277
278 Specifies whether or not ARP probes and replies should be
279 validated in the active-backup mode. This causes the ARP
280 monitor to examine the incoming ARP requests and replies, and
281 only consider a slave to be up if it is receiving the
282 appropriate ARP traffic.
283
284 Possible values are:
285
286 none or 0
287
288 No validation is performed. This is the default.
289
290 active or 1
291
292 Validation is performed only for the active slave.
293
294 backup or 2
295
296 Validation is performed only for backup slaves.
297
298 all or 3
299
300 Validation is performed for all slaves.
301
302 For the active slave, the validation checks ARP replies to
303 confirm that they were generated by an arp_ip_target. Since
304 backup slaves do not typically receive these replies, the
305 validation performed for backup slaves is on the ARP request
306 sent out via the active slave. It is possible that some
307 switch or network configurations may result in situations
308 wherein the backup slaves do not receive the ARP requests; in
309 such a situation, validation of backup slaves must be
310 disabled.
311
312 This option is useful in network configurations in which
313 multiple bonding hosts are concurrently issuing ARPs to one or
314 more targets beyond a common switch. Should the link between
315 the switch and target fail (but not the switch itself), the
316 probe traffic generated by the multiple bonding instances will
317 fool the standard ARP monitor into considering the links as
318 still up. Use of the arp_validate option can resolve this, as
319 the ARP monitor will only consider ARP requests and replies
320 associated with its own instance of bonding.
321
322 This option was added in bonding version 3.1.0.
323
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324downdelay
325
326 Specifies the time, in milliseconds, to wait before disabling
327 a slave after a link failure has been detected. This option
328 is only valid for the miimon link monitor. The downdelay
329 value should be a multiple of the miimon value; if not, it
330 will be rounded down to the nearest multiple. The default
331 value is 0.
332
Jay Vosburghdd957c52007-10-09 19:57:24 -0700333fail_over_mac
334
335 Specifies whether active-backup mode should set all slaves to
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700336 the same MAC address at enslavement (the traditional
337 behavior), or, when enabled, perform special handling of the
338 bond's MAC address in accordance with the selected policy.
Jay Vosburghdd957c52007-10-09 19:57:24 -0700339
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700340 Possible values are:
Jay Vosburghdd957c52007-10-09 19:57:24 -0700341
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700342 none or 0
Jay Vosburghdd957c52007-10-09 19:57:24 -0700343
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700344 This setting disables fail_over_mac, and causes
345 bonding to set all slaves of an active-backup bond to
346 the same MAC address at enslavement time. This is the
347 default.
Jay Vosburghdd957c52007-10-09 19:57:24 -0700348
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700349 active or 1
Jay Vosburghdd957c52007-10-09 19:57:24 -0700350
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700351 The "active" fail_over_mac policy indicates that the
352 MAC address of the bond should always be the MAC
353 address of the currently active slave. The MAC
354 address of the slaves is not changed; instead, the MAC
355 address of the bond changes during a failover.
356
357 This policy is useful for devices that cannot ever
358 alter their MAC address, or for devices that refuse
359 incoming broadcasts with their own source MAC (which
360 interferes with the ARP monitor).
361
362 The down side of this policy is that every device on
363 the network must be updated via gratuitous ARP,
364 vs. just updating a switch or set of switches (which
365 often takes place for any traffic, not just ARP
366 traffic, if the switch snoops incoming traffic to
367 update its tables) for the traditional method. If the
368 gratuitous ARP is lost, communication may be
369 disrupted.
370
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300371 When this policy is used in conjunction with the mii
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700372 monitor, devices which assert link up prior to being
373 able to actually transmit and receive are particularly
Matt LaPlante19f59462009-04-27 15:06:31 +0200374 susceptible to loss of the gratuitous ARP, and an
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700375 appropriate updelay setting may be required.
376
377 follow or 2
378
379 The "follow" fail_over_mac policy causes the MAC
380 address of the bond to be selected normally (normally
381 the MAC address of the first slave added to the bond).
382 However, the second and subsequent slaves are not set
383 to this MAC address while they are in a backup role; a
384 slave is programmed with the bond's MAC address at
385 failover time (and the formerly active slave receives
386 the newly active slave's MAC address).
387
388 This policy is useful for multiport devices that
389 either become confused or incur a performance penalty
390 when multiple ports are programmed with the same MAC
391 address.
392
393
394 The default policy is none, unless the first slave cannot
395 change its MAC address, in which case the active policy is
396 selected by default.
397
398 This option may be modified via sysfs only when no slaves are
399 present in the bond.
400
401 This option was added in bonding version 3.2.0. The "follow"
402 policy was added in bonding version 3.3.0.
Jay Vosburghdd957c52007-10-09 19:57:24 -0700403
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404lacp_rate
405
406 Option specifying the rate in which we'll ask our link partner
407 to transmit LACPDU packets in 802.3ad mode. Possible values
408 are:
409
410 slow or 0
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700411 Request partner to transmit LACPDUs every 30 seconds
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412
413 fast or 1
414 Request partner to transmit LACPDUs every 1 second
415
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700416 The default is slow.
417
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418max_bonds
419
420 Specifies the number of bonding devices to create for this
421 instance of the bonding driver. E.g., if max_bonds is 3, and
422 the bonding driver is not already loaded, then bond0, bond1
Jay Vosburghb8a97872008-06-13 18:12:04 -0700423 and bond2 will be created. The default value is 1. Specifying
424 a value of 0 will load bonding, but will not create any devices.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425
426miimon
427
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700428 Specifies the MII link monitoring frequency in milliseconds.
429 This determines how often the link state of each slave is
430 inspected for link failures. A value of zero disables MII
431 link monitoring. A value of 100 is a good starting point.
432 The use_carrier option, below, affects how the link state is
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 determined. See the High Availability section for additional
434 information. The default value is 0.
435
436mode
437
438 Specifies one of the bonding policies. The default is
439 balance-rr (round robin). Possible values are:
440
441 balance-rr or 0
442
443 Round-robin policy: Transmit packets in sequential
444 order from the first available slave through the
445 last. This mode provides load balancing and fault
446 tolerance.
447
448 active-backup or 1
449
450 Active-backup policy: Only one slave in the bond is
451 active. A different slave becomes active if, and only
452 if, the active slave fails. The bond's MAC address is
453 externally visible on only one port (network adapter)
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700454 to avoid confusing the switch.
455
456 In bonding version 2.6.2 or later, when a failover
457 occurs in active-backup mode, bonding will issue one
458 or more gratuitous ARPs on the newly active slave.
Auke Kok6224e012006-06-08 11:15:35 -0700459 One gratuitous ARP is issued for the bonding master
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700460 interface and each VLAN interfaces configured above
461 it, provided that the interface has at least one IP
462 address configured. Gratuitous ARPs issued for VLAN
463 interfaces are tagged with the appropriate VLAN id.
464
465 This mode provides fault tolerance. The primary
466 option, documented below, affects the behavior of this
467 mode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468
469 balance-xor or 2
470
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700471 XOR policy: Transmit based on the selected transmit
472 hash policy. The default policy is a simple [(source
473 MAC address XOR'd with destination MAC address) modulo
474 slave count]. Alternate transmit policies may be
475 selected via the xmit_hash_policy option, described
476 below.
477
478 This mode provides load balancing and fault tolerance.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479
480 broadcast or 3
481
482 Broadcast policy: transmits everything on all slave
483 interfaces. This mode provides fault tolerance.
484
485 802.3ad or 4
486
487 IEEE 802.3ad Dynamic link aggregation. Creates
488 aggregation groups that share the same speed and
489 duplex settings. Utilizes all slaves in the active
490 aggregator according to the 802.3ad specification.
491
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700492 Slave selection for outgoing traffic is done according
493 to the transmit hash policy, which may be changed from
494 the default simple XOR policy via the xmit_hash_policy
495 option, documented below. Note that not all transmit
496 policies may be 802.3ad compliant, particularly in
497 regards to the packet mis-ordering requirements of
498 section 43.2.4 of the 802.3ad standard. Differing
499 peer implementations will have varying tolerances for
500 noncompliance.
501
502 Prerequisites:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503
504 1. Ethtool support in the base drivers for retrieving
505 the speed and duplex of each slave.
506
507 2. A switch that supports IEEE 802.3ad Dynamic link
508 aggregation.
509
510 Most switches will require some type of configuration
511 to enable 802.3ad mode.
512
513 balance-tlb or 5
514
515 Adaptive transmit load balancing: channel bonding that
516 does not require any special switch support. The
517 outgoing traffic is distributed according to the
518 current load (computed relative to the speed) on each
519 slave. Incoming traffic is received by the current
520 slave. If the receiving slave fails, another slave
521 takes over the MAC address of the failed receiving
522 slave.
523
524 Prerequisite:
525
526 Ethtool support in the base drivers for retrieving the
527 speed of each slave.
528
529 balance-alb or 6
530
531 Adaptive load balancing: includes balance-tlb plus
532 receive load balancing (rlb) for IPV4 traffic, and
533 does not require any special switch support. The
534 receive load balancing is achieved by ARP negotiation.
535 The bonding driver intercepts the ARP Replies sent by
536 the local system on their way out and overwrites the
537 source hardware address with the unique hardware
538 address of one of the slaves in the bond such that
539 different peers use different hardware addresses for
540 the server.
541
542 Receive traffic from connections created by the server
543 is also balanced. When the local system sends an ARP
544 Request the bonding driver copies and saves the peer's
545 IP information from the ARP packet. When the ARP
546 Reply arrives from the peer, its hardware address is
547 retrieved and the bonding driver initiates an ARP
548 reply to this peer assigning it to one of the slaves
549 in the bond. A problematic outcome of using ARP
550 negotiation for balancing is that each time that an
551 ARP request is broadcast it uses the hardware address
552 of the bond. Hence, peers learn the hardware address
553 of the bond and the balancing of receive traffic
554 collapses to the current slave. This is handled by
555 sending updates (ARP Replies) to all the peers with
556 their individually assigned hardware address such that
557 the traffic is redistributed. Receive traffic is also
558 redistributed when a new slave is added to the bond
559 and when an inactive slave is re-activated. The
560 receive load is distributed sequentially (round robin)
561 among the group of highest speed slaves in the bond.
562
563 When a link is reconnected or a new slave joins the
564 bond the receive traffic is redistributed among all
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700565 active slaves in the bond by initiating ARP Replies
Auke Kok6224e012006-06-08 11:15:35 -0700566 with the selected MAC address to each of the
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 clients. The updelay parameter (detailed below) must
568 be set to a value equal or greater than the switch's
569 forwarding delay so that the ARP Replies sent to the
570 peers will not be blocked by the switch.
571
572 Prerequisites:
573
574 1. Ethtool support in the base drivers for retrieving
575 the speed of each slave.
576
577 2. Base driver support for setting the hardware
578 address of a device while it is open. This is
579 required so that there will always be one slave in the
580 team using the bond hardware address (the
581 curr_active_slave) while having a unique hardware
582 address for each slave in the bond. If the
583 curr_active_slave fails its hardware address is
584 swapped with the new curr_active_slave that was
585 chosen.
586
Jay Vosburghb59f9f72008-06-13 18:12:03 -0700587num_grat_arp
588
589 Specifies the number of gratuitous ARPs to be issued after a
590 failover event. One gratuitous ARP is issued immediately after
591 the failover, subsequent ARPs are sent at a rate of one per link
592 monitor interval (arp_interval or miimon, whichever is active).
593
594 The valid range is 0 - 255; the default value is 1. This option
595 affects only the active-backup mode. This option was added for
596 bonding version 3.3.0.
597
Brian Haley305d5522008-11-04 17:51:14 -0800598num_unsol_na
599
600 Specifies the number of unsolicited IPv6 Neighbor Advertisements
601 to be issued after a failover event. One unsolicited NA is issued
602 immediately after the failover.
603
604 The valid range is 0 - 255; the default value is 1. This option
605 affects only the active-backup mode. This option was added for
606 bonding version 3.4.0.
607
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608primary
609
610 A string (eth0, eth2, etc) specifying which slave is the
611 primary device. The specified device will always be the
612 active slave while it is available. Only when the primary is
613 off-line will alternate devices be used. This is useful when
614 one slave is preferred over another, e.g., when one slave has
615 higher throughput than another.
616
617 The primary option is only valid for active-backup mode.
618
Jiri Pirkoa5499522009-09-25 03:28:09 +0000619primary_reselect
620
621 Specifies the reselection policy for the primary slave. This
622 affects how the primary slave is chosen to become the active slave
623 when failure of the active slave or recovery of the primary slave
624 occurs. This option is designed to prevent flip-flopping between
625 the primary slave and other slaves. Possible values are:
626
627 always or 0 (default)
628
629 The primary slave becomes the active slave whenever it
630 comes back up.
631
632 better or 1
633
634 The primary slave becomes the active slave when it comes
635 back up, if the speed and duplex of the primary slave is
636 better than the speed and duplex of the current active
637 slave.
638
639 failure or 2
640
641 The primary slave becomes the active slave only if the
642 current active slave fails and the primary slave is up.
643
644 The primary_reselect setting is ignored in two cases:
645
646 If no slaves are active, the first slave to recover is
647 made the active slave.
648
649 When initially enslaved, the primary slave is always made
650 the active slave.
651
652 Changing the primary_reselect policy via sysfs will cause an
653 immediate selection of the best active slave according to the new
654 policy. This may or may not result in a change of the active
655 slave, depending upon the circumstances.
656
657 This option was added for bonding version 3.6.0.
658
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659updelay
660
661 Specifies the time, in milliseconds, to wait before enabling a
662 slave after a link recovery has been detected. This option is
663 only valid for the miimon link monitor. The updelay value
664 should be a multiple of the miimon value; if not, it will be
665 rounded down to the nearest multiple. The default value is 0.
666
667use_carrier
668
669 Specifies whether or not miimon should use MII or ETHTOOL
670 ioctls vs. netif_carrier_ok() to determine the link
671 status. The MII or ETHTOOL ioctls are less efficient and
672 utilize a deprecated calling sequence within the kernel. The
673 netif_carrier_ok() relies on the device driver to maintain its
674 state with netif_carrier_on/off; at this writing, most, but
675 not all, device drivers support this facility.
676
677 If bonding insists that the link is up when it should not be,
678 it may be that your network device driver does not support
679 netif_carrier_on/off. The default state for netif_carrier is
680 "carrier on," so if a driver does not support netif_carrier,
681 it will appear as if the link is always up. In this case,
682 setting use_carrier to 0 will cause bonding to revert to the
683 MII / ETHTOOL ioctl method to determine the link state.
684
685 A value of 1 enables the use of netif_carrier_ok(), a value of
686 0 will use the deprecated MII / ETHTOOL ioctls. The default
687 value is 1.
688
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700689xmit_hash_policy
690
691 Selects the transmit hash policy to use for slave selection in
692 balance-xor and 802.3ad modes. Possible values are:
693
694 layer2
695
696 Uses XOR of hardware MAC addresses to generate the
697 hash. The formula is
698
699 (source MAC XOR destination MAC) modulo slave count
700
701 This algorithm will place all traffic to a particular
702 network peer on the same slave.
703
704 This algorithm is 802.3ad compliant.
705
Jay Vosburgh6f6652b2007-12-06 23:40:34 -0800706 layer2+3
707
708 This policy uses a combination of layer2 and layer3
709 protocol information to generate the hash.
710
711 Uses XOR of hardware MAC addresses and IP addresses to
712 generate the hash. The formula is
713
714 (((source IP XOR dest IP) AND 0xffff) XOR
715 ( source MAC XOR destination MAC ))
716 modulo slave count
717
718 This algorithm will place all traffic to a particular
719 network peer on the same slave. For non-IP traffic,
720 the formula is the same as for the layer2 transmit
721 hash policy.
722
723 This policy is intended to provide a more balanced
724 distribution of traffic than layer2 alone, especially
725 in environments where a layer3 gateway device is
726 required to reach most destinations.
727
Matt LaPlanted9195882008-07-25 19:45:33 -0700728 This algorithm is 802.3ad compliant.
Jay Vosburgh6f6652b2007-12-06 23:40:34 -0800729
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700730 layer3+4
731
732 This policy uses upper layer protocol information,
733 when available, to generate the hash. This allows for
734 traffic to a particular network peer to span multiple
735 slaves, although a single connection will not span
736 multiple slaves.
737
738 The formula for unfragmented TCP and UDP packets is
739
740 ((source port XOR dest port) XOR
741 ((source IP XOR dest IP) AND 0xffff)
742 modulo slave count
743
744 For fragmented TCP or UDP packets and all other IP
745 protocol traffic, the source and destination port
746 information is omitted. For non-IP traffic, the
747 formula is the same as for the layer2 transmit hash
748 policy.
749
750 This policy is intended to mimic the behavior of
751 certain switches, notably Cisco switches with PFC2 as
752 well as some Foundry and IBM products.
753
754 This algorithm is not fully 802.3ad compliant. A
755 single TCP or UDP conversation containing both
756 fragmented and unfragmented packets will see packets
757 striped across two interfaces. This may result in out
758 of order delivery. Most traffic types will not meet
759 this criteria, as TCP rarely fragments traffic, and
760 most UDP traffic is not involved in extended
761 conversations. Other implementations of 802.3ad may
762 or may not tolerate this noncompliance.
763
764 The default value is layer2. This option was added in bonding
Jay Vosburgh6f6652b2007-12-06 23:40:34 -0800765 version 2.6.3. In earlier versions of bonding, this parameter
766 does not exist, and the layer2 policy is the only policy. The
767 layer2+3 value was added for bonding version 3.2.2.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768
Flavio Leitnerc2952c32010-10-05 14:23:59 +0000769resend_igmp
770
771 Specifies the number of IGMP membership reports to be issued after
772 a failover event. One membership report is issued immediately after
773 the failover, subsequent packets are sent in each 200ms interval.
774
775 The valid range is 0 - 255; the default value is 1. This option
776 was added for bonding version 3.7.0.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777
7783. Configuring Bonding Devices
779==============================
780
Auke Kok6224e012006-06-08 11:15:35 -0700781 You can configure bonding using either your distro's network
782initialization scripts, or manually using either ifenslave or the
Nicolas de Pesloüande221bd2011-01-24 13:21:37 +0000783sysfs interface. Distros generally use one of three packages for the
784network initialization scripts: initscripts, sysconfig or interfaces.
785Recent versions of these packages have support for bonding, while older
Auke Kok6224e012006-06-08 11:15:35 -0700786versions do not.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787
788 We will first describe the options for configuring bonding for
Nicolas de Pesloüande221bd2011-01-24 13:21:37 +0000789distros using versions of initscripts, sysconfig and interfaces with full
790or partial support for bonding, then provide information on enabling
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791bonding without support from the network initialization scripts (i.e.,
792older versions of initscripts or sysconfig).
793
Nicolas de Pesloüande221bd2011-01-24 13:21:37 +0000794 If you're unsure whether your distro uses sysconfig,
795initscripts or interfaces, or don't know if it's new enough, have no fear.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796Determining this is fairly straightforward.
797
Nicolas de Pesloüande221bd2011-01-24 13:21:37 +0000798 First, look for a file called interfaces in /etc/network directory.
799If this file is present in your system, then your system use interfaces. See
800Configuration with Interfaces Support.
801
802 Else, issue the command:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803
804$ rpm -qf /sbin/ifup
805
806 It will respond with a line of text starting with either
807"initscripts" or "sysconfig," followed by some numbers. This is the
808package that provides your network initialization scripts.
809
810 Next, to determine if your installation supports bonding,
811issue the command:
812
813$ grep ifenslave /sbin/ifup
814
815 If this returns any matches, then your initscripts or
816sysconfig has support for bonding.
817
Auke Kok6224e012006-06-08 11:15:35 -07008183.1 Configuration with Sysconfig Support
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819----------------------------------------
820
821 This section applies to distros using a version of sysconfig
822with bonding support, for example, SuSE Linux Enterprise Server 9.
823
824 SuSE SLES 9's networking configuration system does support
825bonding, however, at this writing, the YaST system configuration
Auke Kok6224e012006-06-08 11:15:35 -0700826front end does not provide any means to work with bonding devices.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827Bonding devices can be managed by hand, however, as follows.
828
829 First, if they have not already been configured, configure the
830slave devices. On SLES 9, this is most easily done by running the
831yast2 sysconfig configuration utility. The goal is for to create an
832ifcfg-id file for each slave device. The simplest way to accomplish
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700833this is to configure the devices for DHCP (this is only to get the
834file ifcfg-id file created; see below for some issues with DHCP). The
835name of the configuration file for each device will be of the form:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836
837ifcfg-id-xx:xx:xx:xx:xx:xx
838
839 Where the "xx" portion will be replaced with the digits from
840the device's permanent MAC address.
841
842 Once the set of ifcfg-id-xx:xx:xx:xx:xx:xx files has been
843created, it is necessary to edit the configuration files for the slave
844devices (the MAC addresses correspond to those of the slave devices).
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700845Before editing, the file will contain multiple lines, and will look
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846something like this:
847
848BOOTPROTO='dhcp'
849STARTMODE='on'
850USERCTL='no'
851UNIQUE='XNzu.WeZGOGF+4wE'
852_nm_name='bus-pci-0001:61:01.0'
853
854 Change the BOOTPROTO and STARTMODE lines to the following:
855
856BOOTPROTO='none'
857STARTMODE='off'
858
859 Do not alter the UNIQUE or _nm_name lines. Remove any other
860lines (USERCTL, etc).
861
862 Once the ifcfg-id-xx:xx:xx:xx:xx:xx files have been modified,
863it's time to create the configuration file for the bonding device
864itself. This file is named ifcfg-bondX, where X is the number of the
865bonding device to create, starting at 0. The first such file is
866ifcfg-bond0, the second is ifcfg-bond1, and so on. The sysconfig
867network configuration system will correctly start multiple instances
868of bonding.
869
870 The contents of the ifcfg-bondX file is as follows:
871
872BOOTPROTO="static"
873BROADCAST="10.0.2.255"
874IPADDR="10.0.2.10"
875NETMASK="255.255.0.0"
876NETWORK="10.0.2.0"
877REMOTE_IPADDR=""
878STARTMODE="onboot"
879BONDING_MASTER="yes"
880BONDING_MODULE_OPTS="mode=active-backup miimon=100"
881BONDING_SLAVE0="eth0"
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700882BONDING_SLAVE1="bus-pci-0000:06:08.1"
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883
884 Replace the sample BROADCAST, IPADDR, NETMASK and NETWORK
885values with the appropriate values for your network.
886
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 The STARTMODE specifies when the device is brought online.
888The possible values are:
889
890 onboot: The device is started at boot time. If you're not
891 sure, this is probably what you want.
892
893 manual: The device is started only when ifup is called
894 manually. Bonding devices may be configured this
895 way if you do not wish them to start automatically
896 at boot for some reason.
897
898 hotplug: The device is started by a hotplug event. This is not
899 a valid choice for a bonding device.
900
901 off or ignore: The device configuration is ignored.
902
903 The line BONDING_MASTER='yes' indicates that the device is a
904bonding master device. The only useful value is "yes."
905
906 The contents of BONDING_MODULE_OPTS are supplied to the
907instance of the bonding module for this device. Specify the options
908for the bonding mode, link monitoring, and so on here. Do not include
909the max_bonds bonding parameter; this will confuse the configuration
910system if you have multiple bonding devices.
911
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700912 Finally, supply one BONDING_SLAVEn="slave device" for each
913slave. where "n" is an increasing value, one for each slave. The
914"slave device" is either an interface name, e.g., "eth0", or a device
915specifier for the network device. The interface name is easier to
916find, but the ethN names are subject to change at boot time if, e.g.,
917a device early in the sequence has failed. The device specifiers
918(bus-pci-0000:06:08.1 in the example above) specify the physical
919network device, and will not change unless the device's bus location
920changes (for example, it is moved from one PCI slot to another). The
921example above uses one of each type for demonstration purposes; most
922configurations will choose one or the other for all slave devices.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923
924 When all configuration files have been modified or created,
925networking must be restarted for the configuration changes to take
926effect. This can be accomplished via the following:
927
928# /etc/init.d/network restart
929
930 Note that the network control script (/sbin/ifdown) will
931remove the bonding module as part of the network shutdown processing,
932so it is not necessary to remove the module by hand if, e.g., the
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700933module parameters have changed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934
935 Also, at this writing, YaST/YaST2 will not manage bonding
936devices (they do not show bonding interfaces on its list of network
937devices). It is necessary to edit the configuration file by hand to
938change the bonding configuration.
939
940 Additional general options and details of the ifcfg file
941format can be found in an example ifcfg template file:
942
943/etc/sysconfig/network/ifcfg.template
944
945 Note that the template does not document the various BONDING_
946settings described above, but does describe many of the other options.
947
Auke Kok6224e012006-06-08 11:15:35 -07009483.1.1 Using DHCP with Sysconfig
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700949-------------------------------
950
951 Under sysconfig, configuring a device with BOOTPROTO='dhcp'
952will cause it to query DHCP for its IP address information. At this
953writing, this does not function for bonding devices; the scripts
954attempt to obtain the device address from DHCP prior to adding any of
955the slave devices. Without active slaves, the DHCP requests are not
956sent to the network.
957
Auke Kok6224e012006-06-08 11:15:35 -07009583.1.2 Configuring Multiple Bonds with Sysconfig
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700959-----------------------------------------------
960
961 The sysconfig network initialization system is capable of
962handling multiple bonding devices. All that is necessary is for each
963bonding instance to have an appropriately configured ifcfg-bondX file
964(as described above). Do not specify the "max_bonds" parameter to any
965instance of bonding, as this will confuse sysconfig. If you require
966multiple bonding devices with identical parameters, create multiple
967ifcfg-bondX files.
968
969 Because the sysconfig scripts supply the bonding module
970options in the ifcfg-bondX file, it is not necessary to add them to
971the system /etc/modules.conf or /etc/modprobe.conf configuration file.
972
Auke Kok6224e012006-06-08 11:15:35 -07009733.2 Configuration with Initscripts Support
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974------------------------------------------
975
Jay Vosburgh9a6c6862007-11-13 20:25:48 -0800976 This section applies to distros using a recent version of
977initscripts with bonding support, for example, Red Hat Enterprise Linux
978version 3 or later, Fedora, etc. On these systems, the network
979initialization scripts have knowledge of bonding, and can be configured to
980control bonding devices. Note that older versions of the initscripts
981package have lower levels of support for bonding; this will be noted where
982applicable.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983
984 These distros will not automatically load the network adapter
985driver unless the ethX device is configured with an IP address.
986Because of this constraint, users must manually configure a
987network-script file for all physical adapters that will be members of
988a bondX link. Network script files are located in the directory:
989
990/etc/sysconfig/network-scripts
991
992 The file name must be prefixed with "ifcfg-eth" and suffixed
993with the adapter's physical adapter number. For example, the script
994for eth0 would be named /etc/sysconfig/network-scripts/ifcfg-eth0.
995Place the following text in the file:
996
997DEVICE=eth0
998USERCTL=no
999ONBOOT=yes
1000MASTER=bond0
1001SLAVE=yes
1002BOOTPROTO=none
1003
1004 The DEVICE= line will be different for every ethX device and
1005must correspond with the name of the file, i.e., ifcfg-eth1 must have
1006a device line of DEVICE=eth1. The setting of the MASTER= line will
1007also depend on the final bonding interface name chosen for your bond.
1008As with other network devices, these typically start at 0, and go up
1009one for each device, i.e., the first bonding instance is bond0, the
1010second is bond1, and so on.
1011
1012 Next, create a bond network script. The file name for this
1013script will be /etc/sysconfig/network-scripts/ifcfg-bondX where X is
1014the number of the bond. For bond0 the file is named "ifcfg-bond0",
1015for bond1 it is named "ifcfg-bond1", and so on. Within that file,
1016place the following text:
1017
1018DEVICE=bond0
1019IPADDR=192.168.1.1
1020NETMASK=255.255.255.0
1021NETWORK=192.168.1.0
1022BROADCAST=192.168.1.255
1023ONBOOT=yes
1024BOOTPROTO=none
1025USERCTL=no
1026
1027 Be sure to change the networking specific lines (IPADDR,
1028NETMASK, NETWORK and BROADCAST) to match your network configuration.
1029
Jay Vosburgh9a6c6862007-11-13 20:25:48 -08001030 For later versions of initscripts, such as that found with Fedora
Andy Gospodarek3f8b4b12008-10-22 11:19:48 +000010317 (or later) and Red Hat Enterprise Linux version 5 (or later), it is possible,
1032and, indeed, preferable, to specify the bonding options in the ifcfg-bond0
Jay Vosburgh9a6c6862007-11-13 20:25:48 -08001033file, e.g. a line of the format:
1034
Andy Gospodarek3f8b4b12008-10-22 11:19:48 +00001035BONDING_OPTS="mode=active-backup arp_interval=60 arp_ip_target=192.168.1.254"
Jay Vosburgh9a6c6862007-11-13 20:25:48 -08001036
1037 will configure the bond with the specified options. The options
1038specified in BONDING_OPTS are identical to the bonding module parameters
Andy Gospodarek3f8b4b12008-10-22 11:19:48 +00001039except for the arp_ip_target field when using versions of initscripts older
1040than and 8.57 (Fedora 8) and 8.45.19 (Red Hat Enterprise Linux 5.2). When
1041using older versions each target should be included as a separate option and
1042should be preceded by a '+' to indicate it should be added to the list of
1043queried targets, e.g.,
Jay Vosburgh9a6c6862007-11-13 20:25:48 -08001044
1045 arp_ip_target=+192.168.1.1 arp_ip_target=+192.168.1.2
1046
1047 is the proper syntax to specify multiple targets. When specifying
1048options via BONDING_OPTS, it is not necessary to edit /etc/modules.conf or
1049/etc/modprobe.conf.
1050
Andy Gospodarek3f8b4b12008-10-22 11:19:48 +00001051 For even older versions of initscripts that do not support
Jay Vosburgh9a6c6862007-11-13 20:25:48 -08001052BONDING_OPTS, it is necessary to edit /etc/modules.conf (or
1053/etc/modprobe.conf, depending upon your distro) to load the bonding module
1054with your desired options when the bond0 interface is brought up. The
1055following lines in /etc/modules.conf (or modprobe.conf) will load the
1056bonding module, and select its options:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057
1058alias bond0 bonding
1059options bond0 mode=balance-alb miimon=100
1060
1061 Replace the sample parameters with the appropriate set of
1062options for your configuration.
1063
1064 Finally run "/etc/rc.d/init.d/network restart" as root. This
1065will restart the networking subsystem and your bond link should be now
1066up and running.
1067
Auke Kok6224e012006-06-08 11:15:35 -070010683.2.1 Using DHCP with Initscripts
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001069---------------------------------
1070
Jay Vosburgh9a6c6862007-11-13 20:25:48 -08001071 Recent versions of initscripts (the versions supplied with Fedora
1072Core 3 and Red Hat Enterprise Linux 4, or later versions, are reported to
1073work) have support for assigning IP information to bonding devices via
1074DHCP.
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001075
1076 To configure bonding for DHCP, configure it as described
1077above, except replace the line "BOOTPROTO=none" with "BOOTPROTO=dhcp"
1078and add a line consisting of "TYPE=Bonding". Note that the TYPE value
1079is case sensitive.
1080
Auke Kok6224e012006-06-08 11:15:35 -070010813.2.2 Configuring Multiple Bonds with Initscripts
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001082-------------------------------------------------
1083
Jay Vosburgh9a6c6862007-11-13 20:25:48 -08001084 Initscripts packages that are included with Fedora 7 and Red Hat
1085Enterprise Linux 5 support multiple bonding interfaces by simply
1086specifying the appropriate BONDING_OPTS= in ifcfg-bondX where X is the
1087number of the bond. This support requires sysfs support in the kernel,
1088and a bonding driver of version 3.0.0 or later. Other configurations may
1089not support this method for specifying multiple bonding interfaces; for
1090those instances, see the "Configuring Multiple Bonds Manually" section,
1091below.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092
Auke Kok6224e012006-06-08 11:15:35 -070010933.3 Configuring Bonding Manually with Ifenslave
1094-----------------------------------------------
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095
1096 This section applies to distros whose network initialization
1097scripts (the sysconfig or initscripts package) do not have specific
1098knowledge of bonding. One such distro is SuSE Linux Enterprise Server
1099version 8.
1100
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001101 The general method for these systems is to place the bonding
1102module parameters into /etc/modules.conf or /etc/modprobe.conf (as
1103appropriate for the installed distro), then add modprobe and/or
1104ifenslave commands to the system's global init script. The name of
1105the global init script differs; for sysconfig, it is
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106/etc/init.d/boot.local and for initscripts it is /etc/rc.d/rc.local.
1107
1108 For example, if you wanted to make a simple bond of two e100
1109devices (presumed to be eth0 and eth1), and have it persist across
1110reboots, edit the appropriate file (/etc/init.d/boot.local or
1111/etc/rc.d/rc.local), and add the following:
1112
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001113modprobe bonding mode=balance-alb miimon=100
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114modprobe e100
1115ifconfig bond0 192.168.1.1 netmask 255.255.255.0 up
1116ifenslave bond0 eth0
1117ifenslave bond0 eth1
1118
1119 Replace the example bonding module parameters and bond0
1120network configuration (IP address, netmask, etc) with the appropriate
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001121values for your configuration.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122
1123 Unfortunately, this method will not provide support for the
1124ifup and ifdown scripts on the bond devices. To reload the bonding
1125configuration, it is necessary to run the initialization script, e.g.,
1126
1127# /etc/init.d/boot.local
1128
1129 or
1130
1131# /etc/rc.d/rc.local
1132
1133 It may be desirable in such a case to create a separate script
1134which only initializes the bonding configuration, then call that
1135separate script from within boot.local. This allows for bonding to be
1136enabled without re-running the entire global init script.
1137
1138 To shut down the bonding devices, it is necessary to first
1139mark the bonding device itself as being down, then remove the
1140appropriate device driver modules. For our example above, you can do
1141the following:
1142
1143# ifconfig bond0 down
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001144# rmmod bonding
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145# rmmod e100
1146
1147 Again, for convenience, it may be desirable to create a script
1148with these commands.
1149
1150
Jay Vosburgh00354cf2005-07-21 12:18:02 -070011513.3.1 Configuring Multiple Bonds Manually
1152-----------------------------------------
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153
1154 This section contains information on configuring multiple
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001155bonding devices with differing options for those systems whose network
1156initialization scripts lack support for configuring multiple bonds.
1157
1158 If you require multiple bonding devices, but all with the same
1159options, you may wish to use the "max_bonds" module parameter,
1160documented above.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161
Jay Vosburgh9a6c6862007-11-13 20:25:48 -08001162 To create multiple bonding devices with differing options, it is
1163preferrable to use bonding parameters exported by sysfs, documented in the
1164section below.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165
Jay Vosburgh9a6c6862007-11-13 20:25:48 -08001166 For versions of bonding without sysfs support, the only means to
1167provide multiple instances of bonding with differing options is to load
1168the bonding driver multiple times. Note that current versions of the
1169sysconfig network initialization scripts handle this automatically; if
1170your distro uses these scripts, no special action is needed. See the
1171section Configuring Bonding Devices, above, if you're not sure about your
1172network initialization scripts.
1173
1174 To load multiple instances of the module, it is necessary to
1175specify a different name for each instance (the module loading system
1176requires that every loaded module, even multiple instances of the same
1177module, have a unique name). This is accomplished by supplying multiple
1178sets of bonding options in /etc/modprobe.conf, for example:
1179
1180alias bond0 bonding
1181options bond0 -o bond0 mode=balance-rr miimon=100
1182
1183alias bond1 bonding
1184options bond1 -o bond1 mode=balance-alb miimon=50
1185
1186 will load the bonding module two times. The first instance is
1187named "bond0" and creates the bond0 device in balance-rr mode with an
1188miimon of 100. The second instance is named "bond1" and creates the
1189bond1 device in balance-alb mode with an miimon of 50.
1190
1191 In some circumstances (typically with older distributions),
1192the above does not work, and the second bonding instance never sees
1193its options. In that case, the second options line can be substituted
1194as follows:
1195
1196install bond1 /sbin/modprobe --ignore-install bonding -o bond1 \
1197 mode=balance-alb miimon=50
1198
1199 This may be repeated any number of times, specifying a new and
1200unique name in place of bond1 for each subsequent instance.
1201
1202 It has been observed that some Red Hat supplied kernels are unable
1203to rename modules at load time (the "-o bond1" part). Attempts to pass
1204that option to modprobe will produce an "Operation not permitted" error.
1205This has been reported on some Fedora Core kernels, and has been seen on
1206RHEL 4 as well. On kernels exhibiting this problem, it will be impossible
1207to configure multiple bonds with differing parameters (as they are older
1208kernels, and also lack sysfs support).
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001209
Auke Kok6224e012006-06-08 11:15:35 -070012103.4 Configuring Bonding Manually via Sysfs
1211------------------------------------------
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212
Jay Vosburgh9a6c6862007-11-13 20:25:48 -08001213 Starting with version 3.0.0, Channel Bonding may be configured
Auke Kok6224e012006-06-08 11:15:35 -07001214via the sysfs interface. This interface allows dynamic configuration
1215of all bonds in the system without unloading the module. It also
1216allows for adding and removing bonds at runtime. Ifenslave is no
1217longer required, though it is still supported.
1218
1219 Use of the sysfs interface allows you to use multiple bonds
1220with different configurations without having to reload the module.
1221It also allows you to use multiple, differently configured bonds when
1222bonding is compiled into the kernel.
1223
1224 You must have the sysfs filesystem mounted to configure
1225bonding this way. The examples in this document assume that you
1226are using the standard mount point for sysfs, e.g. /sys. If your
1227sysfs filesystem is mounted elsewhere, you will need to adjust the
1228example paths accordingly.
1229
1230Creating and Destroying Bonds
1231-----------------------------
1232To add a new bond foo:
1233# echo +foo > /sys/class/net/bonding_masters
1234
1235To remove an existing bond bar:
1236# echo -bar > /sys/class/net/bonding_masters
1237
1238To show all existing bonds:
1239# cat /sys/class/net/bonding_masters
1240
1241NOTE: due to 4K size limitation of sysfs files, this list may be
1242truncated if you have more than a few hundred bonds. This is unlikely
1243to occur under normal operating conditions.
1244
1245Adding and Removing Slaves
1246--------------------------
1247 Interfaces may be enslaved to a bond using the file
1248/sys/class/net/<bond>/bonding/slaves. The semantics for this file
1249are the same as for the bonding_masters file.
1250
1251To enslave interface eth0 to bond bond0:
1252# ifconfig bond0 up
1253# echo +eth0 > /sys/class/net/bond0/bonding/slaves
1254
1255To free slave eth0 from bond bond0:
1256# echo -eth0 > /sys/class/net/bond0/bonding/slaves
1257
Auke Kok6224e012006-06-08 11:15:35 -07001258 When an interface is enslaved to a bond, symlinks between the
1259two are created in the sysfs filesystem. In this case, you would get
1260/sys/class/net/bond0/slave_eth0 pointing to /sys/class/net/eth0, and
1261/sys/class/net/eth0/master pointing to /sys/class/net/bond0.
1262
1263 This means that you can tell quickly whether or not an
1264interface is enslaved by looking for the master symlink. Thus:
1265# echo -eth0 > /sys/class/net/eth0/master/bonding/slaves
1266will free eth0 from whatever bond it is enslaved to, regardless of
1267the name of the bond interface.
1268
1269Changing a Bond's Configuration
1270-------------------------------
1271 Each bond may be configured individually by manipulating the
1272files located in /sys/class/net/<bond name>/bonding
1273
1274 The names of these files correspond directly with the command-
Paolo Ornati670e9f32006-10-03 22:57:56 +02001275line parameters described elsewhere in this file, and, with the
Auke Kok6224e012006-06-08 11:15:35 -07001276exception of arp_ip_target, they accept the same values. To see the
1277current setting, simply cat the appropriate file.
1278
1279 A few examples will be given here; for specific usage
1280guidelines for each parameter, see the appropriate section in this
1281document.
1282
1283To configure bond0 for balance-alb mode:
1284# ifconfig bond0 down
1285# echo 6 > /sys/class/net/bond0/bonding/mode
1286 - or -
1287# echo balance-alb > /sys/class/net/bond0/bonding/mode
1288 NOTE: The bond interface must be down before the mode can be
1289changed.
1290
1291To enable MII monitoring on bond0 with a 1 second interval:
1292# echo 1000 > /sys/class/net/bond0/bonding/miimon
1293 NOTE: If ARP monitoring is enabled, it will disabled when MII
1294monitoring is enabled, and vice-versa.
1295
1296To add ARP targets:
1297# echo +192.168.0.100 > /sys/class/net/bond0/bonding/arp_ip_target
1298# echo +192.168.0.101 > /sys/class/net/bond0/bonding/arp_ip_target
Brian Haley5a31bec2009-04-13 00:11:30 -07001299 NOTE: up to 16 target addresses may be specified.
Auke Kok6224e012006-06-08 11:15:35 -07001300
1301To remove an ARP target:
1302# echo -192.168.0.100 > /sys/class/net/bond0/bonding/arp_ip_target
1303
1304Example Configuration
1305---------------------
1306 We begin with the same example that is shown in section 3.3,
1307executed with sysfs, and without using ifenslave.
1308
1309 To make a simple bond of two e100 devices (presumed to be eth0
1310and eth1), and have it persist across reboots, edit the appropriate
1311file (/etc/init.d/boot.local or /etc/rc.d/rc.local), and add the
1312following:
1313
1314modprobe bonding
1315modprobe e100
1316echo balance-alb > /sys/class/net/bond0/bonding/mode
1317ifconfig bond0 192.168.1.1 netmask 255.255.255.0 up
1318echo 100 > /sys/class/net/bond0/bonding/miimon
1319echo +eth0 > /sys/class/net/bond0/bonding/slaves
1320echo +eth1 > /sys/class/net/bond0/bonding/slaves
1321
1322 To add a second bond, with two e1000 interfaces in
1323active-backup mode, using ARP monitoring, add the following lines to
1324your init script:
1325
1326modprobe e1000
1327echo +bond1 > /sys/class/net/bonding_masters
1328echo active-backup > /sys/class/net/bond1/bonding/mode
1329ifconfig bond1 192.168.2.1 netmask 255.255.255.0 up
1330echo +192.168.2.100 /sys/class/net/bond1/bonding/arp_ip_target
1331echo 2000 > /sys/class/net/bond1/bonding/arp_interval
1332echo +eth2 > /sys/class/net/bond1/bonding/slaves
1333echo +eth3 > /sys/class/net/bond1/bonding/slaves
1334
Nicolas de Pesloüande221bd2011-01-24 13:21:37 +000013353.5 Configuration with Interfaces Support
1336-----------------------------------------
1337
1338 This section applies to distros which use /etc/network/interfaces file
1339to describe network interface configuration, most notably Debian and it's
1340derivatives.
1341
1342 The ifup and ifdown commands on Debian don't support bonding out of
1343the box. The ifenslave-2.6 package should be installed to provide bonding
1344support. Once installed, this package will provide bond-* options to be used
1345into /etc/network/interfaces.
1346
1347 Note that ifenslave-2.6 package will load the bonding module and use
1348the ifenslave command when appropriate.
1349
1350Example Configurations
1351----------------------
1352
1353In /etc/network/interfaces, the following stanza will configure bond0, in
1354active-backup mode, with eth0 and eth1 as slaves.
1355
1356auto bond0
1357iface bond0 inet dhcp
1358 bond-slaves eth0 eth1
1359 bond-mode active-backup
1360 bond-miimon 100
1361 bond-primary eth0 eth1
1362
1363If the above configuration doesn't work, you might have a system using
1364upstart for system startup. This is most notably true for recent
1365Ubuntu versions. The following stanza in /etc/network/interfaces will
1366produce the same result on those systems.
1367
1368auto bond0
1369iface bond0 inet dhcp
1370 bond-slaves none
1371 bond-mode active-backup
1372 bond-miimon 100
1373
1374auto eth0
1375iface eth0 inet manual
1376 bond-master bond0
1377 bond-primary eth0 eth1
1378
1379auto eth1
1380iface eth1 inet manual
1381 bond-master bond0
1382 bond-primary eth0 eth1
1383
1384For a full list of bond-* supported options in /etc/network/interfaces and some
1385more advanced examples tailored to you particular distros, see the files in
1386/usr/share/doc/ifenslave-2.6.
1387
13883.6 Overriding Configuration for Special Cases
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00001389----------------------------------------------
Nicolas de Pesloüande221bd2011-01-24 13:21:37 +00001390
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00001391When using the bonding driver, the physical port which transmits a frame is
1392typically selected by the bonding driver, and is not relevant to the user or
1393system administrator. The output port is simply selected using the policies of
1394the selected bonding mode. On occasion however, it is helpful to direct certain
1395classes of traffic to certain physical interfaces on output to implement
1396slightly more complex policies. For example, to reach a web server over a
1397bonded interface in which eth0 connects to a private network, while eth1
1398connects via a public network, it may be desirous to bias the bond to send said
1399traffic over eth0 first, using eth1 only as a fall back, while all other traffic
1400can safely be sent over either interface. Such configurations may be achieved
1401using the traffic control utilities inherent in linux.
Auke Kok6224e012006-06-08 11:15:35 -07001402
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00001403By default the bonding driver is multiqueue aware and 16 queues are created
1404when the driver initializes (see Documentation/networking/multiqueue.txt
1405for details). If more or less queues are desired the module parameter
1406tx_queues can be used to change this value. There is no sysfs parameter
1407available as the allocation is done at module init time.
1408
1409The output of the file /proc/net/bonding/bondX has changed so the output Queue
1410ID is now printed for each slave:
1411
1412Bonding Mode: fault-tolerance (active-backup)
1413Primary Slave: None
1414Currently Active Slave: eth0
1415MII Status: up
1416MII Polling Interval (ms): 0
1417Up Delay (ms): 0
1418Down Delay (ms): 0
1419
1420Slave Interface: eth0
1421MII Status: up
1422Link Failure Count: 0
1423Permanent HW addr: 00:1a:a0:12:8f:cb
1424Slave queue ID: 0
1425
1426Slave Interface: eth1
1427MII Status: up
1428Link Failure Count: 0
1429Permanent HW addr: 00:1a:a0:12:8f:cc
1430Slave queue ID: 2
1431
1432The queue_id for a slave can be set using the command:
1433
1434# echo "eth1:2" > /sys/class/net/bond0/bonding/queue_id
1435
1436Any interface that needs a queue_id set should set it with multiple calls
1437like the one above until proper priorities are set for all interfaces. On
1438distributions that allow configuration via initscripts, multiple 'queue_id'
1439arguments can be added to BONDING_OPTS to set all needed slave queues.
1440
1441These queue id's can be used in conjunction with the tc utility to configure
1442a multiqueue qdisc and filters to bias certain traffic to transmit on certain
1443slave devices. For instance, say we wanted, in the above configuration to
1444force all traffic bound to 192.168.1.100 to use eth1 in the bond as its output
1445device. The following commands would accomplish this:
1446
1447# tc qdisc add dev bond0 handle 1 root multiq
1448
1449# tc filter add dev bond0 protocol ip parent 1: prio 1 u32 match ip dst \
1450 192.168.1.100 action skbedit queue_mapping 2
1451
1452These commands tell the kernel to attach a multiqueue queue discipline to the
1453bond0 interface and filter traffic enqueued to it, such that packets with a dst
1454ip of 192.168.1.100 have their output queue mapping value overwritten to 2.
1455This value is then passed into the driver, causing the normal output path
1456selection policy to be overridden, selecting instead qid 2, which maps to eth1.
1457
1458Note that qid values begin at 1. Qid 0 is reserved to initiate to the driver
1459that normal output policy selection should take place. One benefit to simply
1460leaving the qid for a slave to 0 is the multiqueue awareness in the bonding
1461driver that is now present. This awareness allows tc filters to be placed on
1462slave devices as well as bond devices and the bonding driver will simply act as
1463a pass-through for selecting output queues on the slave device rather than
1464output port selection.
1465
1466This feature first appeared in bonding driver version 3.7.0 and support for
1467output slave selection was limited to round-robin and active-backup modes.
1468
14694 Querying Bonding Configuration
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470=================================
1471
Auke Kok6224e012006-06-08 11:15:35 -070014724.1 Bonding Configuration
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473-------------------------
1474
1475 Each bonding device has a read-only file residing in the
1476/proc/net/bonding directory. The file contents include information
1477about the bonding configuration, options and state of each slave.
1478
1479 For example, the contents of /proc/net/bonding/bond0 after the
1480driver is loaded with parameters of mode=0 and miimon=1000 is
1481generally as follows:
1482
1483 Ethernet Channel Bonding Driver: 2.6.1 (October 29, 2004)
1484 Bonding Mode: load balancing (round-robin)
1485 Currently Active Slave: eth0
1486 MII Status: up
1487 MII Polling Interval (ms): 1000
1488 Up Delay (ms): 0
1489 Down Delay (ms): 0
1490
1491 Slave Interface: eth1
1492 MII Status: up
1493 Link Failure Count: 1
1494
1495 Slave Interface: eth0
1496 MII Status: up
1497 Link Failure Count: 1
1498
1499 The precise format and contents will change depending upon the
1500bonding configuration, state, and version of the bonding driver.
1501
Auke Kok6224e012006-06-08 11:15:35 -070015024.2 Network configuration
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503-------------------------
1504
1505 The network configuration can be inspected using the ifconfig
1506command. Bonding devices will have the MASTER flag set; Bonding slave
1507devices will have the SLAVE flag set. The ifconfig output does not
1508contain information on which slaves are associated with which masters.
1509
1510 In the example below, the bond0 interface is the master
1511(MASTER) while eth0 and eth1 are slaves (SLAVE). Notice all slaves of
1512bond0 have the same MAC address (HWaddr) as bond0 for all modes except
1513TLB and ALB that require a unique MAC address for each slave.
1514
1515# /sbin/ifconfig
1516bond0 Link encap:Ethernet HWaddr 00:C0:F0:1F:37:B4
1517 inet addr:XXX.XXX.XXX.YYY Bcast:XXX.XXX.XXX.255 Mask:255.255.252.0
1518 UP BROADCAST RUNNING MASTER MULTICAST MTU:1500 Metric:1
1519 RX packets:7224794 errors:0 dropped:0 overruns:0 frame:0
1520 TX packets:3286647 errors:1 dropped:0 overruns:1 carrier:0
1521 collisions:0 txqueuelen:0
1522
1523eth0 Link encap:Ethernet HWaddr 00:C0:F0:1F:37:B4
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524 UP BROADCAST RUNNING SLAVE MULTICAST MTU:1500 Metric:1
1525 RX packets:3573025 errors:0 dropped:0 overruns:0 frame:0
1526 TX packets:1643167 errors:1 dropped:0 overruns:1 carrier:0
1527 collisions:0 txqueuelen:100
1528 Interrupt:10 Base address:0x1080
1529
1530eth1 Link encap:Ethernet HWaddr 00:C0:F0:1F:37:B4
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531 UP BROADCAST RUNNING SLAVE MULTICAST MTU:1500 Metric:1
1532 RX packets:3651769 errors:0 dropped:0 overruns:0 frame:0
1533 TX packets:1643480 errors:0 dropped:0 overruns:0 carrier:0
1534 collisions:0 txqueuelen:100
1535 Interrupt:9 Base address:0x1400
1536
Auke Kok6224e012006-06-08 11:15:35 -070015375. Switch Configuration
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538=======================
1539
1540 For this section, "switch" refers to whatever system the
1541bonded devices are directly connected to (i.e., where the other end of
1542the cable plugs into). This may be an actual dedicated switch device,
1543or it may be another regular system (e.g., another computer running
1544Linux),
1545
1546 The active-backup, balance-tlb and balance-alb modes do not
1547require any specific configuration of the switch.
1548
1549 The 802.3ad mode requires that the switch have the appropriate
1550ports configured as an 802.3ad aggregation. The precise method used
1551to configure this varies from switch to switch, but, for example, a
1552Cisco 3550 series switch requires that the appropriate ports first be
1553grouped together in a single etherchannel instance, then that
1554etherchannel is set to mode "lacp" to enable 802.3ad (instead of
1555standard EtherChannel).
1556
1557 The balance-rr, balance-xor and broadcast modes generally
1558require that the switch have the appropriate ports grouped together.
1559The nomenclature for such a group differs between switches, it may be
1560called an "etherchannel" (as in the Cisco example, above), a "trunk
1561group" or some other similar variation. For these modes, each switch
1562will also have its own configuration options for the switch's transmit
1563policy to the bond. Typical choices include XOR of either the MAC or
1564IP addresses. The transmit policy of the two peers does not need to
1565match. For these three modes, the bonding mode really selects a
1566transmit policy for an EtherChannel group; all three will interoperate
1567with another EtherChannel group.
1568
1569
Auke Kok6224e012006-06-08 11:15:35 -070015706. 802.1q VLAN Support
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571======================
1572
1573 It is possible to configure VLAN devices over a bond interface
1574using the 8021q driver. However, only packets coming from the 8021q
1575driver and passing through bonding will be tagged by default. Self
1576generated packets, for example, bonding's learning packets or ARP
1577packets generated by either ALB mode or the ARP monitor mechanism, are
1578tagged internally by bonding itself. As a result, bonding must
1579"learn" the VLAN IDs configured above it, and use those IDs to tag
1580self generated packets.
1581
1582 For reasons of simplicity, and to support the use of adapters
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001583that can do VLAN hardware acceleration offloading, the bonding
1584interface declares itself as fully hardware offloading capable, it gets
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585the add_vid/kill_vid notifications to gather the necessary
1586information, and it propagates those actions to the slaves. In case
1587of mixed adapter types, hardware accelerated tagged packets that
1588should go through an adapter that is not offloading capable are
1589"un-accelerated" by the bonding driver so the VLAN tag sits in the
1590regular location.
1591
1592 VLAN interfaces *must* be added on top of a bonding interface
1593only after enslaving at least one slave. The bonding interface has a
1594hardware address of 00:00:00:00:00:00 until the first slave is added.
1595If the VLAN interface is created prior to the first enslavement, it
1596would pick up the all-zeroes hardware address. Once the first slave
1597is attached to the bond, the bond device itself will pick up the
1598slave's hardware address, which is then available for the VLAN device.
1599
1600 Also, be aware that a similar problem can occur if all slaves
1601are released from a bond that still has one or more VLAN interfaces on
1602top of it. When a new slave is added, the bonding interface will
1603obtain its hardware address from the first slave, which might not
1604match the hardware address of the VLAN interfaces (which was
1605ultimately copied from an earlier slave).
1606
1607 There are two methods to insure that the VLAN device operates
1608with the correct hardware address if all slaves are removed from a
1609bond interface:
1610
1611 1. Remove all VLAN interfaces then recreate them
1612
1613 2. Set the bonding interface's hardware address so that it
1614matches the hardware address of the VLAN interfaces.
1615
1616 Note that changing a VLAN interface's HW address would set the
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001617underlying device -- i.e. the bonding interface -- to promiscuous
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618mode, which might not be what you want.
1619
1620
Auke Kok6224e012006-06-08 11:15:35 -070016217. Link Monitoring
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622==================
1623
1624 The bonding driver at present supports two schemes for
1625monitoring a slave device's link state: the ARP monitor and the MII
1626monitor.
1627
1628 At the present time, due to implementation restrictions in the
1629bonding driver itself, it is not possible to enable both ARP and MII
1630monitoring simultaneously.
1631
Auke Kok6224e012006-06-08 11:15:35 -070016327.1 ARP Monitor Operation
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633-------------------------
1634
1635 The ARP monitor operates as its name suggests: it sends ARP
1636queries to one or more designated peer systems on the network, and
1637uses the response as an indication that the link is operating. This
1638gives some assurance that traffic is actually flowing to and from one
1639or more peers on the local network.
1640
1641 The ARP monitor relies on the device driver itself to verify
1642that traffic is flowing. In particular, the driver must keep up to
1643date the last receive time, dev->last_rx, and transmit start time,
1644dev->trans_start. If these are not updated by the driver, then the
1645ARP monitor will immediately fail any slaves using that driver, and
1646those slaves will stay down. If networking monitoring (tcpdump, etc)
1647shows the ARP requests and replies on the network, then it may be that
1648your device driver is not updating last_rx and trans_start.
1649
Auke Kok6224e012006-06-08 11:15:35 -070016507.2 Configuring Multiple ARP Targets
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651------------------------------------
1652
1653 While ARP monitoring can be done with just one target, it can
1654be useful in a High Availability setup to have several targets to
1655monitor. In the case of just one target, the target itself may go
1656down or have a problem making it unresponsive to ARP requests. Having
1657an additional target (or several) increases the reliability of the ARP
1658monitoring.
1659
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001660 Multiple ARP targets must be separated by commas as follows:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661
1662# example options for ARP monitoring with three targets
1663alias bond0 bonding
1664options bond0 arp_interval=60 arp_ip_target=192.168.0.1,192.168.0.3,192.168.0.9
1665
1666 For just a single target the options would resemble:
1667
1668# example options for ARP monitoring with one target
1669alias bond0 bonding
1670options bond0 arp_interval=60 arp_ip_target=192.168.0.100
1671
1672
Auke Kok6224e012006-06-08 11:15:35 -070016737.3 MII Monitor Operation
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674-------------------------
1675
1676 The MII monitor monitors only the carrier state of the local
1677network interface. It accomplishes this in one of three ways: by
1678depending upon the device driver to maintain its carrier state, by
1679querying the device's MII registers, or by making an ethtool query to
1680the device.
1681
1682 If the use_carrier module parameter is 1 (the default value),
1683then the MII monitor will rely on the driver for carrier state
1684information (via the netif_carrier subsystem). As explained in the
1685use_carrier parameter information, above, if the MII monitor fails to
1686detect carrier loss on the device (e.g., when the cable is physically
1687disconnected), it may be that the driver does not support
1688netif_carrier.
1689
1690 If use_carrier is 0, then the MII monitor will first query the
1691device's (via ioctl) MII registers and check the link state. If that
1692request fails (not just that it returns carrier down), then the MII
1693monitor will make an ethtool ETHOOL_GLINK request to attempt to obtain
1694the same information. If both methods fail (i.e., the driver either
1695does not support or had some error in processing both the MII register
1696and ethtool requests), then the MII monitor will assume the link is
1697up.
1698
Auke Kok6224e012006-06-08 11:15:35 -070016998. Potential Sources of Trouble
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700===============================
1701
Auke Kok6224e012006-06-08 11:15:35 -070017028.1 Adventures in Routing
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703-------------------------
1704
1705 When bonding is configured, it is important that the slave
Auke Kok6224e012006-06-08 11:15:35 -07001706devices not have routes that supersede routes of the master (or,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707generally, not have routes at all). For example, suppose the bonding
1708device bond0 has two slaves, eth0 and eth1, and the routing table is
1709as follows:
1710
1711Kernel IP routing table
1712Destination Gateway Genmask Flags MSS Window irtt Iface
171310.0.0.0 0.0.0.0 255.255.0.0 U 40 0 0 eth0
171410.0.0.0 0.0.0.0 255.255.0.0 U 40 0 0 eth1
171510.0.0.0 0.0.0.0 255.255.0.0 U 40 0 0 bond0
1716127.0.0.0 0.0.0.0 255.0.0.0 U 40 0 0 lo
1717
1718 This routing configuration will likely still update the
1719receive/transmit times in the driver (needed by the ARP monitor), but
1720may bypass the bonding driver (because outgoing traffic to, in this
1721case, another host on network 10 would use eth0 or eth1 before bond0).
1722
1723 The ARP monitor (and ARP itself) may become confused by this
1724configuration, because ARP requests (generated by the ARP monitor)
1725will be sent on one interface (bond0), but the corresponding reply
1726will arrive on a different interface (eth0). This reply looks to ARP
1727as an unsolicited ARP reply (because ARP matches replies on an
1728interface basis), and is discarded. The MII monitor is not affected
1729by the state of the routing table.
1730
1731 The solution here is simply to insure that slaves do not have
1732routes of their own, and if for some reason they must, those routes do
Auke Kok6224e012006-06-08 11:15:35 -07001733not supersede routes of their master. This should generally be the
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734case, but unusual configurations or errant manual or automatic static
1735route additions may cause trouble.
1736
Auke Kok6224e012006-06-08 11:15:35 -070017378.2 Ethernet Device Renaming
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738----------------------------
1739
1740 On systems with network configuration scripts that do not
1741associate physical devices directly with network interface names (so
1742that the same physical device always has the same "ethX" name), it may
1743be necessary to add some special logic to either /etc/modules.conf or
1744/etc/modprobe.conf (depending upon which is installed on the system).
1745
1746 For example, given a modules.conf containing the following:
1747
1748alias bond0 bonding
1749options bond0 mode=some-mode miimon=50
1750alias eth0 tg3
1751alias eth1 tg3
1752alias eth2 e1000
1753alias eth3 e1000
1754
1755 If neither eth0 and eth1 are slaves to bond0, then when the
1756bond0 interface comes up, the devices may end up reordered. This
1757happens because bonding is loaded first, then its slave device's
1758drivers are loaded next. Since no other drivers have been loaded,
1759when the e1000 driver loads, it will receive eth0 and eth1 for its
1760devices, but the bonding configuration tries to enslave eth2 and eth3
1761(which may later be assigned to the tg3 devices).
1762
1763 Adding the following:
1764
1765add above bonding e1000 tg3
1766
1767 causes modprobe to load e1000 then tg3, in that order, when
1768bonding is loaded. This command is fully documented in the
1769modules.conf manual page.
1770
1771 On systems utilizing modprobe.conf (or modprobe.conf.local),
1772an equivalent problem can occur. In this case, the following can be
1773added to modprobe.conf (or modprobe.conf.local, as appropriate), as
1774follows (all on one line; it has been split here for clarity):
1775
1776install bonding /sbin/modprobe tg3; /sbin/modprobe e1000;
1777 /sbin/modprobe --ignore-install bonding
1778
1779 This will, when loading the bonding module, rather than
1780performing the normal action, instead execute the provided command.
1781This command loads the device drivers in the order needed, then calls
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001782modprobe with --ignore-install to cause the normal action to then take
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783place. Full documentation on this can be found in the modprobe.conf
1784and modprobe manual pages.
1785
Auke Kok6224e012006-06-08 11:15:35 -070017868.3. Painfully Slow Or No Failed Link Detection By Miimon
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787---------------------------------------------------------
1788
1789 By default, bonding enables the use_carrier option, which
1790instructs bonding to trust the driver to maintain carrier state.
1791
1792 As discussed in the options section, above, some drivers do
1793not support the netif_carrier_on/_off link state tracking system.
1794With use_carrier enabled, bonding will always see these links as up,
1795regardless of their actual state.
1796
1797 Additionally, other drivers do support netif_carrier, but do
1798not maintain it in real time, e.g., only polling the link state at
1799some fixed interval. In this case, miimon will detect failures, but
1800only after some long period of time has expired. If it appears that
1801miimon is very slow in detecting link failures, try specifying
1802use_carrier=0 to see if that improves the failure detection time. If
1803it does, then it may be that the driver checks the carrier state at a
1804fixed interval, but does not cache the MII register values (so the
1805use_carrier=0 method of querying the registers directly works). If
1806use_carrier=0 does not improve the failover, then the driver may cache
1807the registers, or the problem may be elsewhere.
1808
1809 Also, remember that miimon only checks for the device's
1810carrier state. It has no way to determine the state of devices on or
1811beyond other ports of a switch, or if a switch is refusing to pass
1812traffic while still maintaining carrier on.
1813
Auke Kok6224e012006-06-08 11:15:35 -070018149. SNMP agents
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815===============
1816
1817 If running SNMP agents, the bonding driver should be loaded
1818before any network drivers participating in a bond. This requirement
Tobias Klauserd533f672005-09-10 00:26:46 -07001819is due to the interface index (ipAdEntIfIndex) being associated to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820the first interface found with a given IP address. That is, there is
1821only one ipAdEntIfIndex for each IP address. For example, if eth0 and
1822eth1 are slaves of bond0 and the driver for eth0 is loaded before the
1823bonding driver, the interface for the IP address will be associated
1824with the eth0 interface. This configuration is shown below, the IP
1825address 192.168.1.1 has an interface index of 2 which indexes to eth0
1826in the ifDescr table (ifDescr.2).
1827
1828 interfaces.ifTable.ifEntry.ifDescr.1 = lo
1829 interfaces.ifTable.ifEntry.ifDescr.2 = eth0
1830 interfaces.ifTable.ifEntry.ifDescr.3 = eth1
1831 interfaces.ifTable.ifEntry.ifDescr.4 = eth2
1832 interfaces.ifTable.ifEntry.ifDescr.5 = eth3
1833 interfaces.ifTable.ifEntry.ifDescr.6 = bond0
1834 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.10.10.10 = 5
1835 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.192.168.1.1 = 2
1836 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.74.20.94 = 4
1837 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.127.0.0.1 = 1
1838
1839 This problem is avoided by loading the bonding driver before
1840any network drivers participating in a bond. Below is an example of
1841loading the bonding driver first, the IP address 192.168.1.1 is
1842correctly associated with ifDescr.2.
1843
1844 interfaces.ifTable.ifEntry.ifDescr.1 = lo
1845 interfaces.ifTable.ifEntry.ifDescr.2 = bond0
1846 interfaces.ifTable.ifEntry.ifDescr.3 = eth0
1847 interfaces.ifTable.ifEntry.ifDescr.4 = eth1
1848 interfaces.ifTable.ifEntry.ifDescr.5 = eth2
1849 interfaces.ifTable.ifEntry.ifDescr.6 = eth3
1850 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.10.10.10 = 6
1851 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.192.168.1.1 = 2
1852 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.74.20.94 = 5
1853 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.127.0.0.1 = 1
1854
1855 While some distributions may not report the interface name in
1856ifDescr, the association between the IP address and IfIndex remains
1857and SNMP functions such as Interface_Scan_Next will report that
1858association.
1859
Auke Kok6224e012006-06-08 11:15:35 -0700186010. Promiscuous mode
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861====================
1862
1863 When running network monitoring tools, e.g., tcpdump, it is
1864common to enable promiscuous mode on the device, so that all traffic
1865is seen (instead of seeing only traffic destined for the local host).
1866The bonding driver handles promiscuous mode changes to the bonding
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001867master device (e.g., bond0), and propagates the setting to the slave
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868devices.
1869
1870 For the balance-rr, balance-xor, broadcast, and 802.3ad modes,
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001871the promiscuous mode setting is propagated to all slaves.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872
1873 For the active-backup, balance-tlb and balance-alb modes, the
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001874promiscuous mode setting is propagated only to the active slave.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875
1876 For balance-tlb mode, the active slave is the slave currently
1877receiving inbound traffic.
1878
1879 For balance-alb mode, the active slave is the slave used as a
1880"primary." This slave is used for mode-specific control traffic, for
1881sending to peers that are unassigned or if the load is unbalanced.
1882
1883 For the active-backup, balance-tlb and balance-alb modes, when
1884the active slave changes (e.g., due to a link failure), the
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001885promiscuous setting will be propagated to the new active slave.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886
Auke Kok6224e012006-06-08 11:15:35 -0700188711. Configuring Bonding for High Availability
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001888=============================================
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889
1890 High Availability refers to configurations that provide
1891maximum network availability by having redundant or backup devices,
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001892links or switches between the host and the rest of the world. The
1893goal is to provide the maximum availability of network connectivity
1894(i.e., the network always works), even though other configurations
1895could provide higher throughput.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896
Auke Kok6224e012006-06-08 11:15:35 -0700189711.1 High Availability in a Single Switch Topology
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898--------------------------------------------------
1899
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001900 If two hosts (or a host and a single switch) are directly
1901connected via multiple physical links, then there is no availability
1902penalty to optimizing for maximum bandwidth. In this case, there is
1903only one switch (or peer), so if it fails, there is no alternative
1904access to fail over to. Additionally, the bonding load balance modes
1905support link monitoring of their members, so if individual links fail,
1906the load will be rebalanced across the remaining devices.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001908 See Section 13, "Configuring Bonding for Maximum Throughput"
1909for information on configuring bonding with one peer device.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910
Auke Kok6224e012006-06-08 11:15:35 -0700191111.2 High Availability in a Multiple Switch Topology
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912----------------------------------------------------
1913
1914 With multiple switches, the configuration of bonding and the
1915network changes dramatically. In multiple switch topologies, there is
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001916a trade off between network availability and usable bandwidth.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917
1918 Below is a sample network, configured to maximize the
1919availability of the network:
1920
1921 | |
1922 |port3 port3|
1923 +-----+----+ +-----+----+
1924 | |port2 ISL port2| |
1925 | switch A +--------------------------+ switch B |
1926 | | | |
1927 +-----+----+ +-----++---+
1928 |port1 port1|
1929 | +-------+ |
1930 +-------------+ host1 +---------------+
1931 eth0 +-------+ eth1
1932
1933 In this configuration, there is a link between the two
1934switches (ISL, or inter switch link), and multiple ports connecting to
1935the outside world ("port3" on each switch). There is no technical
1936reason that this could not be extended to a third switch.
1937
Auke Kok6224e012006-06-08 11:15:35 -0700193811.2.1 HA Bonding Mode Selection for Multiple Switch Topology
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001939-------------------------------------------------------------
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001941 In a topology such as the example above, the active-backup and
1942broadcast modes are the only useful bonding modes when optimizing for
1943availability; the other modes require all links to terminate on the
1944same peer for them to behave rationally.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945
1946active-backup: This is generally the preferred mode, particularly if
1947 the switches have an ISL and play together well. If the
1948 network configuration is such that one switch is specifically
1949 a backup switch (e.g., has lower capacity, higher cost, etc),
1950 then the primary option can be used to insure that the
1951 preferred link is always used when it is available.
1952
1953broadcast: This mode is really a special purpose mode, and is suitable
1954 only for very specific needs. For example, if the two
1955 switches are not connected (no ISL), and the networks beyond
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001956 them are totally independent. In this case, if it is
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957 necessary for some specific one-way traffic to reach both
1958 independent networks, then the broadcast mode may be suitable.
1959
Auke Kok6224e012006-06-08 11:15:35 -0700196011.2.2 HA Link Monitoring Selection for Multiple Switch Topology
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001961----------------------------------------------------------------
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962
1963 The choice of link monitoring ultimately depends upon your
1964switch. If the switch can reliably fail ports in response to other
1965failures, then either the MII or ARP monitors should work. For
1966example, in the above example, if the "port3" link fails at the remote
1967end, the MII monitor has no direct means to detect this. The ARP
1968monitor could be configured with a target at the remote end of port3,
1969thus detecting that failure without switch support.
1970
1971 In general, however, in a multiple switch topology, the ARP
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001972monitor can provide a higher level of reliability in detecting end to
1973end connectivity failures (which may be caused by the failure of any
1974individual component to pass traffic for any reason). Additionally,
1975the ARP monitor should be configured with multiple targets (at least
1976one for each switch in the network). This will insure that,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977regardless of which switch is active, the ARP monitor has a suitable
1978target to query.
1979
Jay Vosburgh9a6c6862007-11-13 20:25:48 -08001980 Note, also, that of late many switches now support a functionality
1981generally referred to as "trunk failover." This is a feature of the
1982switch that causes the link state of a particular switch port to be set
1983down (or up) when the state of another switch port goes down (or up).
Matt LaPlante19f59462009-04-27 15:06:31 +02001984Its purpose is to propagate link failures from logically "exterior" ports
Jay Vosburgh9a6c6862007-11-13 20:25:48 -08001985to the logically "interior" ports that bonding is able to monitor via
1986miimon. Availability and configuration for trunk failover varies by
1987switch, but this can be a viable alternative to the ARP monitor when using
1988suitable switches.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989
Auke Kok6224e012006-06-08 11:15:35 -0700199012. Configuring Bonding for Maximum Throughput
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001991==============================================
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992
Auke Kok6224e012006-06-08 11:15:35 -0700199312.1 Maximizing Throughput in a Single Switch Topology
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001994------------------------------------------------------
1995
1996 In a single switch configuration, the best method to maximize
1997throughput depends upon the application and network environment. The
1998various load balancing modes each have strengths and weaknesses in
1999different environments, as detailed below.
2000
2001 For this discussion, we will break down the topologies into
2002two categories. Depending upon the destination of most traffic, we
2003categorize them into either "gatewayed" or "local" configurations.
2004
2005 In a gatewayed configuration, the "switch" is acting primarily
2006as a router, and the majority of traffic passes through this router to
2007other networks. An example would be the following:
2008
2009
2010 +----------+ +----------+
2011 | |eth0 port1| | to other networks
2012 | Host A +---------------------+ router +------------------->
2013 | +---------------------+ | Hosts B and C are out
2014 | |eth1 port2| | here somewhere
2015 +----------+ +----------+
2016
2017 The router may be a dedicated router device, or another host
2018acting as a gateway. For our discussion, the important point is that
2019the majority of traffic from Host A will pass through the router to
2020some other network before reaching its final destination.
2021
2022 In a gatewayed network configuration, although Host A may
2023communicate with many other systems, all of its traffic will be sent
2024and received via one other peer on the local network, the router.
2025
2026 Note that the case of two systems connected directly via
2027multiple physical links is, for purposes of configuring bonding, the
2028same as a gatewayed configuration. In that case, it happens that all
2029traffic is destined for the "gateway" itself, not some other network
2030beyond the gateway.
2031
2032 In a local configuration, the "switch" is acting primarily as
2033a switch, and the majority of traffic passes through this switch to
2034reach other stations on the same network. An example would be the
2035following:
2036
2037 +----------+ +----------+ +--------+
2038 | |eth0 port1| +-------+ Host B |
2039 | Host A +------------+ switch |port3 +--------+
2040 | +------------+ | +--------+
2041 | |eth1 port2| +------------------+ Host C |
2042 +----------+ +----------+port4 +--------+
2043
2044
2045 Again, the switch may be a dedicated switch device, or another
2046host acting as a gateway. For our discussion, the important point is
2047that the majority of traffic from Host A is destined for other hosts
2048on the same local network (Hosts B and C in the above example).
2049
2050 In summary, in a gatewayed configuration, traffic to and from
2051the bonded device will be to the same MAC level peer on the network
2052(the gateway itself, i.e., the router), regardless of its final
2053destination. In a local configuration, traffic flows directly to and
2054from the final destinations, thus, each destination (Host B, Host C)
2055will be addressed directly by their individual MAC addresses.
2056
2057 This distinction between a gatewayed and a local network
2058configuration is important because many of the load balancing modes
2059available use the MAC addresses of the local network source and
2060destination to make load balancing decisions. The behavior of each
2061mode is described below.
2062
2063
Auke Kok6224e012006-06-08 11:15:35 -0700206412.1.1 MT Bonding Mode Selection for Single Switch Topology
Jay Vosburgh00354cf2005-07-21 12:18:02 -07002065-----------------------------------------------------------
2066
2067 This configuration is the easiest to set up and to understand,
2068although you will have to decide which bonding mode best suits your
2069needs. The trade offs for each mode are detailed below:
2070
2071balance-rr: This mode is the only mode that will permit a single
2072 TCP/IP connection to stripe traffic across multiple
2073 interfaces. It is therefore the only mode that will allow a
2074 single TCP/IP stream to utilize more than one interface's
2075 worth of throughput. This comes at a cost, however: the
Jay Vosburgh9a6c6862007-11-13 20:25:48 -08002076 striping generally results in peer systems receiving packets out
Jay Vosburgh00354cf2005-07-21 12:18:02 -07002077 of order, causing TCP/IP's congestion control system to kick
2078 in, often by retransmitting segments.
2079
2080 It is possible to adjust TCP/IP's congestion limits by
2081 altering the net.ipv4.tcp_reordering sysctl parameter. The
2082 usual default value is 3, and the maximum useful value is 127.
2083 For a four interface balance-rr bond, expect that a single
2084 TCP/IP stream will utilize no more than approximately 2.3
2085 interface's worth of throughput, even after adjusting
2086 tcp_reordering.
2087
Jay Vosburgh9a6c6862007-11-13 20:25:48 -08002088 Note that the fraction of packets that will be delivered out of
2089 order is highly variable, and is unlikely to be zero. The level
2090 of reordering depends upon a variety of factors, including the
2091 networking interfaces, the switch, and the topology of the
2092 configuration. Speaking in general terms, higher speed network
2093 cards produce more reordering (due to factors such as packet
2094 coalescing), and a "many to many" topology will reorder at a
2095 higher rate than a "many slow to one fast" configuration.
2096
2097 Many switches do not support any modes that stripe traffic
2098 (instead choosing a port based upon IP or MAC level addresses);
2099 for those devices, traffic for a particular connection flowing
2100 through the switch to a balance-rr bond will not utilize greater
2101 than one interface's worth of bandwidth.
Jay Vosburgh00354cf2005-07-21 12:18:02 -07002102
2103 If you are utilizing protocols other than TCP/IP, UDP for
2104 example, and your application can tolerate out of order
2105 delivery, then this mode can allow for single stream datagram
2106 performance that scales near linearly as interfaces are added
2107 to the bond.
2108
2109 This mode requires the switch to have the appropriate ports
2110 configured for "etherchannel" or "trunking."
2111
2112active-backup: There is not much advantage in this network topology to
2113 the active-backup mode, as the inactive backup devices are all
2114 connected to the same peer as the primary. In this case, a
2115 load balancing mode (with link monitoring) will provide the
2116 same level of network availability, but with increased
2117 available bandwidth. On the plus side, active-backup mode
2118 does not require any configuration of the switch, so it may
2119 have value if the hardware available does not support any of
2120 the load balance modes.
2121
2122balance-xor: This mode will limit traffic such that packets destined
2123 for specific peers will always be sent over the same
2124 interface. Since the destination is determined by the MAC
2125 addresses involved, this mode works best in a "local" network
2126 configuration (as described above), with destinations all on
2127 the same local network. This mode is likely to be suboptimal
2128 if all your traffic is passed through a single router (i.e., a
2129 "gatewayed" network configuration, as described above).
2130
2131 As with balance-rr, the switch ports need to be configured for
2132 "etherchannel" or "trunking."
2133
2134broadcast: Like active-backup, there is not much advantage to this
2135 mode in this type of network topology.
2136
2137802.3ad: This mode can be a good choice for this type of network
2138 topology. The 802.3ad mode is an IEEE standard, so all peers
2139 that implement 802.3ad should interoperate well. The 802.3ad
2140 protocol includes automatic configuration of the aggregates,
2141 so minimal manual configuration of the switch is needed
2142 (typically only to designate that some set of devices is
2143 available for 802.3ad). The 802.3ad standard also mandates
2144 that frames be delivered in order (within certain limits), so
2145 in general single connections will not see misordering of
2146 packets. The 802.3ad mode does have some drawbacks: the
2147 standard mandates that all devices in the aggregate operate at
2148 the same speed and duplex. Also, as with all bonding load
2149 balance modes other than balance-rr, no single connection will
2150 be able to utilize more than a single interface's worth of
2151 bandwidth.
2152
2153 Additionally, the linux bonding 802.3ad implementation
2154 distributes traffic by peer (using an XOR of MAC addresses),
2155 so in a "gatewayed" configuration, all outgoing traffic will
2156 generally use the same device. Incoming traffic may also end
2157 up on a single device, but that is dependent upon the
2158 balancing policy of the peer's 8023.ad implementation. In a
2159 "local" configuration, traffic will be distributed across the
2160 devices in the bond.
2161
2162 Finally, the 802.3ad mode mandates the use of the MII monitor,
2163 therefore, the ARP monitor is not available in this mode.
2164
2165balance-tlb: The balance-tlb mode balances outgoing traffic by peer.
2166 Since the balancing is done according to MAC address, in a
2167 "gatewayed" configuration (as described above), this mode will
2168 send all traffic across a single device. However, in a
2169 "local" network configuration, this mode balances multiple
2170 local network peers across devices in a vaguely intelligent
2171 manner (not a simple XOR as in balance-xor or 802.3ad mode),
2172 so that mathematically unlucky MAC addresses (i.e., ones that
2173 XOR to the same value) will not all "bunch up" on a single
2174 interface.
2175
2176 Unlike 802.3ad, interfaces may be of differing speeds, and no
2177 special switch configuration is required. On the down side,
2178 in this mode all incoming traffic arrives over a single
2179 interface, this mode requires certain ethtool support in the
2180 network device driver of the slave interfaces, and the ARP
2181 monitor is not available.
2182
2183balance-alb: This mode is everything that balance-tlb is, and more.
2184 It has all of the features (and restrictions) of balance-tlb,
2185 and will also balance incoming traffic from local network
2186 peers (as described in the Bonding Module Options section,
2187 above).
2188
2189 The only additional down side to this mode is that the network
2190 device driver must support changing the hardware address while
2191 the device is open.
2192
Auke Kok6224e012006-06-08 11:15:35 -0700219312.1.2 MT Link Monitoring for Single Switch Topology
Jay Vosburgh00354cf2005-07-21 12:18:02 -07002194----------------------------------------------------
2195
2196 The choice of link monitoring may largely depend upon which
2197mode you choose to use. The more advanced load balancing modes do not
2198support the use of the ARP monitor, and are thus restricted to using
2199the MII monitor (which does not provide as high a level of end to end
2200assurance as the ARP monitor).
2201
Auke Kok6224e012006-06-08 11:15:35 -0700220212.2 Maximum Throughput in a Multiple Switch Topology
Jay Vosburgh00354cf2005-07-21 12:18:02 -07002203-----------------------------------------------------
2204
2205 Multiple switches may be utilized to optimize for throughput
2206when they are configured in parallel as part of an isolated network
2207between two or more systems, for example:
2208
2209 +-----------+
2210 | Host A |
2211 +-+---+---+-+
2212 | | |
2213 +--------+ | +---------+
2214 | | |
2215 +------+---+ +-----+----+ +-----+----+
2216 | Switch A | | Switch B | | Switch C |
2217 +------+---+ +-----+----+ +-----+----+
2218 | | |
2219 +--------+ | +---------+
2220 | | |
2221 +-+---+---+-+
2222 | Host B |
2223 +-----------+
2224
2225 In this configuration, the switches are isolated from one
2226another. One reason to employ a topology such as this is for an
2227isolated network with many hosts (a cluster configured for high
2228performance, for example), using multiple smaller switches can be more
2229cost effective than a single larger switch, e.g., on a network with 24
2230hosts, three 24 port switches can be significantly less expensive than
2231a single 72 port switch.
2232
2233 If access beyond the network is required, an individual host
2234can be equipped with an additional network device connected to an
2235external network; this host then additionally acts as a gateway.
2236
Auke Kok6224e012006-06-08 11:15:35 -0700223712.2.1 MT Bonding Mode Selection for Multiple Switch Topology
Jay Vosburgh00354cf2005-07-21 12:18:02 -07002238-------------------------------------------------------------
2239
2240 In actual practice, the bonding mode typically employed in
2241configurations of this type is balance-rr. Historically, in this
2242network configuration, the usual caveats about out of order packet
2243delivery are mitigated by the use of network adapters that do not do
2244any kind of packet coalescing (via the use of NAPI, or because the
2245device itself does not generate interrupts until some number of
2246packets has arrived). When employed in this fashion, the balance-rr
2247mode allows individual connections between two hosts to effectively
2248utilize greater than one interface's bandwidth.
2249
Auke Kok6224e012006-06-08 11:15:35 -0700225012.2.2 MT Link Monitoring for Multiple Switch Topology
Jay Vosburgh00354cf2005-07-21 12:18:02 -07002251------------------------------------------------------
2252
2253 Again, in actual practice, the MII monitor is most often used
2254in this configuration, as performance is given preference over
2255availability. The ARP monitor will function in this topology, but its
2256advantages over the MII monitor are mitigated by the volume of probes
2257needed as the number of systems involved grows (remember that each
2258host in the network is configured with bonding).
2259
Auke Kok6224e012006-06-08 11:15:35 -0700226013. Switch Behavior Issues
Jay Vosburgh00354cf2005-07-21 12:18:02 -07002261==========================
2262
Auke Kok6224e012006-06-08 11:15:35 -0700226313.1 Link Establishment and Failover Delays
Jay Vosburgh00354cf2005-07-21 12:18:02 -07002264-------------------------------------------
2265
2266 Some switches exhibit undesirable behavior with regard to the
2267timing of link up and down reporting by the switch.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002268
2269 First, when a link comes up, some switches may indicate that
2270the link is up (carrier available), but not pass traffic over the
2271interface for some period of time. This delay is typically due to
2272some type of autonegotiation or routing protocol, but may also occur
2273during switch initialization (e.g., during recovery after a switch
2274failure). If you find this to be a problem, specify an appropriate
2275value to the updelay bonding module option to delay the use of the
2276relevant interface(s).
2277
2278 Second, some switches may "bounce" the link state one or more
2279times while a link is changing state. This occurs most commonly while
2280the switch is initializing. Again, an appropriate updelay value may
Jay Vosburgh00354cf2005-07-21 12:18:02 -07002281help.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002282
2283 Note that when a bonding interface has no active links, the
Jay Vosburgh00354cf2005-07-21 12:18:02 -07002284driver will immediately reuse the first link that goes up, even if the
2285updelay parameter has been specified (the updelay is ignored in this
2286case). If there are slave interfaces waiting for the updelay timeout
2287to expire, the interface that first went into that state will be
2288immediately reused. This reduces down time of the network if the
2289value of updelay has been overestimated, and since this occurs only in
2290cases with no connectivity, there is no additional penalty for
2291ignoring the updelay.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292
2293 In addition to the concerns about switch timings, if your
2294switches take a long time to go into backup mode, it may be desirable
2295to not activate a backup interface immediately after a link goes down.
2296Failover may be delayed via the downdelay bonding module option.
2297
Auke Kok6224e012006-06-08 11:15:35 -0700229813.2 Duplicated Incoming Packets
Jay Vosburgh00354cf2005-07-21 12:18:02 -07002299--------------------------------
2300
Jay Vosburgh9a6c6862007-11-13 20:25:48 -08002301 NOTE: Starting with version 3.0.2, the bonding driver has logic to
2302suppress duplicate packets, which should largely eliminate this problem.
2303The following description is kept for reference.
2304
Jay Vosburgh00354cf2005-07-21 12:18:02 -07002305 It is not uncommon to observe a short burst of duplicated
2306traffic when the bonding device is first used, or after it has been
2307idle for some period of time. This is most easily observed by issuing
2308a "ping" to some other host on the network, and noticing that the
2309output from ping flags duplicates (typically one per slave).
2310
2311 For example, on a bond in active-backup mode with five slaves
2312all connected to one switch, the output may appear as follows:
2313
2314# ping -n 10.0.4.2
2315PING 10.0.4.2 (10.0.4.2) from 10.0.3.10 : 56(84) bytes of data.
231664 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.7 ms
231764 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
231864 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
231964 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
232064 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
232164 bytes from 10.0.4.2: icmp_seq=2 ttl=64 time=0.216 ms
232264 bytes from 10.0.4.2: icmp_seq=3 ttl=64 time=0.267 ms
232364 bytes from 10.0.4.2: icmp_seq=4 ttl=64 time=0.222 ms
2324
2325 This is not due to an error in the bonding driver, rather, it
2326is a side effect of how many switches update their MAC forwarding
2327tables. Initially, the switch does not associate the MAC address in
2328the packet with a particular switch port, and so it may send the
2329traffic to all ports until its MAC forwarding table is updated. Since
2330the interfaces attached to the bond may occupy multiple ports on a
2331single switch, when the switch (temporarily) floods the traffic to all
2332ports, the bond device receives multiple copies of the same packet
2333(one per slave device).
2334
2335 The duplicated packet behavior is switch dependent, some
2336switches exhibit this, and some do not. On switches that display this
2337behavior, it can be induced by clearing the MAC forwarding table (on
2338most Cisco switches, the privileged command "clear mac address-table
2339dynamic" will accomplish this).
2340
Auke Kok6224e012006-06-08 11:15:35 -0700234114. Hardware Specific Considerations
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342====================================
2343
2344 This section contains additional information for configuring
2345bonding on specific hardware platforms, or for interfacing bonding
2346with particular switches or other devices.
2347
Auke Kok6224e012006-06-08 11:15:35 -0700234814.1 IBM BladeCenter
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349--------------------
2350
2351 This applies to the JS20 and similar systems.
2352
2353 On the JS20 blades, the bonding driver supports only
2354balance-rr, active-backup, balance-tlb and balance-alb modes. This is
2355largely due to the network topology inside the BladeCenter, detailed
2356below.
2357
2358JS20 network adapter information
2359--------------------------------
2360
2361 All JS20s come with two Broadcom Gigabit Ethernet ports
Jay Vosburgh00354cf2005-07-21 12:18:02 -07002362integrated on the planar (that's "motherboard" in IBM-speak). In the
2363BladeCenter chassis, the eth0 port of all JS20 blades is hard wired to
2364I/O Module #1; similarly, all eth1 ports are wired to I/O Module #2.
2365An add-on Broadcom daughter card can be installed on a JS20 to provide
2366two more Gigabit Ethernet ports. These ports, eth2 and eth3, are
2367wired to I/O Modules 3 and 4, respectively.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002368
2369 Each I/O Module may contain either a switch or a passthrough
2370module (which allows ports to be directly connected to an external
2371switch). Some bonding modes require a specific BladeCenter internal
2372network topology in order to function; these are detailed below.
2373
2374 Additional BladeCenter-specific networking information can be
2375found in two IBM Redbooks (www.ibm.com/redbooks):
2376
2377"IBM eServer BladeCenter Networking Options"
2378"IBM eServer BladeCenter Layer 2-7 Network Switching"
2379
2380BladeCenter networking configuration
2381------------------------------------
2382
2383 Because a BladeCenter can be configured in a very large number
2384of ways, this discussion will be confined to describing basic
2385configurations.
2386
Jay Vosburgh00354cf2005-07-21 12:18:02 -07002387 Normally, Ethernet Switch Modules (ESMs) are used in I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388modules 1 and 2. In this configuration, the eth0 and eth1 ports of a
2389JS20 will be connected to different internal switches (in the
2390respective I/O modules).
2391
Jay Vosburgh00354cf2005-07-21 12:18:02 -07002392 A passthrough module (OPM or CPM, optical or copper,
2393passthrough module) connects the I/O module directly to an external
2394switch. By using PMs in I/O module #1 and #2, the eth0 and eth1
2395interfaces of a JS20 can be redirected to the outside world and
2396connected to a common external switch.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002397
Jay Vosburgh00354cf2005-07-21 12:18:02 -07002398 Depending upon the mix of ESMs and PMs, the network will
2399appear to bonding as either a single switch topology (all PMs) or as a
2400multiple switch topology (one or more ESMs, zero or more PMs). It is
2401also possible to connect ESMs together, resulting in a configuration
2402much like the example in "High Availability in a Multiple Switch
2403Topology," above.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404
Jay Vosburgh00354cf2005-07-21 12:18:02 -07002405Requirements for specific modes
2406-------------------------------
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407
Jay Vosburgh00354cf2005-07-21 12:18:02 -07002408 The balance-rr mode requires the use of passthrough modules
2409for devices in the bond, all connected to an common external switch.
2410That switch must be configured for "etherchannel" or "trunking" on the
Linus Torvalds1da177e2005-04-16 15:20:36 -07002411appropriate ports, as is usual for balance-rr.
2412
2413 The balance-alb and balance-tlb modes will function with
2414either switch modules or passthrough modules (or a mix). The only
2415specific requirement for these modes is that all network interfaces
2416must be able to reach all destinations for traffic sent over the
2417bonding device (i.e., the network must converge at some point outside
2418the BladeCenter).
2419
2420 The active-backup mode has no additional requirements.
2421
2422Link monitoring issues
2423----------------------
2424
2425 When an Ethernet Switch Module is in place, only the ARP
2426monitor will reliably detect link loss to an external switch. This is
2427nothing unusual, but examination of the BladeCenter cabinet would
2428suggest that the "external" network ports are the ethernet ports for
2429the system, when it fact there is a switch between these "external"
2430ports and the devices on the JS20 system itself. The MII monitor is
2431only able to detect link failures between the ESM and the JS20 system.
2432
2433 When a passthrough module is in place, the MII monitor does
2434detect failures to the "external" port, which is then directly
2435connected to the JS20 system.
2436
2437Other concerns
2438--------------
2439
Jay Vosburgh00354cf2005-07-21 12:18:02 -07002440 The Serial Over LAN (SoL) link is established over the primary
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441ethernet (eth0) only, therefore, any loss of link to eth0 will result
2442in losing your SoL connection. It will not fail over with other
Jay Vosburgh00354cf2005-07-21 12:18:02 -07002443network traffic, as the SoL system is beyond the control of the
2444bonding driver.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002445
2446 It may be desirable to disable spanning tree on the switch
2447(either the internal Ethernet Switch Module, or an external switch) to
Jay Vosburgh00354cf2005-07-21 12:18:02 -07002448avoid fail-over delay issues when using bonding.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449
2450
Auke Kok6224e012006-06-08 11:15:35 -0700245115. Frequently Asked Questions
Linus Torvalds1da177e2005-04-16 15:20:36 -07002452==============================
2453
24541. Is it SMP safe?
2455
2456 Yes. The old 2.0.xx channel bonding patch was not SMP safe.
2457The new driver was designed to be SMP safe from the start.
2458
24592. What type of cards will work with it?
2460
2461 Any Ethernet type cards (you can even mix cards - a Intel
Jay Vosburgh00354cf2005-07-21 12:18:02 -07002462EtherExpress PRO/100 and a 3com 3c905b, for example). For most modes,
2463devices need not be of the same speed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002464
Jay Vosburgh9a6c6862007-11-13 20:25:48 -08002465 Starting with version 3.2.1, bonding also supports Infiniband
2466slaves in active-backup mode.
2467
Linus Torvalds1da177e2005-04-16 15:20:36 -070024683. How many bonding devices can I have?
2469
2470 There is no limit.
2471
24724. How many slaves can a bonding device have?
2473
2474 This is limited only by the number of network interfaces Linux
2475supports and/or the number of network cards you can place in your
2476system.
2477
24785. What happens when a slave link dies?
2479
2480 If link monitoring is enabled, then the failing device will be
2481disabled. The active-backup mode will fail over to a backup link, and
2482other modes will ignore the failed link. The link will continue to be
2483monitored, and should it recover, it will rejoin the bond (in whatever
Jay Vosburgh00354cf2005-07-21 12:18:02 -07002484manner is appropriate for the mode). See the sections on High
2485Availability and the documentation for each mode for additional
2486information.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002487
2488 Link monitoring can be enabled via either the miimon or
Jay Vosburgh00354cf2005-07-21 12:18:02 -07002489arp_interval parameters (described in the module parameters section,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002490above). In general, miimon monitors the carrier state as sensed by
2491the underlying network device, and the arp monitor (arp_interval)
2492monitors connectivity to another host on the local network.
2493
2494 If no link monitoring is configured, the bonding driver will
2495be unable to detect link failures, and will assume that all links are
2496always available. This will likely result in lost packets, and a
Jay Vosburgh00354cf2005-07-21 12:18:02 -07002497resulting degradation of performance. The precise performance loss
Linus Torvalds1da177e2005-04-16 15:20:36 -07002498depends upon the bonding mode and network configuration.
2499
25006. Can bonding be used for High Availability?
2501
2502 Yes. See the section on High Availability for details.
2503
25047. Which switches/systems does it work with?
2505
2506 The full answer to this depends upon the desired mode.
2507
2508 In the basic balance modes (balance-rr and balance-xor), it
2509works with any system that supports etherchannel (also called
2510trunking). Most managed switches currently available have such
Jay Vosburgh00354cf2005-07-21 12:18:02 -07002511support, and many unmanaged switches as well.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002512
2513 The advanced balance modes (balance-tlb and balance-alb) do
2514not have special switch requirements, but do need device drivers that
2515support specific features (described in the appropriate section under
Jay Vosburgh00354cf2005-07-21 12:18:02 -07002516module parameters, above).
Linus Torvalds1da177e2005-04-16 15:20:36 -07002517
Auke Kok6224e012006-06-08 11:15:35 -07002518 In 802.3ad mode, it works with systems that support IEEE
Linus Torvalds1da177e2005-04-16 15:20:36 -07002519802.3ad Dynamic Link Aggregation. Most managed and many unmanaged
2520switches currently available support 802.3ad.
2521
2522 The active-backup mode should work with any Layer-II switch.
2523
25248. Where does a bonding device get its MAC address from?
2525
Jay Vosburgh9a6c6862007-11-13 20:25:48 -08002526 When using slave devices that have fixed MAC addresses, or when
2527the fail_over_mac option is enabled, the bonding device's MAC address is
2528the MAC address of the active slave.
2529
2530 For other configurations, if not explicitly configured (with
2531ifconfig or ip link), the MAC address of the bonding device is taken from
2532its first slave device. This MAC address is then passed to all following
2533slaves and remains persistent (even if the first slave is removed) until
2534the bonding device is brought down or reconfigured.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002535
2536 If you wish to change the MAC address, you can set it with
Jay Vosburgh00354cf2005-07-21 12:18:02 -07002537ifconfig or ip link:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538
2539# ifconfig bond0 hw ether 00:11:22:33:44:55
2540
Jay Vosburgh00354cf2005-07-21 12:18:02 -07002541# ip link set bond0 address 66:77:88:99:aa:bb
2542
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543 The MAC address can be also changed by bringing down/up the
2544device and then changing its slaves (or their order):
2545
2546# ifconfig bond0 down ; modprobe -r bonding
2547# ifconfig bond0 .... up
2548# ifenslave bond0 eth...
2549
2550 This method will automatically take the address from the next
2551slave that is added.
2552
2553 To restore your slaves' MAC addresses, you need to detach them
2554from the bond (`ifenslave -d bond0 eth0'). The bonding driver will
2555then restore the MAC addresses that the slaves had before they were
2556enslaved.
2557
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700255816. Resources and Links
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559=======================
2560
Nicolas de Pesloüana23c37f12011-03-13 10:34:22 +00002561 The latest version of the bonding driver can be found in the latest
Linus Torvalds1da177e2005-04-16 15:20:36 -07002562version of the linux kernel, found on http://kernel.org
2563
Nicolas de Pesloüana23c37f12011-03-13 10:34:22 +00002564 The latest version of this document can be found in the latest kernel
2565source (named Documentation/networking/bonding.txt).
Jay Vosburgh00354cf2005-07-21 12:18:02 -07002566
Nicolas de Pesloüana23c37f12011-03-13 10:34:22 +00002567 Discussions regarding the usage of the bonding driver take place on the
2568bonding-devel mailing list, hosted at sourceforge.net. If you have questions or
2569problems, post them to the list. The list address is:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570
2571bonding-devel@lists.sourceforge.net
2572
Jay Vosburgh00354cf2005-07-21 12:18:02 -07002573 The administrative interface (to subscribe or unsubscribe) can
2574be found at:
2575
Linus Torvalds1da177e2005-04-16 15:20:36 -07002576https://lists.sourceforge.net/lists/listinfo/bonding-devel
2577
Nicolas de Pesloüana23c37f12011-03-13 10:34:22 +00002578 Discussions regarding the developpement of the bonding driver take place
2579on the main Linux network mailing list, hosted at vger.kernel.org. The list
2580address is:
2581
2582netdev@vger.kernel.org
2583
2584 The administrative interface (to subscribe or unsubscribe) can
2585be found at:
2586
2587http://vger.kernel.org/vger-lists.html#netdev
2588
Linus Torvalds1da177e2005-04-16 15:20:36 -07002589Donald Becker's Ethernet Drivers and diag programs may be found at :
Justin P. Mattock0ea6e612010-07-23 20:51:24 -07002590 - http://web.archive.org/web/*/http://www.scyld.com/network/
Linus Torvalds1da177e2005-04-16 15:20:36 -07002591
2592You will also find a lot of information regarding Ethernet, NWay, MII,
2593etc. at www.scyld.com.
2594
2595-- END --