blob: 24d029455baadabc3acc398e3970ff8052e3ab1d [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
4 Latest update: 21 June 2005
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
15
Jay Vosburgh00354cf2005-07-21 12:18:02 -070016Introduction
17============
Linus Torvalds1da177e2005-04-16 15:20:36 -070018
Jay Vosburgh00354cf2005-07-21 12:18:02 -070019 The Linux bonding driver provides a method for aggregating
20multiple network interfaces into a single logical "bonded" interface.
21The behavior of the bonded interfaces depends upon the mode; generally
22speaking, modes provide either hot standby or load balancing services.
23Additionally, link integrity monitoring may be performed.
24
25 The bonding driver originally came from Donald Becker's
26beowulf patches for kernel 2.0. It has changed quite a bit since, and
27the original tools from extreme-linux and beowulf sites will not work
28with this version of the driver.
29
30 For new versions of the driver, updated userspace tools, and
31who to ask for help, please follow the links at the end of this file.
Linus Torvalds1da177e2005-04-16 15:20:36 -070032
33Table of Contents
34=================
35
361. Bonding Driver Installation
37
382. Bonding Driver Options
39
403. Configuring Bonding Devices
413.1 Configuration with sysconfig support
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700423.1.1 Using DHCP with sysconfig
433.1.2 Configuring Multiple Bonds with sysconfig
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443.2 Configuration with initscripts support
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700453.2.1 Using DHCP with initscripts
463.2.2 Configuring Multiple Bonds with initscripts
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473.3 Configuring Bonding Manually
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700483.3.1 Configuring Multiple Bonds Manually
Linus Torvalds1da177e2005-04-16 15:20:36 -070049
505. Querying Bonding Configuration
515.1 Bonding Configuration
525.2 Network Configuration
53
546. Switch Configuration
55
567. 802.1q VLAN Support
57
588. Link Monitoring
598.1 ARP Monitor Operation
608.2 Configuring Multiple ARP Targets
618.3 MII Monitor Operation
62
639. Potential Trouble Sources
649.1 Adventures in Routing
659.2 Ethernet Device Renaming
669.3 Painfully Slow Or No Failed Link Detection By Miimon
67
6810. SNMP agents
69
7011. Promiscuous mode
71
Jay Vosburgh00354cf2005-07-21 12:18:02 -07007212. Configuring Bonding for High Availability
Linus Torvalds1da177e2005-04-16 15:20:36 -07007312.1 High Availability in a Single Switch Topology
Linus Torvalds1da177e2005-04-16 15:20:36 -07007412.2 High Availability in a Multiple Switch Topology
Jay Vosburgh00354cf2005-07-21 12:18:02 -07007512.2.1 HA Bonding Mode Selection for Multiple Switch Topology
7612.2.2 HA Link Monitoring for Multiple Switch Topology
Linus Torvalds1da177e2005-04-16 15:20:36 -070077
Jay Vosburgh00354cf2005-07-21 12:18:02 -07007813. Configuring Bonding for Maximum Throughput
7913.1 Maximum Throughput in a Single Switch Topology
8013.1.1 MT Bonding Mode Selection for Single Switch Topology
8113.1.2 MT Link Monitoring for Single Switch Topology
8213.2 Maximum Throughput in a Multiple Switch Topology
8313.2.1 MT Bonding Mode Selection for Multiple Switch Topology
8413.2.2 MT Link Monitoring for Multiple Switch Topology
Linus Torvalds1da177e2005-04-16 15:20:36 -070085
Jay Vosburgh00354cf2005-07-21 12:18:02 -07008614. Switch Behavior Issues
8714.1 Link Establishment and Failover Delays
8814.2 Duplicated Incoming Packets
Linus Torvalds1da177e2005-04-16 15:20:36 -070089
Jay Vosburgh00354cf2005-07-21 12:18:02 -07009015. Hardware Specific Considerations
9115.1 IBM BladeCenter
92
9316. Frequently Asked Questions
94
9517. Resources and Links
Linus Torvalds1da177e2005-04-16 15:20:36 -070096
97
981. Bonding Driver Installation
99==============================
100
101 Most popular distro kernels ship with the bonding driver
102already available as a module and the ifenslave user level control
103program installed and ready for use. If your distro does not, or you
104have need to compile bonding from source (e.g., configuring and
105installing a mainline kernel from kernel.org), you'll need to perform
106the following steps:
107
1081.1 Configure and build the kernel with bonding
109-----------------------------------------------
110
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700111 The current version of the bonding driver is available in the
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112drivers/net/bonding subdirectory of the most recent kernel source
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700113(which is available on http://kernel.org). Most users "rolling their
114own" will want to use the most recent kernel from kernel.org.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115
116 Configure kernel with "make menuconfig" (or "make xconfig" or
117"make config"), then select "Bonding driver support" in the "Network
118device support" section. It is recommended that you configure the
119driver as module since it is currently the only way to pass parameters
120to the driver or configure more than one bonding device.
121
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700122 Build and install the new kernel and modules, then continue
123below to install ifenslave.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124
1251.2 Install ifenslave Control Utility
126-------------------------------------
127
128 The ifenslave user level control program is included in the
129kernel source tree, in the file Documentation/networking/ifenslave.c.
130It is generally recommended that you use the ifenslave that
131corresponds to the kernel that you are using (either from the same
132source tree or supplied with the distro), however, ifenslave
133executables from older kernels should function (but features newer
134than the ifenslave release are not supported). Running an ifenslave
135that is newer than the kernel is not supported, and may or may not
136work.
137
138 To install ifenslave, do the following:
139
140# gcc -Wall -O -I/usr/src/linux/include ifenslave.c -o ifenslave
141# cp ifenslave /sbin/ifenslave
142
143 If your kernel source is not in "/usr/src/linux," then replace
144"/usr/src/linux/include" in the above with the location of your kernel
145source include directory.
146
147 You may wish to back up any existing /sbin/ifenslave, or, for
148testing or informal use, tag the ifenslave to the kernel version
149(e.g., name the ifenslave executable /sbin/ifenslave-2.6.10).
150
151IMPORTANT NOTE:
152
153 If you omit the "-I" or specify an incorrect directory, you
154may end up with an ifenslave that is incompatible with the kernel
155you're trying to build it for. Some distros (e.g., Red Hat from 7.1
156onwards) do not have /usr/include/linux symbolically linked to the
157default kernel source include directory.
158
159
1602. Bonding Driver Options
161=========================
162
163 Options for the bonding driver are supplied as parameters to
164the bonding module at load time. They may be given as command line
165arguments to the insmod or modprobe command, but are usually specified
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700166in either the /etc/modules.conf or /etc/modprobe.conf configuration
167file, or in a distro-specific configuration file (some of which are
168detailed in the next section).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169
170 The available bonding driver parameters are listed below. If a
171parameter is not specified the default value is used. When initially
172configuring a bond, it is recommended "tail -f /var/log/messages" be
173run in a separate window to watch for bonding driver error messages.
174
175 It is critical that either the miimon or arp_interval and
176arp_ip_target parameters be specified, otherwise serious network
177degradation will occur during link failures. Very few devices do not
178support at least miimon, so there is really no reason not to use it.
179
180 Options with textual values will accept either the text name
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700181or, for backwards compatibility, the option value. E.g.,
182"mode=802.3ad" and "mode=4" set the same mode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183
184 The parameters are as follows:
185
186arp_interval
187
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700188 Specifies the ARP link monitoring frequency in milliseconds.
189 If ARP monitoring is used in an etherchannel compatible mode
190 (modes 0 and 2), the switch should be configured in a mode
191 that evenly distributes packets across all links. If the
192 switch is configured to distribute the packets in an XOR
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 fashion, all replies from the ARP targets will be received on
194 the same link which could cause the other team members to
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700195 fail. ARP monitoring should not be used in conjunction with
196 miimon. A value of 0 disables ARP monitoring. The default
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 value is 0.
198
199arp_ip_target
200
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700201 Specifies the IP addresses to use as ARP monitoring peers when
202 arp_interval is > 0. These are the targets of the ARP request
203 sent to determine the health of the link to the targets.
204 Specify these values in ddd.ddd.ddd.ddd format. Multiple IP
205 addresses must be separated by a comma. At least one IP
206 address must be given for ARP monitoring to function. The
207 maximum number of targets that can be specified is 16. The
208 default value is no IP addresses.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209
210downdelay
211
212 Specifies the time, in milliseconds, to wait before disabling
213 a slave after a link failure has been detected. This option
214 is only valid for the miimon link monitor. The downdelay
215 value should be a multiple of the miimon value; if not, it
216 will be rounded down to the nearest multiple. The default
217 value is 0.
218
219lacp_rate
220
221 Option specifying the rate in which we'll ask our link partner
222 to transmit LACPDU packets in 802.3ad mode. Possible values
223 are:
224
225 slow or 0
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700226 Request partner to transmit LACPDUs every 30 seconds
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227
228 fast or 1
229 Request partner to transmit LACPDUs every 1 second
230
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700231 The default is slow.
232
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233max_bonds
234
235 Specifies the number of bonding devices to create for this
236 instance of the bonding driver. E.g., if max_bonds is 3, and
237 the bonding driver is not already loaded, then bond0, bond1
238 and bond2 will be created. The default value is 1.
239
240miimon
241
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700242 Specifies the MII link monitoring frequency in milliseconds.
243 This determines how often the link state of each slave is
244 inspected for link failures. A value of zero disables MII
245 link monitoring. A value of 100 is a good starting point.
246 The use_carrier option, below, affects how the link state is
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 determined. See the High Availability section for additional
248 information. The default value is 0.
249
250mode
251
252 Specifies one of the bonding policies. The default is
253 balance-rr (round robin). Possible values are:
254
255 balance-rr or 0
256
257 Round-robin policy: Transmit packets in sequential
258 order from the first available slave through the
259 last. This mode provides load balancing and fault
260 tolerance.
261
262 active-backup or 1
263
264 Active-backup policy: Only one slave in the bond is
265 active. A different slave becomes active if, and only
266 if, the active slave fails. The bond's MAC address is
267 externally visible on only one port (network adapter)
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700268 to avoid confusing the switch.
269
270 In bonding version 2.6.2 or later, when a failover
271 occurs in active-backup mode, bonding will issue one
272 or more gratuitous ARPs on the newly active slave.
273 One gratutious ARP is issued for the bonding master
274 interface and each VLAN interfaces configured above
275 it, provided that the interface has at least one IP
276 address configured. Gratuitous ARPs issued for VLAN
277 interfaces are tagged with the appropriate VLAN id.
278
279 This mode provides fault tolerance. The primary
280 option, documented below, affects the behavior of this
281 mode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282
283 balance-xor or 2
284
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700285 XOR policy: Transmit based on the selected transmit
286 hash policy. The default policy is a simple [(source
287 MAC address XOR'd with destination MAC address) modulo
288 slave count]. Alternate transmit policies may be
289 selected via the xmit_hash_policy option, described
290 below.
291
292 This mode provides load balancing and fault tolerance.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293
294 broadcast or 3
295
296 Broadcast policy: transmits everything on all slave
297 interfaces. This mode provides fault tolerance.
298
299 802.3ad or 4
300
301 IEEE 802.3ad Dynamic link aggregation. Creates
302 aggregation groups that share the same speed and
303 duplex settings. Utilizes all slaves in the active
304 aggregator according to the 802.3ad specification.
305
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700306 Slave selection for outgoing traffic is done according
307 to the transmit hash policy, which may be changed from
308 the default simple XOR policy via the xmit_hash_policy
309 option, documented below. Note that not all transmit
310 policies may be 802.3ad compliant, particularly in
311 regards to the packet mis-ordering requirements of
312 section 43.2.4 of the 802.3ad standard. Differing
313 peer implementations will have varying tolerances for
314 noncompliance.
315
316 Prerequisites:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317
318 1. Ethtool support in the base drivers for retrieving
319 the speed and duplex of each slave.
320
321 2. A switch that supports IEEE 802.3ad Dynamic link
322 aggregation.
323
324 Most switches will require some type of configuration
325 to enable 802.3ad mode.
326
327 balance-tlb or 5
328
329 Adaptive transmit load balancing: channel bonding that
330 does not require any special switch support. The
331 outgoing traffic is distributed according to the
332 current load (computed relative to the speed) on each
333 slave. Incoming traffic is received by the current
334 slave. If the receiving slave fails, another slave
335 takes over the MAC address of the failed receiving
336 slave.
337
338 Prerequisite:
339
340 Ethtool support in the base drivers for retrieving the
341 speed of each slave.
342
343 balance-alb or 6
344
345 Adaptive load balancing: includes balance-tlb plus
346 receive load balancing (rlb) for IPV4 traffic, and
347 does not require any special switch support. The
348 receive load balancing is achieved by ARP negotiation.
349 The bonding driver intercepts the ARP Replies sent by
350 the local system on their way out and overwrites the
351 source hardware address with the unique hardware
352 address of one of the slaves in the bond such that
353 different peers use different hardware addresses for
354 the server.
355
356 Receive traffic from connections created by the server
357 is also balanced. When the local system sends an ARP
358 Request the bonding driver copies and saves the peer's
359 IP information from the ARP packet. When the ARP
360 Reply arrives from the peer, its hardware address is
361 retrieved and the bonding driver initiates an ARP
362 reply to this peer assigning it to one of the slaves
363 in the bond. A problematic outcome of using ARP
364 negotiation for balancing is that each time that an
365 ARP request is broadcast it uses the hardware address
366 of the bond. Hence, peers learn the hardware address
367 of the bond and the balancing of receive traffic
368 collapses to the current slave. This is handled by
369 sending updates (ARP Replies) to all the peers with
370 their individually assigned hardware address such that
371 the traffic is redistributed. Receive traffic is also
372 redistributed when a new slave is added to the bond
373 and when an inactive slave is re-activated. The
374 receive load is distributed sequentially (round robin)
375 among the group of highest speed slaves in the bond.
376
377 When a link is reconnected or a new slave joins the
378 bond the receive traffic is redistributed among all
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700379 active slaves in the bond by initiating ARP Replies
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380 with the selected mac address to each of the
381 clients. The updelay parameter (detailed below) must
382 be set to a value equal or greater than the switch's
383 forwarding delay so that the ARP Replies sent to the
384 peers will not be blocked by the switch.
385
386 Prerequisites:
387
388 1. Ethtool support in the base drivers for retrieving
389 the speed of each slave.
390
391 2. Base driver support for setting the hardware
392 address of a device while it is open. This is
393 required so that there will always be one slave in the
394 team using the bond hardware address (the
395 curr_active_slave) while having a unique hardware
396 address for each slave in the bond. If the
397 curr_active_slave fails its hardware address is
398 swapped with the new curr_active_slave that was
399 chosen.
400
401primary
402
403 A string (eth0, eth2, etc) specifying which slave is the
404 primary device. The specified device will always be the
405 active slave while it is available. Only when the primary is
406 off-line will alternate devices be used. This is useful when
407 one slave is preferred over another, e.g., when one slave has
408 higher throughput than another.
409
410 The primary option is only valid for active-backup mode.
411
412updelay
413
414 Specifies the time, in milliseconds, to wait before enabling a
415 slave after a link recovery has been detected. This option is
416 only valid for the miimon link monitor. The updelay value
417 should be a multiple of the miimon value; if not, it will be
418 rounded down to the nearest multiple. The default value is 0.
419
420use_carrier
421
422 Specifies whether or not miimon should use MII or ETHTOOL
423 ioctls vs. netif_carrier_ok() to determine the link
424 status. The MII or ETHTOOL ioctls are less efficient and
425 utilize a deprecated calling sequence within the kernel. The
426 netif_carrier_ok() relies on the device driver to maintain its
427 state with netif_carrier_on/off; at this writing, most, but
428 not all, device drivers support this facility.
429
430 If bonding insists that the link is up when it should not be,
431 it may be that your network device driver does not support
432 netif_carrier_on/off. The default state for netif_carrier is
433 "carrier on," so if a driver does not support netif_carrier,
434 it will appear as if the link is always up. In this case,
435 setting use_carrier to 0 will cause bonding to revert to the
436 MII / ETHTOOL ioctl method to determine the link state.
437
438 A value of 1 enables the use of netif_carrier_ok(), a value of
439 0 will use the deprecated MII / ETHTOOL ioctls. The default
440 value is 1.
441
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700442xmit_hash_policy
443
444 Selects the transmit hash policy to use for slave selection in
445 balance-xor and 802.3ad modes. Possible values are:
446
447 layer2
448
449 Uses XOR of hardware MAC addresses to generate the
450 hash. The formula is
451
452 (source MAC XOR destination MAC) modulo slave count
453
454 This algorithm will place all traffic to a particular
455 network peer on the same slave.
456
457 This algorithm is 802.3ad compliant.
458
459 layer3+4
460
461 This policy uses upper layer protocol information,
462 when available, to generate the hash. This allows for
463 traffic to a particular network peer to span multiple
464 slaves, although a single connection will not span
465 multiple slaves.
466
467 The formula for unfragmented TCP and UDP packets is
468
469 ((source port XOR dest port) XOR
470 ((source IP XOR dest IP) AND 0xffff)
471 modulo slave count
472
473 For fragmented TCP or UDP packets and all other IP
474 protocol traffic, the source and destination port
475 information is omitted. For non-IP traffic, the
476 formula is the same as for the layer2 transmit hash
477 policy.
478
479 This policy is intended to mimic the behavior of
480 certain switches, notably Cisco switches with PFC2 as
481 well as some Foundry and IBM products.
482
483 This algorithm is not fully 802.3ad compliant. A
484 single TCP or UDP conversation containing both
485 fragmented and unfragmented packets will see packets
486 striped across two interfaces. This may result in out
487 of order delivery. Most traffic types will not meet
488 this criteria, as TCP rarely fragments traffic, and
489 most UDP traffic is not involved in extended
490 conversations. Other implementations of 802.3ad may
491 or may not tolerate this noncompliance.
492
493 The default value is layer2. This option was added in bonding
494version 2.6.3. In earlier versions of bonding, this parameter does
495not exist, and the layer2 policy is the only policy.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496
497
4983. Configuring Bonding Devices
499==============================
500
501 There are, essentially, two methods for configuring bonding:
502with support from the distro's network initialization scripts, and
503without. Distros generally use one of two packages for the network
504initialization scripts: initscripts or sysconfig. Recent versions of
505these packages have support for bonding, while older versions do not.
506
507 We will first describe the options for configuring bonding for
508distros using versions of initscripts and sysconfig with full or
509partial support for bonding, then provide information on enabling
510bonding without support from the network initialization scripts (i.e.,
511older versions of initscripts or sysconfig).
512
513 If you're unsure whether your distro uses sysconfig or
514initscripts, or don't know if it's new enough, have no fear.
515Determining this is fairly straightforward.
516
517 First, issue the command:
518
519$ rpm -qf /sbin/ifup
520
521 It will respond with a line of text starting with either
522"initscripts" or "sysconfig," followed by some numbers. This is the
523package that provides your network initialization scripts.
524
525 Next, to determine if your installation supports bonding,
526issue the command:
527
528$ grep ifenslave /sbin/ifup
529
530 If this returns any matches, then your initscripts or
531sysconfig has support for bonding.
532
5333.1 Configuration with sysconfig support
534----------------------------------------
535
536 This section applies to distros using a version of sysconfig
537with bonding support, for example, SuSE Linux Enterprise Server 9.
538
539 SuSE SLES 9's networking configuration system does support
540bonding, however, at this writing, the YaST system configuration
541frontend does not provide any means to work with bonding devices.
542Bonding devices can be managed by hand, however, as follows.
543
544 First, if they have not already been configured, configure the
545slave devices. On SLES 9, this is most easily done by running the
546yast2 sysconfig configuration utility. The goal is for to create an
547ifcfg-id file for each slave device. The simplest way to accomplish
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700548this is to configure the devices for DHCP (this is only to get the
549file ifcfg-id file created; see below for some issues with DHCP). The
550name of the configuration file for each device will be of the form:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551
552ifcfg-id-xx:xx:xx:xx:xx:xx
553
554 Where the "xx" portion will be replaced with the digits from
555the device's permanent MAC address.
556
557 Once the set of ifcfg-id-xx:xx:xx:xx:xx:xx files has been
558created, it is necessary to edit the configuration files for the slave
559devices (the MAC addresses correspond to those of the slave devices).
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700560Before editing, the file will contain multiple lines, and will look
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561something like this:
562
563BOOTPROTO='dhcp'
564STARTMODE='on'
565USERCTL='no'
566UNIQUE='XNzu.WeZGOGF+4wE'
567_nm_name='bus-pci-0001:61:01.0'
568
569 Change the BOOTPROTO and STARTMODE lines to the following:
570
571BOOTPROTO='none'
572STARTMODE='off'
573
574 Do not alter the UNIQUE or _nm_name lines. Remove any other
575lines (USERCTL, etc).
576
577 Once the ifcfg-id-xx:xx:xx:xx:xx:xx files have been modified,
578it's time to create the configuration file for the bonding device
579itself. This file is named ifcfg-bondX, where X is the number of the
580bonding device to create, starting at 0. The first such file is
581ifcfg-bond0, the second is ifcfg-bond1, and so on. The sysconfig
582network configuration system will correctly start multiple instances
583of bonding.
584
585 The contents of the ifcfg-bondX file is as follows:
586
587BOOTPROTO="static"
588BROADCAST="10.0.2.255"
589IPADDR="10.0.2.10"
590NETMASK="255.255.0.0"
591NETWORK="10.0.2.0"
592REMOTE_IPADDR=""
593STARTMODE="onboot"
594BONDING_MASTER="yes"
595BONDING_MODULE_OPTS="mode=active-backup miimon=100"
596BONDING_SLAVE0="eth0"
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700597BONDING_SLAVE1="bus-pci-0000:06:08.1"
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598
599 Replace the sample BROADCAST, IPADDR, NETMASK and NETWORK
600values with the appropriate values for your network.
601
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 The STARTMODE specifies when the device is brought online.
603The possible values are:
604
605 onboot: The device is started at boot time. If you're not
606 sure, this is probably what you want.
607
608 manual: The device is started only when ifup is called
609 manually. Bonding devices may be configured this
610 way if you do not wish them to start automatically
611 at boot for some reason.
612
613 hotplug: The device is started by a hotplug event. This is not
614 a valid choice for a bonding device.
615
616 off or ignore: The device configuration is ignored.
617
618 The line BONDING_MASTER='yes' indicates that the device is a
619bonding master device. The only useful value is "yes."
620
621 The contents of BONDING_MODULE_OPTS are supplied to the
622instance of the bonding module for this device. Specify the options
623for the bonding mode, link monitoring, and so on here. Do not include
624the max_bonds bonding parameter; this will confuse the configuration
625system if you have multiple bonding devices.
626
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700627 Finally, supply one BONDING_SLAVEn="slave device" for each
628slave. where "n" is an increasing value, one for each slave. The
629"slave device" is either an interface name, e.g., "eth0", or a device
630specifier for the network device. The interface name is easier to
631find, but the ethN names are subject to change at boot time if, e.g.,
632a device early in the sequence has failed. The device specifiers
633(bus-pci-0000:06:08.1 in the example above) specify the physical
634network device, and will not change unless the device's bus location
635changes (for example, it is moved from one PCI slot to another). The
636example above uses one of each type for demonstration purposes; most
637configurations will choose one or the other for all slave devices.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638
639 When all configuration files have been modified or created,
640networking must be restarted for the configuration changes to take
641effect. This can be accomplished via the following:
642
643# /etc/init.d/network restart
644
645 Note that the network control script (/sbin/ifdown) will
646remove the bonding module as part of the network shutdown processing,
647so it is not necessary to remove the module by hand if, e.g., the
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700648module parameters have changed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649
650 Also, at this writing, YaST/YaST2 will not manage bonding
651devices (they do not show bonding interfaces on its list of network
652devices). It is necessary to edit the configuration file by hand to
653change the bonding configuration.
654
655 Additional general options and details of the ifcfg file
656format can be found in an example ifcfg template file:
657
658/etc/sysconfig/network/ifcfg.template
659
660 Note that the template does not document the various BONDING_
661settings described above, but does describe many of the other options.
662
Jay Vosburgh00354cf2005-07-21 12:18:02 -07006633.1.1 Using DHCP with sysconfig
664-------------------------------
665
666 Under sysconfig, configuring a device with BOOTPROTO='dhcp'
667will cause it to query DHCP for its IP address information. At this
668writing, this does not function for bonding devices; the scripts
669attempt to obtain the device address from DHCP prior to adding any of
670the slave devices. Without active slaves, the DHCP requests are not
671sent to the network.
672
6733.1.2 Configuring Multiple Bonds with sysconfig
674-----------------------------------------------
675
676 The sysconfig network initialization system is capable of
677handling multiple bonding devices. All that is necessary is for each
678bonding instance to have an appropriately configured ifcfg-bondX file
679(as described above). Do not specify the "max_bonds" parameter to any
680instance of bonding, as this will confuse sysconfig. If you require
681multiple bonding devices with identical parameters, create multiple
682ifcfg-bondX files.
683
684 Because the sysconfig scripts supply the bonding module
685options in the ifcfg-bondX file, it is not necessary to add them to
686the system /etc/modules.conf or /etc/modprobe.conf configuration file.
687
Linus Torvalds1da177e2005-04-16 15:20:36 -07006883.2 Configuration with initscripts support
689------------------------------------------
690
691 This section applies to distros using a version of initscripts
692with bonding support, for example, Red Hat Linux 9 or Red Hat
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700693Enterprise Linux version 3 or 4. On these systems, the network
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694initialization scripts have some knowledge of bonding, and can be
695configured to control bonding devices.
696
697 These distros will not automatically load the network adapter
698driver unless the ethX device is configured with an IP address.
699Because of this constraint, users must manually configure a
700network-script file for all physical adapters that will be members of
701a bondX link. Network script files are located in the directory:
702
703/etc/sysconfig/network-scripts
704
705 The file name must be prefixed with "ifcfg-eth" and suffixed
706with the adapter's physical adapter number. For example, the script
707for eth0 would be named /etc/sysconfig/network-scripts/ifcfg-eth0.
708Place the following text in the file:
709
710DEVICE=eth0
711USERCTL=no
712ONBOOT=yes
713MASTER=bond0
714SLAVE=yes
715BOOTPROTO=none
716
717 The DEVICE= line will be different for every ethX device and
718must correspond with the name of the file, i.e., ifcfg-eth1 must have
719a device line of DEVICE=eth1. The setting of the MASTER= line will
720also depend on the final bonding interface name chosen for your bond.
721As with other network devices, these typically start at 0, and go up
722one for each device, i.e., the first bonding instance is bond0, the
723second is bond1, and so on.
724
725 Next, create a bond network script. The file name for this
726script will be /etc/sysconfig/network-scripts/ifcfg-bondX where X is
727the number of the bond. For bond0 the file is named "ifcfg-bond0",
728for bond1 it is named "ifcfg-bond1", and so on. Within that file,
729place the following text:
730
731DEVICE=bond0
732IPADDR=192.168.1.1
733NETMASK=255.255.255.0
734NETWORK=192.168.1.0
735BROADCAST=192.168.1.255
736ONBOOT=yes
737BOOTPROTO=none
738USERCTL=no
739
740 Be sure to change the networking specific lines (IPADDR,
741NETMASK, NETWORK and BROADCAST) to match your network configuration.
742
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700743 Finally, it is necessary to edit /etc/modules.conf (or
744/etc/modprobe.conf, depending upon your distro) to load the bonding
745module with your desired options when the bond0 interface is brought
746up. The following lines in /etc/modules.conf (or modprobe.conf) will
747load the bonding module, and select its options:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748
749alias bond0 bonding
750options bond0 mode=balance-alb miimon=100
751
752 Replace the sample parameters with the appropriate set of
753options for your configuration.
754
755 Finally run "/etc/rc.d/init.d/network restart" as root. This
756will restart the networking subsystem and your bond link should be now
757up and running.
758
Jay Vosburgh00354cf2005-07-21 12:18:02 -07007593.2.1 Using DHCP with initscripts
760---------------------------------
761
762 Recent versions of initscripts (the version supplied with
763Fedora Core 3 and Red Hat Enterprise Linux 4 is reported to work) do
764have support for assigning IP information to bonding devices via DHCP.
765
766 To configure bonding for DHCP, configure it as described
767above, except replace the line "BOOTPROTO=none" with "BOOTPROTO=dhcp"
768and add a line consisting of "TYPE=Bonding". Note that the TYPE value
769is case sensitive.
770
7713.2.2 Configuring Multiple Bonds with initscripts
772-------------------------------------------------
773
774 At this writing, the initscripts package does not directly
775support loading the bonding driver multiple times, so the process for
776doing so is the same as described in the "Configuring Multiple Bonds
777Manually" section, below.
778
779 NOTE: It has been observed that some Red Hat supplied kernels
780are apparently unable to rename modules at load time (the "-obonding1"
781part). Attempts to pass that option to modprobe will produce an
782"Operation not permitted" error. This has been reported on some
783Fedora Core kernels, and has been seen on RHEL 4 as well. On kernels
784exhibiting this problem, it will be impossible to configure multiple
785bonds with differing parameters.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786
7873.3 Configuring Bonding Manually
788--------------------------------
789
790 This section applies to distros whose network initialization
791scripts (the sysconfig or initscripts package) do not have specific
792knowledge of bonding. One such distro is SuSE Linux Enterprise Server
793version 8.
794
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700795 The general method for these systems is to place the bonding
796module parameters into /etc/modules.conf or /etc/modprobe.conf (as
797appropriate for the installed distro), then add modprobe and/or
798ifenslave commands to the system's global init script. The name of
799the global init script differs; for sysconfig, it is
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800/etc/init.d/boot.local and for initscripts it is /etc/rc.d/rc.local.
801
802 For example, if you wanted to make a simple bond of two e100
803devices (presumed to be eth0 and eth1), and have it persist across
804reboots, edit the appropriate file (/etc/init.d/boot.local or
805/etc/rc.d/rc.local), and add the following:
806
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700807modprobe bonding mode=balance-alb miimon=100
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808modprobe e100
809ifconfig bond0 192.168.1.1 netmask 255.255.255.0 up
810ifenslave bond0 eth0
811ifenslave bond0 eth1
812
813 Replace the example bonding module parameters and bond0
814network configuration (IP address, netmask, etc) with the appropriate
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700815values for your configuration.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816
817 Unfortunately, this method will not provide support for the
818ifup and ifdown scripts on the bond devices. To reload the bonding
819configuration, it is necessary to run the initialization script, e.g.,
820
821# /etc/init.d/boot.local
822
823 or
824
825# /etc/rc.d/rc.local
826
827 It may be desirable in such a case to create a separate script
828which only initializes the bonding configuration, then call that
829separate script from within boot.local. This allows for bonding to be
830enabled without re-running the entire global init script.
831
832 To shut down the bonding devices, it is necessary to first
833mark the bonding device itself as being down, then remove the
834appropriate device driver modules. For our example above, you can do
835the following:
836
837# ifconfig bond0 down
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700838# rmmod bonding
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839# rmmod e100
840
841 Again, for convenience, it may be desirable to create a script
842with these commands.
843
844
Jay Vosburgh00354cf2005-07-21 12:18:02 -07008453.3.1 Configuring Multiple Bonds Manually
846-----------------------------------------
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847
848 This section contains information on configuring multiple
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700849bonding devices with differing options for those systems whose network
850initialization scripts lack support for configuring multiple bonds.
851
852 If you require multiple bonding devices, but all with the same
853options, you may wish to use the "max_bonds" module parameter,
854documented above.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855
856 To create multiple bonding devices with differing options, it
857is necessary to load the bonding driver multiple times. Note that
858current versions of the sysconfig network initialization scripts
859handle this automatically; if your distro uses these scripts, no
860special action is needed. See the section Configuring Bonding
861Devices, above, if you're not sure about your network initialization
862scripts.
863
864 To load multiple instances of the module, it is necessary to
865specify a different name for each instance (the module loading system
866requires that every loaded module, even multiple instances of the same
867module, have a unique name). This is accomplished by supplying
868multiple sets of bonding options in /etc/modprobe.conf, for example:
869
870alias bond0 bonding
871options bond0 -o bond0 mode=balance-rr miimon=100
872
873alias bond1 bonding
874options bond1 -o bond1 mode=balance-alb miimon=50
875
876 will load the bonding module two times. The first instance is
877named "bond0" and creates the bond0 device in balance-rr mode with an
878miimon of 100. The second instance is named "bond1" and creates the
879bond1 device in balance-alb mode with an miimon of 50.
880
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700881 In some circumstances (typically with older distributions),
882the above does not work, and the second bonding instance never sees
883its options. In that case, the second options line can be substituted
884as follows:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700886install bonding1 /sbin/modprobe bonding -obond1 mode=balance-alb miimon=50
887
888 This may be repeated any number of times, specifying a new and
889unique name in place of bond1 for each subsequent instance.
890
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891
8925. Querying Bonding Configuration
893=================================
894
8955.1 Bonding Configuration
896-------------------------
897
898 Each bonding device has a read-only file residing in the
899/proc/net/bonding directory. The file contents include information
900about the bonding configuration, options and state of each slave.
901
902 For example, the contents of /proc/net/bonding/bond0 after the
903driver is loaded with parameters of mode=0 and miimon=1000 is
904generally as follows:
905
906 Ethernet Channel Bonding Driver: 2.6.1 (October 29, 2004)
907 Bonding Mode: load balancing (round-robin)
908 Currently Active Slave: eth0
909 MII Status: up
910 MII Polling Interval (ms): 1000
911 Up Delay (ms): 0
912 Down Delay (ms): 0
913
914 Slave Interface: eth1
915 MII Status: up
916 Link Failure Count: 1
917
918 Slave Interface: eth0
919 MII Status: up
920 Link Failure Count: 1
921
922 The precise format and contents will change depending upon the
923bonding configuration, state, and version of the bonding driver.
924
9255.2 Network configuration
926-------------------------
927
928 The network configuration can be inspected using the ifconfig
929command. Bonding devices will have the MASTER flag set; Bonding slave
930devices will have the SLAVE flag set. The ifconfig output does not
931contain information on which slaves are associated with which masters.
932
933 In the example below, the bond0 interface is the master
934(MASTER) while eth0 and eth1 are slaves (SLAVE). Notice all slaves of
935bond0 have the same MAC address (HWaddr) as bond0 for all modes except
936TLB and ALB that require a unique MAC address for each slave.
937
938# /sbin/ifconfig
939bond0 Link encap:Ethernet HWaddr 00:C0:F0:1F:37:B4
940 inet addr:XXX.XXX.XXX.YYY Bcast:XXX.XXX.XXX.255 Mask:255.255.252.0
941 UP BROADCAST RUNNING MASTER MULTICAST MTU:1500 Metric:1
942 RX packets:7224794 errors:0 dropped:0 overruns:0 frame:0
943 TX packets:3286647 errors:1 dropped:0 overruns:1 carrier:0
944 collisions:0 txqueuelen:0
945
946eth0 Link encap:Ethernet HWaddr 00:C0:F0:1F:37:B4
947 inet addr:XXX.XXX.XXX.YYY Bcast:XXX.XXX.XXX.255 Mask:255.255.252.0
948 UP BROADCAST RUNNING SLAVE MULTICAST MTU:1500 Metric:1
949 RX packets:3573025 errors:0 dropped:0 overruns:0 frame:0
950 TX packets:1643167 errors:1 dropped:0 overruns:1 carrier:0
951 collisions:0 txqueuelen:100
952 Interrupt:10 Base address:0x1080
953
954eth1 Link encap:Ethernet HWaddr 00:C0:F0:1F:37:B4
955 inet addr:XXX.XXX.XXX.YYY Bcast:XXX.XXX.XXX.255 Mask:255.255.252.0
956 UP BROADCAST RUNNING SLAVE MULTICAST MTU:1500 Metric:1
957 RX packets:3651769 errors:0 dropped:0 overruns:0 frame:0
958 TX packets:1643480 errors:0 dropped:0 overruns:0 carrier:0
959 collisions:0 txqueuelen:100
960 Interrupt:9 Base address:0x1400
961
9626. Switch Configuration
963=======================
964
965 For this section, "switch" refers to whatever system the
966bonded devices are directly connected to (i.e., where the other end of
967the cable plugs into). This may be an actual dedicated switch device,
968or it may be another regular system (e.g., another computer running
969Linux),
970
971 The active-backup, balance-tlb and balance-alb modes do not
972require any specific configuration of the switch.
973
974 The 802.3ad mode requires that the switch have the appropriate
975ports configured as an 802.3ad aggregation. The precise method used
976to configure this varies from switch to switch, but, for example, a
977Cisco 3550 series switch requires that the appropriate ports first be
978grouped together in a single etherchannel instance, then that
979etherchannel is set to mode "lacp" to enable 802.3ad (instead of
980standard EtherChannel).
981
982 The balance-rr, balance-xor and broadcast modes generally
983require that the switch have the appropriate ports grouped together.
984The nomenclature for such a group differs between switches, it may be
985called an "etherchannel" (as in the Cisco example, above), a "trunk
986group" or some other similar variation. For these modes, each switch
987will also have its own configuration options for the switch's transmit
988policy to the bond. Typical choices include XOR of either the MAC or
989IP addresses. The transmit policy of the two peers does not need to
990match. For these three modes, the bonding mode really selects a
991transmit policy for an EtherChannel group; all three will interoperate
992with another EtherChannel group.
993
994
9957. 802.1q VLAN Support
996======================
997
998 It is possible to configure VLAN devices over a bond interface
999using the 8021q driver. However, only packets coming from the 8021q
1000driver and passing through bonding will be tagged by default. Self
1001generated packets, for example, bonding's learning packets or ARP
1002packets generated by either ALB mode or the ARP monitor mechanism, are
1003tagged internally by bonding itself. As a result, bonding must
1004"learn" the VLAN IDs configured above it, and use those IDs to tag
1005self generated packets.
1006
1007 For reasons of simplicity, and to support the use of adapters
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001008that can do VLAN hardware acceleration offloading, the bonding
1009interface declares itself as fully hardware offloading capable, it gets
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010the add_vid/kill_vid notifications to gather the necessary
1011information, and it propagates those actions to the slaves. In case
1012of mixed adapter types, hardware accelerated tagged packets that
1013should go through an adapter that is not offloading capable are
1014"un-accelerated" by the bonding driver so the VLAN tag sits in the
1015regular location.
1016
1017 VLAN interfaces *must* be added on top of a bonding interface
1018only after enslaving at least one slave. The bonding interface has a
1019hardware address of 00:00:00:00:00:00 until the first slave is added.
1020If the VLAN interface is created prior to the first enslavement, it
1021would pick up the all-zeroes hardware address. Once the first slave
1022is attached to the bond, the bond device itself will pick up the
1023slave's hardware address, which is then available for the VLAN device.
1024
1025 Also, be aware that a similar problem can occur if all slaves
1026are released from a bond that still has one or more VLAN interfaces on
1027top of it. When a new slave is added, the bonding interface will
1028obtain its hardware address from the first slave, which might not
1029match the hardware address of the VLAN interfaces (which was
1030ultimately copied from an earlier slave).
1031
1032 There are two methods to insure that the VLAN device operates
1033with the correct hardware address if all slaves are removed from a
1034bond interface:
1035
1036 1. Remove all VLAN interfaces then recreate them
1037
1038 2. Set the bonding interface's hardware address so that it
1039matches the hardware address of the VLAN interfaces.
1040
1041 Note that changing a VLAN interface's HW address would set the
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001042underlying device -- i.e. the bonding interface -- to promiscuous
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043mode, which might not be what you want.
1044
1045
10468. Link Monitoring
1047==================
1048
1049 The bonding driver at present supports two schemes for
1050monitoring a slave device's link state: the ARP monitor and the MII
1051monitor.
1052
1053 At the present time, due to implementation restrictions in the
1054bonding driver itself, it is not possible to enable both ARP and MII
1055monitoring simultaneously.
1056
10578.1 ARP Monitor Operation
1058-------------------------
1059
1060 The ARP monitor operates as its name suggests: it sends ARP
1061queries to one or more designated peer systems on the network, and
1062uses the response as an indication that the link is operating. This
1063gives some assurance that traffic is actually flowing to and from one
1064or more peers on the local network.
1065
1066 The ARP monitor relies on the device driver itself to verify
1067that traffic is flowing. In particular, the driver must keep up to
1068date the last receive time, dev->last_rx, and transmit start time,
1069dev->trans_start. If these are not updated by the driver, then the
1070ARP monitor will immediately fail any slaves using that driver, and
1071those slaves will stay down. If networking monitoring (tcpdump, etc)
1072shows the ARP requests and replies on the network, then it may be that
1073your device driver is not updating last_rx and trans_start.
1074
10758.2 Configuring Multiple ARP Targets
1076------------------------------------
1077
1078 While ARP monitoring can be done with just one target, it can
1079be useful in a High Availability setup to have several targets to
1080monitor. In the case of just one target, the target itself may go
1081down or have a problem making it unresponsive to ARP requests. Having
1082an additional target (or several) increases the reliability of the ARP
1083monitoring.
1084
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001085 Multiple ARP targets must be separated by commas as follows:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086
1087# example options for ARP monitoring with three targets
1088alias bond0 bonding
1089options bond0 arp_interval=60 arp_ip_target=192.168.0.1,192.168.0.3,192.168.0.9
1090
1091 For just a single target the options would resemble:
1092
1093# example options for ARP monitoring with one target
1094alias bond0 bonding
1095options bond0 arp_interval=60 arp_ip_target=192.168.0.100
1096
1097
10988.3 MII Monitor Operation
1099-------------------------
1100
1101 The MII monitor monitors only the carrier state of the local
1102network interface. It accomplishes this in one of three ways: by
1103depending upon the device driver to maintain its carrier state, by
1104querying the device's MII registers, or by making an ethtool query to
1105the device.
1106
1107 If the use_carrier module parameter is 1 (the default value),
1108then the MII monitor will rely on the driver for carrier state
1109information (via the netif_carrier subsystem). As explained in the
1110use_carrier parameter information, above, if the MII monitor fails to
1111detect carrier loss on the device (e.g., when the cable is physically
1112disconnected), it may be that the driver does not support
1113netif_carrier.
1114
1115 If use_carrier is 0, then the MII monitor will first query the
1116device's (via ioctl) MII registers and check the link state. If that
1117request fails (not just that it returns carrier down), then the MII
1118monitor will make an ethtool ETHOOL_GLINK request to attempt to obtain
1119the same information. If both methods fail (i.e., the driver either
1120does not support or had some error in processing both the MII register
1121and ethtool requests), then the MII monitor will assume the link is
1122up.
1123
11249. Potential Sources of Trouble
1125===============================
1126
11279.1 Adventures in Routing
1128-------------------------
1129
1130 When bonding is configured, it is important that the slave
1131devices not have routes that supercede routes of the master (or,
1132generally, not have routes at all). For example, suppose the bonding
1133device bond0 has two slaves, eth0 and eth1, and the routing table is
1134as follows:
1135
1136Kernel IP routing table
1137Destination Gateway Genmask Flags MSS Window irtt Iface
113810.0.0.0 0.0.0.0 255.255.0.0 U 40 0 0 eth0
113910.0.0.0 0.0.0.0 255.255.0.0 U 40 0 0 eth1
114010.0.0.0 0.0.0.0 255.255.0.0 U 40 0 0 bond0
1141127.0.0.0 0.0.0.0 255.0.0.0 U 40 0 0 lo
1142
1143 This routing configuration will likely still update the
1144receive/transmit times in the driver (needed by the ARP monitor), but
1145may bypass the bonding driver (because outgoing traffic to, in this
1146case, another host on network 10 would use eth0 or eth1 before bond0).
1147
1148 The ARP monitor (and ARP itself) may become confused by this
1149configuration, because ARP requests (generated by the ARP monitor)
1150will be sent on one interface (bond0), but the corresponding reply
1151will arrive on a different interface (eth0). This reply looks to ARP
1152as an unsolicited ARP reply (because ARP matches replies on an
1153interface basis), and is discarded. The MII monitor is not affected
1154by the state of the routing table.
1155
1156 The solution here is simply to insure that slaves do not have
1157routes of their own, and if for some reason they must, those routes do
1158not supercede routes of their master. This should generally be the
1159case, but unusual configurations or errant manual or automatic static
1160route additions may cause trouble.
1161
11629.2 Ethernet Device Renaming
1163----------------------------
1164
1165 On systems with network configuration scripts that do not
1166associate physical devices directly with network interface names (so
1167that the same physical device always has the same "ethX" name), it may
1168be necessary to add some special logic to either /etc/modules.conf or
1169/etc/modprobe.conf (depending upon which is installed on the system).
1170
1171 For example, given a modules.conf containing the following:
1172
1173alias bond0 bonding
1174options bond0 mode=some-mode miimon=50
1175alias eth0 tg3
1176alias eth1 tg3
1177alias eth2 e1000
1178alias eth3 e1000
1179
1180 If neither eth0 and eth1 are slaves to bond0, then when the
1181bond0 interface comes up, the devices may end up reordered. This
1182happens because bonding is loaded first, then its slave device's
1183drivers are loaded next. Since no other drivers have been loaded,
1184when the e1000 driver loads, it will receive eth0 and eth1 for its
1185devices, but the bonding configuration tries to enslave eth2 and eth3
1186(which may later be assigned to the tg3 devices).
1187
1188 Adding the following:
1189
1190add above bonding e1000 tg3
1191
1192 causes modprobe to load e1000 then tg3, in that order, when
1193bonding is loaded. This command is fully documented in the
1194modules.conf manual page.
1195
1196 On systems utilizing modprobe.conf (or modprobe.conf.local),
1197an equivalent problem can occur. In this case, the following can be
1198added to modprobe.conf (or modprobe.conf.local, as appropriate), as
1199follows (all on one line; it has been split here for clarity):
1200
1201install bonding /sbin/modprobe tg3; /sbin/modprobe e1000;
1202 /sbin/modprobe --ignore-install bonding
1203
1204 This will, when loading the bonding module, rather than
1205performing the normal action, instead execute the provided command.
1206This command loads the device drivers in the order needed, then calls
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001207modprobe with --ignore-install to cause the normal action to then take
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208place. Full documentation on this can be found in the modprobe.conf
1209and modprobe manual pages.
1210
12119.3. Painfully Slow Or No Failed Link Detection By Miimon
1212---------------------------------------------------------
1213
1214 By default, bonding enables the use_carrier option, which
1215instructs bonding to trust the driver to maintain carrier state.
1216
1217 As discussed in the options section, above, some drivers do
1218not support the netif_carrier_on/_off link state tracking system.
1219With use_carrier enabled, bonding will always see these links as up,
1220regardless of their actual state.
1221
1222 Additionally, other drivers do support netif_carrier, but do
1223not maintain it in real time, e.g., only polling the link state at
1224some fixed interval. In this case, miimon will detect failures, but
1225only after some long period of time has expired. If it appears that
1226miimon is very slow in detecting link failures, try specifying
1227use_carrier=0 to see if that improves the failure detection time. If
1228it does, then it may be that the driver checks the carrier state at a
1229fixed interval, but does not cache the MII register values (so the
1230use_carrier=0 method of querying the registers directly works). If
1231use_carrier=0 does not improve the failover, then the driver may cache
1232the registers, or the problem may be elsewhere.
1233
1234 Also, remember that miimon only checks for the device's
1235carrier state. It has no way to determine the state of devices on or
1236beyond other ports of a switch, or if a switch is refusing to pass
1237traffic while still maintaining carrier on.
1238
123910. SNMP agents
1240===============
1241
1242 If running SNMP agents, the bonding driver should be loaded
1243before any network drivers participating in a bond. This requirement
1244is due to the the interface index (ipAdEntIfIndex) being associated to
1245the first interface found with a given IP address. That is, there is
1246only one ipAdEntIfIndex for each IP address. For example, if eth0 and
1247eth1 are slaves of bond0 and the driver for eth0 is loaded before the
1248bonding driver, the interface for the IP address will be associated
1249with the eth0 interface. This configuration is shown below, the IP
1250address 192.168.1.1 has an interface index of 2 which indexes to eth0
1251in the ifDescr table (ifDescr.2).
1252
1253 interfaces.ifTable.ifEntry.ifDescr.1 = lo
1254 interfaces.ifTable.ifEntry.ifDescr.2 = eth0
1255 interfaces.ifTable.ifEntry.ifDescr.3 = eth1
1256 interfaces.ifTable.ifEntry.ifDescr.4 = eth2
1257 interfaces.ifTable.ifEntry.ifDescr.5 = eth3
1258 interfaces.ifTable.ifEntry.ifDescr.6 = bond0
1259 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.10.10.10 = 5
1260 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.192.168.1.1 = 2
1261 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.74.20.94 = 4
1262 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.127.0.0.1 = 1
1263
1264 This problem is avoided by loading the bonding driver before
1265any network drivers participating in a bond. Below is an example of
1266loading the bonding driver first, the IP address 192.168.1.1 is
1267correctly associated with ifDescr.2.
1268
1269 interfaces.ifTable.ifEntry.ifDescr.1 = lo
1270 interfaces.ifTable.ifEntry.ifDescr.2 = bond0
1271 interfaces.ifTable.ifEntry.ifDescr.3 = eth0
1272 interfaces.ifTable.ifEntry.ifDescr.4 = eth1
1273 interfaces.ifTable.ifEntry.ifDescr.5 = eth2
1274 interfaces.ifTable.ifEntry.ifDescr.6 = eth3
1275 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.10.10.10 = 6
1276 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.192.168.1.1 = 2
1277 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.74.20.94 = 5
1278 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.127.0.0.1 = 1
1279
1280 While some distributions may not report the interface name in
1281ifDescr, the association between the IP address and IfIndex remains
1282and SNMP functions such as Interface_Scan_Next will report that
1283association.
1284
128511. Promiscuous mode
1286====================
1287
1288 When running network monitoring tools, e.g., tcpdump, it is
1289common to enable promiscuous mode on the device, so that all traffic
1290is seen (instead of seeing only traffic destined for the local host).
1291The bonding driver handles promiscuous mode changes to the bonding
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001292master device (e.g., bond0), and propagates the setting to the slave
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293devices.
1294
1295 For the balance-rr, balance-xor, broadcast, and 802.3ad modes,
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001296the promiscuous mode setting is propagated to all slaves.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297
1298 For the active-backup, balance-tlb and balance-alb modes, the
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001299promiscuous mode setting is propagated only to the active slave.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300
1301 For balance-tlb mode, the active slave is the slave currently
1302receiving inbound traffic.
1303
1304 For balance-alb mode, the active slave is the slave used as a
1305"primary." This slave is used for mode-specific control traffic, for
1306sending to peers that are unassigned or if the load is unbalanced.
1307
1308 For the active-backup, balance-tlb and balance-alb modes, when
1309the active slave changes (e.g., due to a link failure), the
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001310promiscuous setting will be propagated to the new active slave.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700131212. Configuring Bonding for High Availability
1313=============================================
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314
1315 High Availability refers to configurations that provide
1316maximum network availability by having redundant or backup devices,
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001317links or switches between the host and the rest of the world. The
1318goal is to provide the maximum availability of network connectivity
1319(i.e., the network always works), even though other configurations
1320could provide higher throughput.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321
132212.1 High Availability in a Single Switch Topology
1323--------------------------------------------------
1324
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001325 If two hosts (or a host and a single switch) are directly
1326connected via multiple physical links, then there is no availability
1327penalty to optimizing for maximum bandwidth. In this case, there is
1328only one switch (or peer), so if it fails, there is no alternative
1329access to fail over to. Additionally, the bonding load balance modes
1330support link monitoring of their members, so if individual links fail,
1331the load will be rebalanced across the remaining devices.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001333 See Section 13, "Configuring Bonding for Maximum Throughput"
1334for information on configuring bonding with one peer device.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335
133612.2 High Availability in a Multiple Switch Topology
1337----------------------------------------------------
1338
1339 With multiple switches, the configuration of bonding and the
1340network changes dramatically. In multiple switch topologies, there is
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001341a trade off between network availability and usable bandwidth.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342
1343 Below is a sample network, configured to maximize the
1344availability of the network:
1345
1346 | |
1347 |port3 port3|
1348 +-----+----+ +-----+----+
1349 | |port2 ISL port2| |
1350 | switch A +--------------------------+ switch B |
1351 | | | |
1352 +-----+----+ +-----++---+
1353 |port1 port1|
1354 | +-------+ |
1355 +-------------+ host1 +---------------+
1356 eth0 +-------+ eth1
1357
1358 In this configuration, there is a link between the two
1359switches (ISL, or inter switch link), and multiple ports connecting to
1360the outside world ("port3" on each switch). There is no technical
1361reason that this could not be extended to a third switch.
1362
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700136312.2.1 HA Bonding Mode Selection for Multiple Switch Topology
1364-------------------------------------------------------------
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001366 In a topology such as the example above, the active-backup and
1367broadcast modes are the only useful bonding modes when optimizing for
1368availability; the other modes require all links to terminate on the
1369same peer for them to behave rationally.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370
1371active-backup: This is generally the preferred mode, particularly if
1372 the switches have an ISL and play together well. If the
1373 network configuration is such that one switch is specifically
1374 a backup switch (e.g., has lower capacity, higher cost, etc),
1375 then the primary option can be used to insure that the
1376 preferred link is always used when it is available.
1377
1378broadcast: This mode is really a special purpose mode, and is suitable
1379 only for very specific needs. For example, if the two
1380 switches are not connected (no ISL), and the networks beyond
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001381 them are totally independent. In this case, if it is
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382 necessary for some specific one-way traffic to reach both
1383 independent networks, then the broadcast mode may be suitable.
1384
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700138512.2.2 HA Link Monitoring Selection for Multiple Switch Topology
1386----------------------------------------------------------------
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387
1388 The choice of link monitoring ultimately depends upon your
1389switch. If the switch can reliably fail ports in response to other
1390failures, then either the MII or ARP monitors should work. For
1391example, in the above example, if the "port3" link fails at the remote
1392end, the MII monitor has no direct means to detect this. The ARP
1393monitor could be configured with a target at the remote end of port3,
1394thus detecting that failure without switch support.
1395
1396 In general, however, in a multiple switch topology, the ARP
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001397monitor can provide a higher level of reliability in detecting end to
1398end connectivity failures (which may be caused by the failure of any
1399individual component to pass traffic for any reason). Additionally,
1400the ARP monitor should be configured with multiple targets (at least
1401one for each switch in the network). This will insure that,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402regardless of which switch is active, the ARP monitor has a suitable
1403target to query.
1404
1405
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700140613. Configuring Bonding for Maximum Throughput
1407==============================================
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700140913.1 Maximizing Throughput in a Single Switch Topology
1410------------------------------------------------------
1411
1412 In a single switch configuration, the best method to maximize
1413throughput depends upon the application and network environment. The
1414various load balancing modes each have strengths and weaknesses in
1415different environments, as detailed below.
1416
1417 For this discussion, we will break down the topologies into
1418two categories. Depending upon the destination of most traffic, we
1419categorize them into either "gatewayed" or "local" configurations.
1420
1421 In a gatewayed configuration, the "switch" is acting primarily
1422as a router, and the majority of traffic passes through this router to
1423other networks. An example would be the following:
1424
1425
1426 +----------+ +----------+
1427 | |eth0 port1| | to other networks
1428 | Host A +---------------------+ router +------------------->
1429 | +---------------------+ | Hosts B and C are out
1430 | |eth1 port2| | here somewhere
1431 +----------+ +----------+
1432
1433 The router may be a dedicated router device, or another host
1434acting as a gateway. For our discussion, the important point is that
1435the majority of traffic from Host A will pass through the router to
1436some other network before reaching its final destination.
1437
1438 In a gatewayed network configuration, although Host A may
1439communicate with many other systems, all of its traffic will be sent
1440and received via one other peer on the local network, the router.
1441
1442 Note that the case of two systems connected directly via
1443multiple physical links is, for purposes of configuring bonding, the
1444same as a gatewayed configuration. In that case, it happens that all
1445traffic is destined for the "gateway" itself, not some other network
1446beyond the gateway.
1447
1448 In a local configuration, the "switch" is acting primarily as
1449a switch, and the majority of traffic passes through this switch to
1450reach other stations on the same network. An example would be the
1451following:
1452
1453 +----------+ +----------+ +--------+
1454 | |eth0 port1| +-------+ Host B |
1455 | Host A +------------+ switch |port3 +--------+
1456 | +------------+ | +--------+
1457 | |eth1 port2| +------------------+ Host C |
1458 +----------+ +----------+port4 +--------+
1459
1460
1461 Again, the switch may be a dedicated switch device, or another
1462host acting as a gateway. For our discussion, the important point is
1463that the majority of traffic from Host A is destined for other hosts
1464on the same local network (Hosts B and C in the above example).
1465
1466 In summary, in a gatewayed configuration, traffic to and from
1467the bonded device will be to the same MAC level peer on the network
1468(the gateway itself, i.e., the router), regardless of its final
1469destination. In a local configuration, traffic flows directly to and
1470from the final destinations, thus, each destination (Host B, Host C)
1471will be addressed directly by their individual MAC addresses.
1472
1473 This distinction between a gatewayed and a local network
1474configuration is important because many of the load balancing modes
1475available use the MAC addresses of the local network source and
1476destination to make load balancing decisions. The behavior of each
1477mode is described below.
1478
1479
148013.1.1 MT Bonding Mode Selection for Single Switch Topology
1481-----------------------------------------------------------
1482
1483 This configuration is the easiest to set up and to understand,
1484although you will have to decide which bonding mode best suits your
1485needs. The trade offs for each mode are detailed below:
1486
1487balance-rr: This mode is the only mode that will permit a single
1488 TCP/IP connection to stripe traffic across multiple
1489 interfaces. It is therefore the only mode that will allow a
1490 single TCP/IP stream to utilize more than one interface's
1491 worth of throughput. This comes at a cost, however: the
1492 striping often results in peer systems receiving packets out
1493 of order, causing TCP/IP's congestion control system to kick
1494 in, often by retransmitting segments.
1495
1496 It is possible to adjust TCP/IP's congestion limits by
1497 altering the net.ipv4.tcp_reordering sysctl parameter. The
1498 usual default value is 3, and the maximum useful value is 127.
1499 For a four interface balance-rr bond, expect that a single
1500 TCP/IP stream will utilize no more than approximately 2.3
1501 interface's worth of throughput, even after adjusting
1502 tcp_reordering.
1503
1504 Note that this out of order delivery occurs when both the
1505 sending and receiving systems are utilizing a multiple
1506 interface bond. Consider a configuration in which a
1507 balance-rr bond feeds into a single higher capacity network
1508 channel (e.g., multiple 100Mb/sec ethernets feeding a single
1509 gigabit ethernet via an etherchannel capable switch). In this
1510 configuration, traffic sent from the multiple 100Mb devices to
1511 a destination connected to the gigabit device will not see
1512 packets out of order. However, traffic sent from the gigabit
1513 device to the multiple 100Mb devices may or may not see
1514 traffic out of order, depending upon the balance policy of the
1515 switch. Many switches do not support any modes that stripe
1516 traffic (instead choosing a port based upon IP or MAC level
1517 addresses); for those devices, traffic flowing from the
1518 gigabit device to the many 100Mb devices will only utilize one
1519 interface.
1520
1521 If you are utilizing protocols other than TCP/IP, UDP for
1522 example, and your application can tolerate out of order
1523 delivery, then this mode can allow for single stream datagram
1524 performance that scales near linearly as interfaces are added
1525 to the bond.
1526
1527 This mode requires the switch to have the appropriate ports
1528 configured for "etherchannel" or "trunking."
1529
1530active-backup: There is not much advantage in this network topology to
1531 the active-backup mode, as the inactive backup devices are all
1532 connected to the same peer as the primary. In this case, a
1533 load balancing mode (with link monitoring) will provide the
1534 same level of network availability, but with increased
1535 available bandwidth. On the plus side, active-backup mode
1536 does not require any configuration of the switch, so it may
1537 have value if the hardware available does not support any of
1538 the load balance modes.
1539
1540balance-xor: This mode will limit traffic such that packets destined
1541 for specific peers will always be sent over the same
1542 interface. Since the destination is determined by the MAC
1543 addresses involved, this mode works best in a "local" network
1544 configuration (as described above), with destinations all on
1545 the same local network. This mode is likely to be suboptimal
1546 if all your traffic is passed through a single router (i.e., a
1547 "gatewayed" network configuration, as described above).
1548
1549 As with balance-rr, the switch ports need to be configured for
1550 "etherchannel" or "trunking."
1551
1552broadcast: Like active-backup, there is not much advantage to this
1553 mode in this type of network topology.
1554
1555802.3ad: This mode can be a good choice for this type of network
1556 topology. The 802.3ad mode is an IEEE standard, so all peers
1557 that implement 802.3ad should interoperate well. The 802.3ad
1558 protocol includes automatic configuration of the aggregates,
1559 so minimal manual configuration of the switch is needed
1560 (typically only to designate that some set of devices is
1561 available for 802.3ad). The 802.3ad standard also mandates
1562 that frames be delivered in order (within certain limits), so
1563 in general single connections will not see misordering of
1564 packets. The 802.3ad mode does have some drawbacks: the
1565 standard mandates that all devices in the aggregate operate at
1566 the same speed and duplex. Also, as with all bonding load
1567 balance modes other than balance-rr, no single connection will
1568 be able to utilize more than a single interface's worth of
1569 bandwidth.
1570
1571 Additionally, the linux bonding 802.3ad implementation
1572 distributes traffic by peer (using an XOR of MAC addresses),
1573 so in a "gatewayed" configuration, all outgoing traffic will
1574 generally use the same device. Incoming traffic may also end
1575 up on a single device, but that is dependent upon the
1576 balancing policy of the peer's 8023.ad implementation. In a
1577 "local" configuration, traffic will be distributed across the
1578 devices in the bond.
1579
1580 Finally, the 802.3ad mode mandates the use of the MII monitor,
1581 therefore, the ARP monitor is not available in this mode.
1582
1583balance-tlb: The balance-tlb mode balances outgoing traffic by peer.
1584 Since the balancing is done according to MAC address, in a
1585 "gatewayed" configuration (as described above), this mode will
1586 send all traffic across a single device. However, in a
1587 "local" network configuration, this mode balances multiple
1588 local network peers across devices in a vaguely intelligent
1589 manner (not a simple XOR as in balance-xor or 802.3ad mode),
1590 so that mathematically unlucky MAC addresses (i.e., ones that
1591 XOR to the same value) will not all "bunch up" on a single
1592 interface.
1593
1594 Unlike 802.3ad, interfaces may be of differing speeds, and no
1595 special switch configuration is required. On the down side,
1596 in this mode all incoming traffic arrives over a single
1597 interface, this mode requires certain ethtool support in the
1598 network device driver of the slave interfaces, and the ARP
1599 monitor is not available.
1600
1601balance-alb: This mode is everything that balance-tlb is, and more.
1602 It has all of the features (and restrictions) of balance-tlb,
1603 and will also balance incoming traffic from local network
1604 peers (as described in the Bonding Module Options section,
1605 above).
1606
1607 The only additional down side to this mode is that the network
1608 device driver must support changing the hardware address while
1609 the device is open.
1610
161113.1.2 MT Link Monitoring for Single Switch Topology
1612----------------------------------------------------
1613
1614 The choice of link monitoring may largely depend upon which
1615mode you choose to use. The more advanced load balancing modes do not
1616support the use of the ARP monitor, and are thus restricted to using
1617the MII monitor (which does not provide as high a level of end to end
1618assurance as the ARP monitor).
1619
162013.2 Maximum Throughput in a Multiple Switch Topology
1621-----------------------------------------------------
1622
1623 Multiple switches may be utilized to optimize for throughput
1624when they are configured in parallel as part of an isolated network
1625between two or more systems, for example:
1626
1627 +-----------+
1628 | Host A |
1629 +-+---+---+-+
1630 | | |
1631 +--------+ | +---------+
1632 | | |
1633 +------+---+ +-----+----+ +-----+----+
1634 | Switch A | | Switch B | | Switch C |
1635 +------+---+ +-----+----+ +-----+----+
1636 | | |
1637 +--------+ | +---------+
1638 | | |
1639 +-+---+---+-+
1640 | Host B |
1641 +-----------+
1642
1643 In this configuration, the switches are isolated from one
1644another. One reason to employ a topology such as this is for an
1645isolated network with many hosts (a cluster configured for high
1646performance, for example), using multiple smaller switches can be more
1647cost effective than a single larger switch, e.g., on a network with 24
1648hosts, three 24 port switches can be significantly less expensive than
1649a single 72 port switch.
1650
1651 If access beyond the network is required, an individual host
1652can be equipped with an additional network device connected to an
1653external network; this host then additionally acts as a gateway.
1654
165513.2.1 MT Bonding Mode Selection for Multiple Switch Topology
1656-------------------------------------------------------------
1657
1658 In actual practice, the bonding mode typically employed in
1659configurations of this type is balance-rr. Historically, in this
1660network configuration, the usual caveats about out of order packet
1661delivery are mitigated by the use of network adapters that do not do
1662any kind of packet coalescing (via the use of NAPI, or because the
1663device itself does not generate interrupts until some number of
1664packets has arrived). When employed in this fashion, the balance-rr
1665mode allows individual connections between two hosts to effectively
1666utilize greater than one interface's bandwidth.
1667
166813.2.2 MT Link Monitoring for Multiple Switch Topology
1669------------------------------------------------------
1670
1671 Again, in actual practice, the MII monitor is most often used
1672in this configuration, as performance is given preference over
1673availability. The ARP monitor will function in this topology, but its
1674advantages over the MII monitor are mitigated by the volume of probes
1675needed as the number of systems involved grows (remember that each
1676host in the network is configured with bonding).
1677
167814. Switch Behavior Issues
1679==========================
1680
168114.1 Link Establishment and Failover Delays
1682-------------------------------------------
1683
1684 Some switches exhibit undesirable behavior with regard to the
1685timing of link up and down reporting by the switch.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686
1687 First, when a link comes up, some switches may indicate that
1688the link is up (carrier available), but not pass traffic over the
1689interface for some period of time. This delay is typically due to
1690some type of autonegotiation or routing protocol, but may also occur
1691during switch initialization (e.g., during recovery after a switch
1692failure). If you find this to be a problem, specify an appropriate
1693value to the updelay bonding module option to delay the use of the
1694relevant interface(s).
1695
1696 Second, some switches may "bounce" the link state one or more
1697times while a link is changing state. This occurs most commonly while
1698the switch is initializing. Again, an appropriate updelay value may
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001699help.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700
1701 Note that when a bonding interface has no active links, the
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001702driver will immediately reuse the first link that goes up, even if the
1703updelay parameter has been specified (the updelay is ignored in this
1704case). If there are slave interfaces waiting for the updelay timeout
1705to expire, the interface that first went into that state will be
1706immediately reused. This reduces down time of the network if the
1707value of updelay has been overestimated, and since this occurs only in
1708cases with no connectivity, there is no additional penalty for
1709ignoring the updelay.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710
1711 In addition to the concerns about switch timings, if your
1712switches take a long time to go into backup mode, it may be desirable
1713to not activate a backup interface immediately after a link goes down.
1714Failover may be delayed via the downdelay bonding module option.
1715
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700171614.2 Duplicated Incoming Packets
1717--------------------------------
1718
1719 It is not uncommon to observe a short burst of duplicated
1720traffic when the bonding device is first used, or after it has been
1721idle for some period of time. This is most easily observed by issuing
1722a "ping" to some other host on the network, and noticing that the
1723output from ping flags duplicates (typically one per slave).
1724
1725 For example, on a bond in active-backup mode with five slaves
1726all connected to one switch, the output may appear as follows:
1727
1728# ping -n 10.0.4.2
1729PING 10.0.4.2 (10.0.4.2) from 10.0.3.10 : 56(84) bytes of data.
173064 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.7 ms
173164 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
173264 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
173364 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
173464 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
173564 bytes from 10.0.4.2: icmp_seq=2 ttl=64 time=0.216 ms
173664 bytes from 10.0.4.2: icmp_seq=3 ttl=64 time=0.267 ms
173764 bytes from 10.0.4.2: icmp_seq=4 ttl=64 time=0.222 ms
1738
1739 This is not due to an error in the bonding driver, rather, it
1740is a side effect of how many switches update their MAC forwarding
1741tables. Initially, the switch does not associate the MAC address in
1742the packet with a particular switch port, and so it may send the
1743traffic to all ports until its MAC forwarding table is updated. Since
1744the interfaces attached to the bond may occupy multiple ports on a
1745single switch, when the switch (temporarily) floods the traffic to all
1746ports, the bond device receives multiple copies of the same packet
1747(one per slave device).
1748
1749 The duplicated packet behavior is switch dependent, some
1750switches exhibit this, and some do not. On switches that display this
1751behavior, it can be induced by clearing the MAC forwarding table (on
1752most Cisco switches, the privileged command "clear mac address-table
1753dynamic" will accomplish this).
1754
175515. Hardware Specific Considerations
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756====================================
1757
1758 This section contains additional information for configuring
1759bonding on specific hardware platforms, or for interfacing bonding
1760with particular switches or other devices.
1761
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700176215.1 IBM BladeCenter
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763--------------------
1764
1765 This applies to the JS20 and similar systems.
1766
1767 On the JS20 blades, the bonding driver supports only
1768balance-rr, active-backup, balance-tlb and balance-alb modes. This is
1769largely due to the network topology inside the BladeCenter, detailed
1770below.
1771
1772JS20 network adapter information
1773--------------------------------
1774
1775 All JS20s come with two Broadcom Gigabit Ethernet ports
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001776integrated on the planar (that's "motherboard" in IBM-speak). In the
1777BladeCenter chassis, the eth0 port of all JS20 blades is hard wired to
1778I/O Module #1; similarly, all eth1 ports are wired to I/O Module #2.
1779An add-on Broadcom daughter card can be installed on a JS20 to provide
1780two more Gigabit Ethernet ports. These ports, eth2 and eth3, are
1781wired to I/O Modules 3 and 4, respectively.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782
1783 Each I/O Module may contain either a switch or a passthrough
1784module (which allows ports to be directly connected to an external
1785switch). Some bonding modes require a specific BladeCenter internal
1786network topology in order to function; these are detailed below.
1787
1788 Additional BladeCenter-specific networking information can be
1789found in two IBM Redbooks (www.ibm.com/redbooks):
1790
1791"IBM eServer BladeCenter Networking Options"
1792"IBM eServer BladeCenter Layer 2-7 Network Switching"
1793
1794BladeCenter networking configuration
1795------------------------------------
1796
1797 Because a BladeCenter can be configured in a very large number
1798of ways, this discussion will be confined to describing basic
1799configurations.
1800
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001801 Normally, Ethernet Switch Modules (ESMs) are used in I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802modules 1 and 2. In this configuration, the eth0 and eth1 ports of a
1803JS20 will be connected to different internal switches (in the
1804respective I/O modules).
1805
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001806 A passthrough module (OPM or CPM, optical or copper,
1807passthrough module) connects the I/O module directly to an external
1808switch. By using PMs in I/O module #1 and #2, the eth0 and eth1
1809interfaces of a JS20 can be redirected to the outside world and
1810connected to a common external switch.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001812 Depending upon the mix of ESMs and PMs, the network will
1813appear to bonding as either a single switch topology (all PMs) or as a
1814multiple switch topology (one or more ESMs, zero or more PMs). It is
1815also possible to connect ESMs together, resulting in a configuration
1816much like the example in "High Availability in a Multiple Switch
1817Topology," above.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001819Requirements for specific modes
1820-------------------------------
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001822 The balance-rr mode requires the use of passthrough modules
1823for devices in the bond, all connected to an common external switch.
1824That switch must be configured for "etherchannel" or "trunking" on the
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825appropriate ports, as is usual for balance-rr.
1826
1827 The balance-alb and balance-tlb modes will function with
1828either switch modules or passthrough modules (or a mix). The only
1829specific requirement for these modes is that all network interfaces
1830must be able to reach all destinations for traffic sent over the
1831bonding device (i.e., the network must converge at some point outside
1832the BladeCenter).
1833
1834 The active-backup mode has no additional requirements.
1835
1836Link monitoring issues
1837----------------------
1838
1839 When an Ethernet Switch Module is in place, only the ARP
1840monitor will reliably detect link loss to an external switch. This is
1841nothing unusual, but examination of the BladeCenter cabinet would
1842suggest that the "external" network ports are the ethernet ports for
1843the system, when it fact there is a switch between these "external"
1844ports and the devices on the JS20 system itself. The MII monitor is
1845only able to detect link failures between the ESM and the JS20 system.
1846
1847 When a passthrough module is in place, the MII monitor does
1848detect failures to the "external" port, which is then directly
1849connected to the JS20 system.
1850
1851Other concerns
1852--------------
1853
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001854 The Serial Over LAN (SoL) link is established over the primary
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855ethernet (eth0) only, therefore, any loss of link to eth0 will result
1856in losing your SoL connection. It will not fail over with other
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001857network traffic, as the SoL system is beyond the control of the
1858bonding driver.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001859
1860 It may be desirable to disable spanning tree on the switch
1861(either the internal Ethernet Switch Module, or an external switch) to
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001862avoid fail-over delay issues when using bonding.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863
1864
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700186516. Frequently Asked Questions
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866==============================
1867
18681. Is it SMP safe?
1869
1870 Yes. The old 2.0.xx channel bonding patch was not SMP safe.
1871The new driver was designed to be SMP safe from the start.
1872
18732. What type of cards will work with it?
1874
1875 Any Ethernet type cards (you can even mix cards - a Intel
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001876EtherExpress PRO/100 and a 3com 3c905b, for example). For most modes,
1877devices need not be of the same speed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878
18793. How many bonding devices can I have?
1880
1881 There is no limit.
1882
18834. How many slaves can a bonding device have?
1884
1885 This is limited only by the number of network interfaces Linux
1886supports and/or the number of network cards you can place in your
1887system.
1888
18895. What happens when a slave link dies?
1890
1891 If link monitoring is enabled, then the failing device will be
1892disabled. The active-backup mode will fail over to a backup link, and
1893other modes will ignore the failed link. The link will continue to be
1894monitored, and should it recover, it will rejoin the bond (in whatever
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001895manner is appropriate for the mode). See the sections on High
1896Availability and the documentation for each mode for additional
1897information.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898
1899 Link monitoring can be enabled via either the miimon or
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001900arp_interval parameters (described in the module parameters section,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901above). In general, miimon monitors the carrier state as sensed by
1902the underlying network device, and the arp monitor (arp_interval)
1903monitors connectivity to another host on the local network.
1904
1905 If no link monitoring is configured, the bonding driver will
1906be unable to detect link failures, and will assume that all links are
1907always available. This will likely result in lost packets, and a
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001908resulting degradation of performance. The precise performance loss
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909depends upon the bonding mode and network configuration.
1910
19116. Can bonding be used for High Availability?
1912
1913 Yes. See the section on High Availability for details.
1914
19157. Which switches/systems does it work with?
1916
1917 The full answer to this depends upon the desired mode.
1918
1919 In the basic balance modes (balance-rr and balance-xor), it
1920works with any system that supports etherchannel (also called
1921trunking). Most managed switches currently available have such
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001922support, and many unmanaged switches as well.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923
1924 The advanced balance modes (balance-tlb and balance-alb) do
1925not have special switch requirements, but do need device drivers that
1926support specific features (described in the appropriate section under
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001927module parameters, above).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928
1929 In 802.3ad mode, it works with with systems that support IEEE
1930802.3ad Dynamic Link Aggregation. Most managed and many unmanaged
1931switches currently available support 802.3ad.
1932
1933 The active-backup mode should work with any Layer-II switch.
1934
19358. Where does a bonding device get its MAC address from?
1936
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001937 If not explicitly configured (with ifconfig or ip link), the
1938MAC address of the bonding device is taken from its first slave
1939device. This MAC address is then passed to all following slaves and
1940remains persistent (even if the the first slave is removed) until the
1941bonding device is brought down or reconfigured.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942
1943 If you wish to change the MAC address, you can set it with
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001944ifconfig or ip link:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945
1946# ifconfig bond0 hw ether 00:11:22:33:44:55
1947
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001948# ip link set bond0 address 66:77:88:99:aa:bb
1949
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950 The MAC address can be also changed by bringing down/up the
1951device and then changing its slaves (or their order):
1952
1953# ifconfig bond0 down ; modprobe -r bonding
1954# ifconfig bond0 .... up
1955# ifenslave bond0 eth...
1956
1957 This method will automatically take the address from the next
1958slave that is added.
1959
1960 To restore your slaves' MAC addresses, you need to detach them
1961from the bond (`ifenslave -d bond0 eth0'). The bonding driver will
1962then restore the MAC addresses that the slaves had before they were
1963enslaved.
1964
Jay Vosburgh00354cf2005-07-21 12:18:02 -0700196516. Resources and Links
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966=======================
1967
1968The latest version of the bonding driver can be found in the latest
1969version of the linux kernel, found on http://kernel.org
1970
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001971The latest version of this document can be found in either the latest
1972kernel source (named Documentation/networking/bonding.txt), or on the
1973bonding sourceforge site:
1974
1975http://www.sourceforge.net/projects/bonding
1976
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977Discussions regarding the bonding driver take place primarily on the
1978bonding-devel mailing list, hosted at sourceforge.net. If you have
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001979questions or problems, post them to the list. The list address is:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980
1981bonding-devel@lists.sourceforge.net
1982
Jay Vosburgh00354cf2005-07-21 12:18:02 -07001983 The administrative interface (to subscribe or unsubscribe) can
1984be found at:
1985
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986https://lists.sourceforge.net/lists/listinfo/bonding-devel
1987
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988Donald Becker's Ethernet Drivers and diag programs may be found at :
1989 - http://www.scyld.com/network/
1990
1991You will also find a lot of information regarding Ethernet, NWay, MII,
1992etc. at www.scyld.com.
1993
1994-- END --