2 Linux Ethernet Bonding Driver HOWTO
4 Latest update: 27 April 2011
6 Initial release : Thomas Davis <tadavis at lbl.gov>
7 Corrections, 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>
14 Reorganized and updated Feb 2005 by Jay Vosburgh
15 Added Sysfs information: 2006/04/24
16 - Mitch Williams <mitch.a.williams at intel.com>
21 The Linux bonding driver provides a method for aggregating
22 multiple network interfaces into a single logical "bonded" interface.
23 The behavior of the bonded interfaces depends upon the mode; generally
24 speaking, modes provide either hot standby or load balancing services.
25 Additionally, link integrity monitoring may be performed.
27 The bonding driver originally came from Donald Becker's
28 beowulf patches for kernel 2.0. It has changed quite a bit since, and
29 the original tools from extreme-linux and beowulf sites will not work
30 with this version of the driver.
32 For new versions of the driver, updated userspace tools, and
33 who to ask for help, please follow the links at the end of this file.
38 1. Bonding Driver Installation
40 2. Bonding Driver Options
42 3. Configuring Bonding Devices
43 3.1 Configuration with Sysconfig Support
44 3.1.1 Using DHCP with Sysconfig
45 3.1.2 Configuring Multiple Bonds with Sysconfig
46 3.2 Configuration with Initscripts Support
47 3.2.1 Using DHCP with Initscripts
48 3.2.2 Configuring Multiple Bonds with Initscripts
49 3.3 Configuring Bonding Manually with Ifenslave
50 3.3.1 Configuring Multiple Bonds Manually
51 3.4 Configuring Bonding Manually via Sysfs
52 3.5 Configuration with Interfaces Support
53 3.6 Overriding Configuration for Special Cases
55 4. Querying Bonding Configuration
56 4.1 Bonding Configuration
57 4.2 Network Configuration
59 5. Switch Configuration
61 6. 802.1q VLAN Support
64 7.1 ARP Monitor Operation
65 7.2 Configuring Multiple ARP Targets
66 7.3 MII Monitor Operation
68 8. Potential Trouble Sources
69 8.1 Adventures in Routing
70 8.2 Ethernet Device Renaming
71 8.3 Painfully Slow Or No Failed Link Detection By Miimon
77 11. Configuring Bonding for High Availability
78 11.1 High Availability in a Single Switch Topology
79 11.2 High Availability in a Multiple Switch Topology
80 11.2.1 HA Bonding Mode Selection for Multiple Switch Topology
81 11.2.2 HA Link Monitoring for Multiple Switch Topology
83 12. Configuring Bonding for Maximum Throughput
84 12.1 Maximum Throughput in a Single Switch Topology
85 12.1.1 MT Bonding Mode Selection for Single Switch Topology
86 12.1.2 MT Link Monitoring for Single Switch Topology
87 12.2 Maximum Throughput in a Multiple Switch Topology
88 12.2.1 MT Bonding Mode Selection for Multiple Switch Topology
89 12.2.2 MT Link Monitoring for Multiple Switch Topology
91 13. Switch Behavior Issues
92 13.1 Link Establishment and Failover Delays
93 13.2 Duplicated Incoming Packets
95 14. Hardware Specific Considerations
98 15. Frequently Asked Questions
100 16. Resources and Links
103 1. Bonding Driver Installation
104 ==============================
106 Most popular distro kernels ship with the bonding driver
107 already available as a module and the ifenslave user level control
108 program installed and ready for use. If your distro does not, or you
109 have need to compile bonding from source (e.g., configuring and
110 installing a mainline kernel from kernel.org), you'll need to perform
113 1.1 Configure and build the kernel with bonding
114 -----------------------------------------------
116 The current version of the bonding driver is available in the
117 drivers/net/bonding subdirectory of the most recent kernel source
118 (which is available on http://kernel.org). Most users "rolling their
119 own" will want to use the most recent kernel from kernel.org.
121 Configure kernel with "make menuconfig" (or "make xconfig" or
122 "make config"), then select "Bonding driver support" in the "Network
123 device support" section. It is recommended that you configure the
124 driver as module since it is currently the only way to pass parameters
125 to the driver or configure more than one bonding device.
127 Build and install the new kernel and modules, then continue
128 below to install ifenslave.
130 1.2 Install ifenslave Control Utility
131 -------------------------------------
133 The ifenslave user level control program is included in the
134 kernel source tree, in the file Documentation/networking/ifenslave.c.
135 It is generally recommended that you use the ifenslave that
136 corresponds to the kernel that you are using (either from the same
137 source tree or supplied with the distro), however, ifenslave
138 executables from older kernels should function (but features newer
139 than the ifenslave release are not supported). Running an ifenslave
140 that is newer than the kernel is not supported, and may or may not
143 To install ifenslave, do the following:
145 # gcc -Wall -O -I/usr/src/linux/include ifenslave.c -o ifenslave
146 # cp ifenslave /sbin/ifenslave
148 If your kernel source is not in "/usr/src/linux," then replace
149 "/usr/src/linux/include" in the above with the location of your kernel
150 source include directory.
152 You may wish to back up any existing /sbin/ifenslave, or, for
153 testing or informal use, tag the ifenslave to the kernel version
154 (e.g., name the ifenslave executable /sbin/ifenslave-2.6.10).
158 If you omit the "-I" or specify an incorrect directory, you
159 may end up with an ifenslave that is incompatible with the kernel
160 you're trying to build it for. Some distros (e.g., Red Hat from 7.1
161 onwards) do not have /usr/include/linux symbolically linked to the
162 default kernel source include directory.
164 SECOND IMPORTANT NOTE:
165 If you plan to configure bonding using sysfs or using the
166 /etc/network/interfaces file, you do not need to use ifenslave.
168 2. Bonding Driver Options
169 =========================
171 Options for the bonding driver are supplied as parameters to the
172 bonding module at load time, or are specified via sysfs.
174 Module options may be given as command line arguments to the
175 insmod or modprobe command, but are usually specified in either the
176 /etc/modules.conf or /etc/modprobe.conf configuration file, or in a
177 distro-specific configuration file (some of which are detailed in the next
180 Details on bonding support for sysfs is provided in the
181 "Configuring Bonding Manually via Sysfs" section, below.
183 The available bonding driver parameters are listed below. If a
184 parameter is not specified the default value is used. When initially
185 configuring a bond, it is recommended "tail -f /var/log/messages" be
186 run in a separate window to watch for bonding driver error messages.
188 It is critical that either the miimon or arp_interval and
189 arp_ip_target parameters be specified, otherwise serious network
190 degradation will occur during link failures. Very few devices do not
191 support at least miimon, so there is really no reason not to use it.
193 Options with textual values will accept either the text name
194 or, for backwards compatibility, the option value. E.g.,
195 "mode=802.3ad" and "mode=4" set the same mode.
197 The parameters are as follows:
201 Specifies the new active slave for modes that support it
202 (active-backup, balance-alb and balance-tlb). Possible values
203 are the name of any currently enslaved interface, or an empty
204 string. If a name is given, the slave and its link must be up in order
205 to be selected as the new active slave. If an empty string is
206 specified, the current active slave is cleared, and a new active
207 slave is selected automatically.
209 Note that this is only available through the sysfs interface. No module
210 parameter by this name exists.
212 The normal value of this option is the name of the currently
213 active slave, or the empty string if there is no active slave or
214 the current mode does not use an active slave.
218 Specifies the 802.3ad aggregation selection logic to use. The
219 possible values and their effects are:
223 The active aggregator is chosen by largest aggregate
226 Reselection of the active aggregator occurs only when all
227 slaves of the active aggregator are down or the active
228 aggregator has no slaves.
230 This is the default value.
234 The active aggregator is chosen by largest aggregate
235 bandwidth. Reselection occurs if:
237 - A slave is added to or removed from the bond
239 - Any slave's link state changes
241 - Any slave's 802.3ad association state changes
243 - The bond's administrative state changes to up
247 The active aggregator is chosen by the largest number of
248 ports (slaves). Reselection occurs as described under the
249 "bandwidth" setting, above.
251 The bandwidth and count selection policies permit failover of
252 802.3ad aggregations when partial failure of the active aggregator
253 occurs. This keeps the aggregator with the highest availability
254 (either in bandwidth or in number of ports) active at all times.
256 This option was added in bonding version 3.4.0.
260 Specifies that duplicate frames (received on inactive ports) should be
261 dropped (0) or delivered (1).
263 Normally, bonding will drop duplicate frames (received on inactive
264 ports), which is desirable for most users. But there are some times
265 it is nice to allow duplicate frames to be delivered.
267 The default value is 0 (drop duplicate frames received on inactive
272 Specifies the ARP link monitoring frequency in milliseconds.
274 The ARP monitor works by periodically checking the slave
275 devices to determine whether they have sent or received
276 traffic recently (the precise criteria depends upon the
277 bonding mode, and the state of the slave). Regular traffic is
278 generated via ARP probes issued for the addresses specified by
279 the arp_ip_target option.
281 This behavior can be modified by the arp_validate option,
284 If ARP monitoring is used in an etherchannel compatible mode
285 (modes 0 and 2), the switch should be configured in a mode
286 that evenly distributes packets across all links. If the
287 switch is configured to distribute the packets in an XOR
288 fashion, all replies from the ARP targets will be received on
289 the same link which could cause the other team members to
290 fail. ARP monitoring should not be used in conjunction with
291 miimon. A value of 0 disables ARP monitoring. The default
296 Specifies the IP addresses to use as ARP monitoring peers when
297 arp_interval is > 0. These are the targets of the ARP request
298 sent to determine the health of the link to the targets.
299 Specify these values in ddd.ddd.ddd.ddd format. Multiple IP
300 addresses must be separated by a comma. At least one IP
301 address must be given for ARP monitoring to function. The
302 maximum number of targets that can be specified is 16. The
303 default value is no IP addresses.
307 Specifies whether or not ARP probes and replies should be
308 validated in the active-backup mode. This causes the ARP
309 monitor to examine the incoming ARP requests and replies, and
310 only consider a slave to be up if it is receiving the
311 appropriate ARP traffic.
317 No validation is performed. This is the default.
321 Validation is performed only for the active slave.
325 Validation is performed only for backup slaves.
329 Validation is performed for all slaves.
331 For the active slave, the validation checks ARP replies to
332 confirm that they were generated by an arp_ip_target. Since
333 backup slaves do not typically receive these replies, the
334 validation performed for backup slaves is on the ARP request
335 sent out via the active slave. It is possible that some
336 switch or network configurations may result in situations
337 wherein the backup slaves do not receive the ARP requests; in
338 such a situation, validation of backup slaves must be
341 This option is useful in network configurations in which
342 multiple bonding hosts are concurrently issuing ARPs to one or
343 more targets beyond a common switch. Should the link between
344 the switch and target fail (but not the switch itself), the
345 probe traffic generated by the multiple bonding instances will
346 fool the standard ARP monitor into considering the links as
347 still up. Use of the arp_validate option can resolve this, as
348 the ARP monitor will only consider ARP requests and replies
349 associated with its own instance of bonding.
351 This option was added in bonding version 3.1.0.
355 Specifies the time, in milliseconds, to wait before disabling
356 a slave after a link failure has been detected. This option
357 is only valid for the miimon link monitor. The downdelay
358 value should be a multiple of the miimon value; if not, it
359 will be rounded down to the nearest multiple. The default
364 Specifies whether active-backup mode should set all slaves to
365 the same MAC address at enslavement (the traditional
366 behavior), or, when enabled, perform special handling of the
367 bond's MAC address in accordance with the selected policy.
373 This setting disables fail_over_mac, and causes
374 bonding to set all slaves of an active-backup bond to
375 the same MAC address at enslavement time. This is the
380 The "active" fail_over_mac policy indicates that the
381 MAC address of the bond should always be the MAC
382 address of the currently active slave. The MAC
383 address of the slaves is not changed; instead, the MAC
384 address of the bond changes during a failover.
386 This policy is useful for devices that cannot ever
387 alter their MAC address, or for devices that refuse
388 incoming broadcasts with their own source MAC (which
389 interferes with the ARP monitor).
391 The down side of this policy is that every device on
392 the network must be updated via gratuitous ARP,
393 vs. just updating a switch or set of switches (which
394 often takes place for any traffic, not just ARP
395 traffic, if the switch snoops incoming traffic to
396 update its tables) for the traditional method. If the
397 gratuitous ARP is lost, communication may be
400 When this policy is used in conjunction with the mii
401 monitor, devices which assert link up prior to being
402 able to actually transmit and receive are particularly
403 susceptible to loss of the gratuitous ARP, and an
404 appropriate updelay setting may be required.
408 The "follow" fail_over_mac policy causes the MAC
409 address of the bond to be selected normally (normally
410 the MAC address of the first slave added to the bond).
411 However, the second and subsequent slaves are not set
412 to this MAC address while they are in a backup role; a
413 slave is programmed with the bond's MAC address at
414 failover time (and the formerly active slave receives
415 the newly active slave's MAC address).
417 This policy is useful for multiport devices that
418 either become confused or incur a performance penalty
419 when multiple ports are programmed with the same MAC
423 The default policy is none, unless the first slave cannot
424 change its MAC address, in which case the active policy is
427 This option may be modified via sysfs only when no slaves are
430 This option was added in bonding version 3.2.0. The "follow"
431 policy was added in bonding version 3.3.0.
435 Option specifying the rate in which we'll ask our link partner
436 to transmit LACPDU packets in 802.3ad mode. Possible values
440 Request partner to transmit LACPDUs every 30 seconds
443 Request partner to transmit LACPDUs every 1 second
449 Specifies the number of bonding devices to create for this
450 instance of the bonding driver. E.g., if max_bonds is 3, and
451 the bonding driver is not already loaded, then bond0, bond1
452 and bond2 will be created. The default value is 1. Specifying
453 a value of 0 will load bonding, but will not create any devices.
457 Specifies the MII link monitoring frequency in milliseconds.
458 This determines how often the link state of each slave is
459 inspected for link failures. A value of zero disables MII
460 link monitoring. A value of 100 is a good starting point.
461 The use_carrier option, below, affects how the link state is
462 determined. See the High Availability section for additional
463 information. The default value is 0.
467 Specifies the minimum number of links that must be active before
468 asserting carrier. It is similar to the Cisco EtherChannel min-links
469 feature. This allows setting the minimum number of member ports that
470 must be up (link-up state) before marking the bond device as up
471 (carrier on). This is useful for situations where higher level services
472 such as clustering want to ensure a minimum number of low bandwidth
473 links are active before switchover. This option only affect 802.3ad
476 The default value is 0. This will cause carrier to be asserted (for
477 802.3ad mode) whenever there is an active aggregator, regardless of the
478 number of available links in that aggregator. Note that, because an
479 aggregator cannot be active without at least one available link,
480 setting this option to 0 or to 1 has the exact same effect.
484 Specifies one of the bonding policies. The default is
485 balance-rr (round robin). Possible values are:
489 Round-robin policy: Transmit packets in sequential
490 order from the first available slave through the
491 last. This mode provides load balancing and fault
496 Active-backup policy: Only one slave in the bond is
497 active. A different slave becomes active if, and only
498 if, the active slave fails. The bond's MAC address is
499 externally visible on only one port (network adapter)
500 to avoid confusing the switch.
502 In bonding version 2.6.2 or later, when a failover
503 occurs in active-backup mode, bonding will issue one
504 or more gratuitous ARPs on the newly active slave.
505 One gratuitous ARP is issued for the bonding master
506 interface and each VLAN interfaces configured above
507 it, provided that the interface has at least one IP
508 address configured. Gratuitous ARPs issued for VLAN
509 interfaces are tagged with the appropriate VLAN id.
511 This mode provides fault tolerance. The primary
512 option, documented below, affects the behavior of this
517 XOR policy: Transmit based on the selected transmit
518 hash policy. The default policy is a simple [(source
519 MAC address XOR'd with destination MAC address) modulo
520 slave count]. Alternate transmit policies may be
521 selected via the xmit_hash_policy option, described
524 This mode provides load balancing and fault tolerance.
528 Broadcast policy: transmits everything on all slave
529 interfaces. This mode provides fault tolerance.
533 IEEE 802.3ad Dynamic link aggregation. Creates
534 aggregation groups that share the same speed and
535 duplex settings. Utilizes all slaves in the active
536 aggregator according to the 802.3ad specification.
538 Slave selection for outgoing traffic is done according
539 to the transmit hash policy, which may be changed from
540 the default simple XOR policy via the xmit_hash_policy
541 option, documented below. Note that not all transmit
542 policies may be 802.3ad compliant, particularly in
543 regards to the packet mis-ordering requirements of
544 section 43.2.4 of the 802.3ad standard. Differing
545 peer implementations will have varying tolerances for
550 1. Ethtool support in the base drivers for retrieving
551 the speed and duplex of each slave.
553 2. A switch that supports IEEE 802.3ad Dynamic link
556 Most switches will require some type of configuration
557 to enable 802.3ad mode.
561 Adaptive transmit load balancing: channel bonding that
562 does not require any special switch support. The
563 outgoing traffic is distributed according to the
564 current load (computed relative to the speed) on each
565 slave. Incoming traffic is received by the current
566 slave. If the receiving slave fails, another slave
567 takes over the MAC address of the failed receiving
572 Ethtool support in the base drivers for retrieving the
577 Adaptive load balancing: includes balance-tlb plus
578 receive load balancing (rlb) for IPV4 traffic, and
579 does not require any special switch support. The
580 receive load balancing is achieved by ARP negotiation.
581 The bonding driver intercepts the ARP Replies sent by
582 the local system on their way out and overwrites the
583 source hardware address with the unique hardware
584 address of one of the slaves in the bond such that
585 different peers use different hardware addresses for
588 Receive traffic from connections created by the server
589 is also balanced. When the local system sends an ARP
590 Request the bonding driver copies and saves the peer's
591 IP information from the ARP packet. When the ARP
592 Reply arrives from the peer, its hardware address is
593 retrieved and the bonding driver initiates an ARP
594 reply to this peer assigning it to one of the slaves
595 in the bond. A problematic outcome of using ARP
596 negotiation for balancing is that each time that an
597 ARP request is broadcast it uses the hardware address
598 of the bond. Hence, peers learn the hardware address
599 of the bond and the balancing of receive traffic
600 collapses to the current slave. This is handled by
601 sending updates (ARP Replies) to all the peers with
602 their individually assigned hardware address such that
603 the traffic is redistributed. Receive traffic is also
604 redistributed when a new slave is added to the bond
605 and when an inactive slave is re-activated. The
606 receive load is distributed sequentially (round robin)
607 among the group of highest speed slaves in the bond.
609 When a link is reconnected or a new slave joins the
610 bond the receive traffic is redistributed among all
611 active slaves in the bond by initiating ARP Replies
612 with the selected MAC address to each of the
613 clients. The updelay parameter (detailed below) must
614 be set to a value equal or greater than the switch's
615 forwarding delay so that the ARP Replies sent to the
616 peers will not be blocked by the switch.
620 1. Ethtool support in the base drivers for retrieving
621 the speed of each slave.
623 2. Base driver support for setting the hardware
624 address of a device while it is open. This is
625 required so that there will always be one slave in the
626 team using the bond hardware address (the
627 curr_active_slave) while having a unique hardware
628 address for each slave in the bond. If the
629 curr_active_slave fails its hardware address is
630 swapped with the new curr_active_slave that was
636 Specify the number of peer notifications (gratuitous ARPs and
637 unsolicited IPv6 Neighbor Advertisements) to be issued after a
638 failover event. As soon as the link is up on the new slave
639 (possibly immediately) a peer notification is sent on the
640 bonding device and each VLAN sub-device. This is repeated at
641 each link monitor interval (arp_interval or miimon, whichever
642 is active) if the number is greater than 1.
644 The valid range is 0 - 255; the default value is 1. These options
645 affect only the active-backup mode. These options were added for
646 bonding versions 3.3.0 and 3.4.0 respectively.
648 From Linux 3.0 and bonding version 3.7.1, these notifications
649 are generated by the ipv4 and ipv6 code and the numbers of
650 repetitions cannot be set independently.
654 A string (eth0, eth2, etc) specifying which slave is the
655 primary device. The specified device will always be the
656 active slave while it is available. Only when the primary is
657 off-line will alternate devices be used. This is useful when
658 one slave is preferred over another, e.g., when one slave has
659 higher throughput than another.
661 The primary option is only valid for active-backup mode.
665 Specifies the reselection policy for the primary slave. This
666 affects how the primary slave is chosen to become the active slave
667 when failure of the active slave or recovery of the primary slave
668 occurs. This option is designed to prevent flip-flopping between
669 the primary slave and other slaves. Possible values are:
671 always or 0 (default)
673 The primary slave becomes the active slave whenever it
678 The primary slave becomes the active slave when it comes
679 back up, if the speed and duplex of the primary slave is
680 better than the speed and duplex of the current active
685 The primary slave becomes the active slave only if the
686 current active slave fails and the primary slave is up.
688 The primary_reselect setting is ignored in two cases:
690 If no slaves are active, the first slave to recover is
691 made the active slave.
693 When initially enslaved, the primary slave is always made
696 Changing the primary_reselect policy via sysfs will cause an
697 immediate selection of the best active slave according to the new
698 policy. This may or may not result in a change of the active
699 slave, depending upon the circumstances.
701 This option was added for bonding version 3.6.0.
705 Specifies the time, in milliseconds, to wait before enabling a
706 slave after a link recovery has been detected. This option is
707 only valid for the miimon link monitor. The updelay value
708 should be a multiple of the miimon value; if not, it will be
709 rounded down to the nearest multiple. The default value is 0.
713 Specifies whether or not miimon should use MII or ETHTOOL
714 ioctls vs. netif_carrier_ok() to determine the link
715 status. The MII or ETHTOOL ioctls are less efficient and
716 utilize a deprecated calling sequence within the kernel. The
717 netif_carrier_ok() relies on the device driver to maintain its
718 state with netif_carrier_on/off; at this writing, most, but
719 not all, device drivers support this facility.
721 If bonding insists that the link is up when it should not be,
722 it may be that your network device driver does not support
723 netif_carrier_on/off. The default state for netif_carrier is
724 "carrier on," so if a driver does not support netif_carrier,
725 it will appear as if the link is always up. In this case,
726 setting use_carrier to 0 will cause bonding to revert to the
727 MII / ETHTOOL ioctl method to determine the link state.
729 A value of 1 enables the use of netif_carrier_ok(), a value of
730 0 will use the deprecated MII / ETHTOOL ioctls. The default
735 Selects the transmit hash policy to use for slave selection in
736 balance-xor and 802.3ad modes. Possible values are:
740 Uses XOR of hardware MAC addresses to generate the
743 (source MAC XOR destination MAC) modulo slave count
745 This algorithm will place all traffic to a particular
746 network peer on the same slave.
748 This algorithm is 802.3ad compliant.
752 This policy uses a combination of layer2 and layer3
753 protocol information to generate the hash.
755 Uses XOR of hardware MAC addresses and IP addresses to
756 generate the hash. The formula is
758 (((source IP XOR dest IP) AND 0xffff) XOR
759 ( source MAC XOR destination MAC ))
762 This algorithm will place all traffic to a particular
763 network peer on the same slave. For non-IP traffic,
764 the formula is the same as for the layer2 transmit
767 This policy is intended to provide a more balanced
768 distribution of traffic than layer2 alone, especially
769 in environments where a layer3 gateway device is
770 required to reach most destinations.
772 This algorithm is 802.3ad compliant.
776 This policy uses upper layer protocol information,
777 when available, to generate the hash. This allows for
778 traffic to a particular network peer to span multiple
779 slaves, although a single connection will not span
782 The formula for unfragmented TCP and UDP packets is
784 ((source port XOR dest port) XOR
785 ((source IP XOR dest IP) AND 0xffff)
788 For fragmented TCP or UDP packets and all other IP
789 protocol traffic, the source and destination port
790 information is omitted. For non-IP traffic, the
791 formula is the same as for the layer2 transmit hash
794 This policy is intended to mimic the behavior of
795 certain switches, notably Cisco switches with PFC2 as
796 well as some Foundry and IBM products.
798 This algorithm is not fully 802.3ad compliant. A
799 single TCP or UDP conversation containing both
800 fragmented and unfragmented packets will see packets
801 striped across two interfaces. This may result in out
802 of order delivery. Most traffic types will not meet
803 this criteria, as TCP rarely fragments traffic, and
804 most UDP traffic is not involved in extended
805 conversations. Other implementations of 802.3ad may
806 or may not tolerate this noncompliance.
808 The default value is layer2. This option was added in bonding
809 version 2.6.3. In earlier versions of bonding, this parameter
810 does not exist, and the layer2 policy is the only policy. The
811 layer2+3 value was added for bonding version 3.2.2.
815 Specifies the number of IGMP membership reports to be issued after
816 a failover event. One membership report is issued immediately after
817 the failover, subsequent packets are sent in each 200ms interval.
819 The valid range is 0 - 255; the default value is 1. A value of 0
820 prevents the IGMP membership report from being issued in response
821 to the failover event.
823 This option is useful for bonding modes balance-rr (0), active-backup
824 (1), balance-tlb (5) and balance-alb (6), in which a failover can
825 switch the IGMP traffic from one slave to another. Therefore a fresh
826 IGMP report must be issued to cause the switch to forward the incoming
827 IGMP traffic over the newly selected slave.
829 This option was added for bonding version 3.7.0.
831 3. Configuring Bonding Devices
832 ==============================
834 You can configure bonding using either your distro's network
835 initialization scripts, or manually using either ifenslave or the
836 sysfs interface. Distros generally use one of three packages for the
837 network initialization scripts: initscripts, sysconfig or interfaces.
838 Recent versions of these packages have support for bonding, while older
841 We will first describe the options for configuring bonding for
842 distros using versions of initscripts, sysconfig and interfaces with full
843 or partial support for bonding, then provide information on enabling
844 bonding without support from the network initialization scripts (i.e.,
845 older versions of initscripts or sysconfig).
847 If you're unsure whether your distro uses sysconfig,
848 initscripts or interfaces, or don't know if it's new enough, have no fear.
849 Determining this is fairly straightforward.
851 First, look for a file called interfaces in /etc/network directory.
852 If this file is present in your system, then your system use interfaces. See
853 Configuration with Interfaces Support.
855 Else, issue the command:
859 It will respond with a line of text starting with either
860 "initscripts" or "sysconfig," followed by some numbers. This is the
861 package that provides your network initialization scripts.
863 Next, to determine if your installation supports bonding,
866 $ grep ifenslave /sbin/ifup
868 If this returns any matches, then your initscripts or
869 sysconfig has support for bonding.
871 3.1 Configuration with Sysconfig Support
872 ----------------------------------------
874 This section applies to distros using a version of sysconfig
875 with bonding support, for example, SuSE Linux Enterprise Server 9.
877 SuSE SLES 9's networking configuration system does support
878 bonding, however, at this writing, the YaST system configuration
879 front end does not provide any means to work with bonding devices.
880 Bonding devices can be managed by hand, however, as follows.
882 First, if they have not already been configured, configure the
883 slave devices. On SLES 9, this is most easily done by running the
884 yast2 sysconfig configuration utility. The goal is for to create an
885 ifcfg-id file for each slave device. The simplest way to accomplish
886 this is to configure the devices for DHCP (this is only to get the
887 file ifcfg-id file created; see below for some issues with DHCP). The
888 name of the configuration file for each device will be of the form:
890 ifcfg-id-xx:xx:xx:xx:xx:xx
892 Where the "xx" portion will be replaced with the digits from
893 the device's permanent MAC address.
895 Once the set of ifcfg-id-xx:xx:xx:xx:xx:xx files has been
896 created, it is necessary to edit the configuration files for the slave
897 devices (the MAC addresses correspond to those of the slave devices).
898 Before editing, the file will contain multiple lines, and will look
904 UNIQUE='XNzu.WeZGOGF+4wE'
905 _nm_name='bus-pci-0001:61:01.0'
907 Change the BOOTPROTO and STARTMODE lines to the following:
912 Do not alter the UNIQUE or _nm_name lines. Remove any other
913 lines (USERCTL, etc).
915 Once the ifcfg-id-xx:xx:xx:xx:xx:xx files have been modified,
916 it's time to create the configuration file for the bonding device
917 itself. This file is named ifcfg-bondX, where X is the number of the
918 bonding device to create, starting at 0. The first such file is
919 ifcfg-bond0, the second is ifcfg-bond1, and so on. The sysconfig
920 network configuration system will correctly start multiple instances
923 The contents of the ifcfg-bondX file is as follows:
926 BROADCAST="10.0.2.255"
928 NETMASK="255.255.0.0"
933 BONDING_MODULE_OPTS="mode=active-backup miimon=100"
934 BONDING_SLAVE0="eth0"
935 BONDING_SLAVE1="bus-pci-0000:06:08.1"
937 Replace the sample BROADCAST, IPADDR, NETMASK and NETWORK
938 values with the appropriate values for your network.
940 The STARTMODE specifies when the device is brought online.
941 The possible values are:
943 onboot: The device is started at boot time. If you're not
944 sure, this is probably what you want.
946 manual: The device is started only when ifup is called
947 manually. Bonding devices may be configured this
948 way if you do not wish them to start automatically
949 at boot for some reason.
951 hotplug: The device is started by a hotplug event. This is not
952 a valid choice for a bonding device.
954 off or ignore: The device configuration is ignored.
956 The line BONDING_MASTER='yes' indicates that the device is a
957 bonding master device. The only useful value is "yes."
959 The contents of BONDING_MODULE_OPTS are supplied to the
960 instance of the bonding module for this device. Specify the options
961 for the bonding mode, link monitoring, and so on here. Do not include
962 the max_bonds bonding parameter; this will confuse the configuration
963 system if you have multiple bonding devices.
965 Finally, supply one BONDING_SLAVEn="slave device" for each
966 slave. where "n" is an increasing value, one for each slave. The
967 "slave device" is either an interface name, e.g., "eth0", or a device
968 specifier for the network device. The interface name is easier to
969 find, but the ethN names are subject to change at boot time if, e.g.,
970 a device early in the sequence has failed. The device specifiers
971 (bus-pci-0000:06:08.1 in the example above) specify the physical
972 network device, and will not change unless the device's bus location
973 changes (for example, it is moved from one PCI slot to another). The
974 example above uses one of each type for demonstration purposes; most
975 configurations will choose one or the other for all slave devices.
977 When all configuration files have been modified or created,
978 networking must be restarted for the configuration changes to take
979 effect. This can be accomplished via the following:
981 # /etc/init.d/network restart
983 Note that the network control script (/sbin/ifdown) will
984 remove the bonding module as part of the network shutdown processing,
985 so it is not necessary to remove the module by hand if, e.g., the
986 module parameters have changed.
988 Also, at this writing, YaST/YaST2 will not manage bonding
989 devices (they do not show bonding interfaces on its list of network
990 devices). It is necessary to edit the configuration file by hand to
991 change the bonding configuration.
993 Additional general options and details of the ifcfg file
994 format can be found in an example ifcfg template file:
996 /etc/sysconfig/network/ifcfg.template
998 Note that the template does not document the various BONDING_
999 settings described above, but does describe many of the other options.
1001 3.1.1 Using DHCP with Sysconfig
1002 -------------------------------
1004 Under sysconfig, configuring a device with BOOTPROTO='dhcp'
1005 will cause it to query DHCP for its IP address information. At this
1006 writing, this does not function for bonding devices; the scripts
1007 attempt to obtain the device address from DHCP prior to adding any of
1008 the slave devices. Without active slaves, the DHCP requests are not
1009 sent to the network.
1011 3.1.2 Configuring Multiple Bonds with Sysconfig
1012 -----------------------------------------------
1014 The sysconfig network initialization system is capable of
1015 handling multiple bonding devices. All that is necessary is for each
1016 bonding instance to have an appropriately configured ifcfg-bondX file
1017 (as described above). Do not specify the "max_bonds" parameter to any
1018 instance of bonding, as this will confuse sysconfig. If you require
1019 multiple bonding devices with identical parameters, create multiple
1022 Because the sysconfig scripts supply the bonding module
1023 options in the ifcfg-bondX file, it is not necessary to add them to
1024 the system /etc/modules.conf or /etc/modprobe.conf configuration file.
1026 3.2 Configuration with Initscripts Support
1027 ------------------------------------------
1029 This section applies to distros using a recent version of
1030 initscripts with bonding support, for example, Red Hat Enterprise Linux
1031 version 3 or later, Fedora, etc. On these systems, the network
1032 initialization scripts have knowledge of bonding, and can be configured to
1033 control bonding devices. Note that older versions of the initscripts
1034 package have lower levels of support for bonding; this will be noted where
1037 These distros will not automatically load the network adapter
1038 driver unless the ethX device is configured with an IP address.
1039 Because of this constraint, users must manually configure a
1040 network-script file for all physical adapters that will be members of
1041 a bondX link. Network script files are located in the directory:
1043 /etc/sysconfig/network-scripts
1045 The file name must be prefixed with "ifcfg-eth" and suffixed
1046 with the adapter's physical adapter number. For example, the script
1047 for eth0 would be named /etc/sysconfig/network-scripts/ifcfg-eth0.
1048 Place the following text in the file:
1057 The DEVICE= line will be different for every ethX device and
1058 must correspond with the name of the file, i.e., ifcfg-eth1 must have
1059 a device line of DEVICE=eth1. The setting of the MASTER= line will
1060 also depend on the final bonding interface name chosen for your bond.
1061 As with other network devices, these typically start at 0, and go up
1062 one for each device, i.e., the first bonding instance is bond0, the
1063 second is bond1, and so on.
1065 Next, create a bond network script. The file name for this
1066 script will be /etc/sysconfig/network-scripts/ifcfg-bondX where X is
1067 the number of the bond. For bond0 the file is named "ifcfg-bond0",
1068 for bond1 it is named "ifcfg-bond1", and so on. Within that file,
1069 place the following text:
1073 NETMASK=255.255.255.0
1075 BROADCAST=192.168.1.255
1080 Be sure to change the networking specific lines (IPADDR,
1081 NETMASK, NETWORK and BROADCAST) to match your network configuration.
1083 For later versions of initscripts, such as that found with Fedora
1084 7 (or later) and Red Hat Enterprise Linux version 5 (or later), it is possible,
1085 and, indeed, preferable, to specify the bonding options in the ifcfg-bond0
1086 file, e.g. a line of the format:
1088 BONDING_OPTS="mode=active-backup arp_interval=60 arp_ip_target=192.168.1.254"
1090 will configure the bond with the specified options. The options
1091 specified in BONDING_OPTS are identical to the bonding module parameters
1092 except for the arp_ip_target field when using versions of initscripts older
1093 than and 8.57 (Fedora 8) and 8.45.19 (Red Hat Enterprise Linux 5.2). When
1094 using older versions each target should be included as a separate option and
1095 should be preceded by a '+' to indicate it should be added to the list of
1096 queried targets, e.g.,
1098 arp_ip_target=+192.168.1.1 arp_ip_target=+192.168.1.2
1100 is the proper syntax to specify multiple targets. When specifying
1101 options via BONDING_OPTS, it is not necessary to edit /etc/modules.conf or
1104 For even older versions of initscripts that do not support
1105 BONDING_OPTS, it is necessary to edit /etc/modules.conf (or
1106 /etc/modprobe.conf, depending upon your distro) to load the bonding module
1107 with your desired options when the bond0 interface is brought up. The
1108 following lines in /etc/modules.conf (or modprobe.conf) will load the
1109 bonding module, and select its options:
1112 options bond0 mode=balance-alb miimon=100
1114 Replace the sample parameters with the appropriate set of
1115 options for your configuration.
1117 Finally run "/etc/rc.d/init.d/network restart" as root. This
1118 will restart the networking subsystem and your bond link should be now
1121 3.2.1 Using DHCP with Initscripts
1122 ---------------------------------
1124 Recent versions of initscripts (the versions supplied with Fedora
1125 Core 3 and Red Hat Enterprise Linux 4, or later versions, are reported to
1126 work) have support for assigning IP information to bonding devices via
1129 To configure bonding for DHCP, configure it as described
1130 above, except replace the line "BOOTPROTO=none" with "BOOTPROTO=dhcp"
1131 and add a line consisting of "TYPE=Bonding". Note that the TYPE value
1134 3.2.2 Configuring Multiple Bonds with Initscripts
1135 -------------------------------------------------
1137 Initscripts packages that are included with Fedora 7 and Red Hat
1138 Enterprise Linux 5 support multiple bonding interfaces by simply
1139 specifying the appropriate BONDING_OPTS= in ifcfg-bondX where X is the
1140 number of the bond. This support requires sysfs support in the kernel,
1141 and a bonding driver of version 3.0.0 or later. Other configurations may
1142 not support this method for specifying multiple bonding interfaces; for
1143 those instances, see the "Configuring Multiple Bonds Manually" section,
1146 3.3 Configuring Bonding Manually with Ifenslave
1147 -----------------------------------------------
1149 This section applies to distros whose network initialization
1150 scripts (the sysconfig or initscripts package) do not have specific
1151 knowledge of bonding. One such distro is SuSE Linux Enterprise Server
1154 The general method for these systems is to place the bonding
1155 module parameters into /etc/modules.conf or /etc/modprobe.conf (as
1156 appropriate for the installed distro), then add modprobe and/or
1157 ifenslave commands to the system's global init script. The name of
1158 the global init script differs; for sysconfig, it is
1159 /etc/init.d/boot.local and for initscripts it is /etc/rc.d/rc.local.
1161 For example, if you wanted to make a simple bond of two e100
1162 devices (presumed to be eth0 and eth1), and have it persist across
1163 reboots, edit the appropriate file (/etc/init.d/boot.local or
1164 /etc/rc.d/rc.local), and add the following:
1166 modprobe bonding mode=balance-alb miimon=100
1168 ifconfig bond0 192.168.1.1 netmask 255.255.255.0 up
1169 ifenslave bond0 eth0
1170 ifenslave bond0 eth1
1172 Replace the example bonding module parameters and bond0
1173 network configuration (IP address, netmask, etc) with the appropriate
1174 values for your configuration.
1176 Unfortunately, this method will not provide support for the
1177 ifup and ifdown scripts on the bond devices. To reload the bonding
1178 configuration, it is necessary to run the initialization script, e.g.,
1180 # /etc/init.d/boot.local
1184 # /etc/rc.d/rc.local
1186 It may be desirable in such a case to create a separate script
1187 which only initializes the bonding configuration, then call that
1188 separate script from within boot.local. This allows for bonding to be
1189 enabled without re-running the entire global init script.
1191 To shut down the bonding devices, it is necessary to first
1192 mark the bonding device itself as being down, then remove the
1193 appropriate device driver modules. For our example above, you can do
1196 # ifconfig bond0 down
1200 Again, for convenience, it may be desirable to create a script
1201 with these commands.
1204 3.3.1 Configuring Multiple Bonds Manually
1205 -----------------------------------------
1207 This section contains information on configuring multiple
1208 bonding devices with differing options for those systems whose network
1209 initialization scripts lack support for configuring multiple bonds.
1211 If you require multiple bonding devices, but all with the same
1212 options, you may wish to use the "max_bonds" module parameter,
1215 To create multiple bonding devices with differing options, it is
1216 preferrable to use bonding parameters exported by sysfs, documented in the
1219 For versions of bonding without sysfs support, the only means to
1220 provide multiple instances of bonding with differing options is to load
1221 the bonding driver multiple times. Note that current versions of the
1222 sysconfig network initialization scripts handle this automatically; if
1223 your distro uses these scripts, no special action is needed. See the
1224 section Configuring Bonding Devices, above, if you're not sure about your
1225 network initialization scripts.
1227 To load multiple instances of the module, it is necessary to
1228 specify a different name for each instance (the module loading system
1229 requires that every loaded module, even multiple instances of the same
1230 module, have a unique name). This is accomplished by supplying multiple
1231 sets of bonding options in /etc/modprobe.conf, for example:
1234 options bond0 -o bond0 mode=balance-rr miimon=100
1237 options bond1 -o bond1 mode=balance-alb miimon=50
1239 will load the bonding module two times. The first instance is
1240 named "bond0" and creates the bond0 device in balance-rr mode with an
1241 miimon of 100. The second instance is named "bond1" and creates the
1242 bond1 device in balance-alb mode with an miimon of 50.
1244 In some circumstances (typically with older distributions),
1245 the above does not work, and the second bonding instance never sees
1246 its options. In that case, the second options line can be substituted
1249 install bond1 /sbin/modprobe --ignore-install bonding -o bond1 \
1250 mode=balance-alb miimon=50
1252 This may be repeated any number of times, specifying a new and
1253 unique name in place of bond1 for each subsequent instance.
1255 It has been observed that some Red Hat supplied kernels are unable
1256 to rename modules at load time (the "-o bond1" part). Attempts to pass
1257 that option to modprobe will produce an "Operation not permitted" error.
1258 This has been reported on some Fedora Core kernels, and has been seen on
1259 RHEL 4 as well. On kernels exhibiting this problem, it will be impossible
1260 to configure multiple bonds with differing parameters (as they are older
1261 kernels, and also lack sysfs support).
1263 3.4 Configuring Bonding Manually via Sysfs
1264 ------------------------------------------
1266 Starting with version 3.0.0, Channel Bonding may be configured
1267 via the sysfs interface. This interface allows dynamic configuration
1268 of all bonds in the system without unloading the module. It also
1269 allows for adding and removing bonds at runtime. Ifenslave is no
1270 longer required, though it is still supported.
1272 Use of the sysfs interface allows you to use multiple bonds
1273 with different configurations without having to reload the module.
1274 It also allows you to use multiple, differently configured bonds when
1275 bonding is compiled into the kernel.
1277 You must have the sysfs filesystem mounted to configure
1278 bonding this way. The examples in this document assume that you
1279 are using the standard mount point for sysfs, e.g. /sys. If your
1280 sysfs filesystem is mounted elsewhere, you will need to adjust the
1281 example paths accordingly.
1283 Creating and Destroying Bonds
1284 -----------------------------
1285 To add a new bond foo:
1286 # echo +foo > /sys/class/net/bonding_masters
1288 To remove an existing bond bar:
1289 # echo -bar > /sys/class/net/bonding_masters
1291 To show all existing bonds:
1292 # cat /sys/class/net/bonding_masters
1294 NOTE: due to 4K size limitation of sysfs files, this list may be
1295 truncated if you have more than a few hundred bonds. This is unlikely
1296 to occur under normal operating conditions.
1298 Adding and Removing Slaves
1299 --------------------------
1300 Interfaces may be enslaved to a bond using the file
1301 /sys/class/net/<bond>/bonding/slaves. The semantics for this file
1302 are the same as for the bonding_masters file.
1304 To enslave interface eth0 to bond bond0:
1306 # echo +eth0 > /sys/class/net/bond0/bonding/slaves
1308 To free slave eth0 from bond bond0:
1309 # echo -eth0 > /sys/class/net/bond0/bonding/slaves
1311 When an interface is enslaved to a bond, symlinks between the
1312 two are created in the sysfs filesystem. In this case, you would get
1313 /sys/class/net/bond0/slave_eth0 pointing to /sys/class/net/eth0, and
1314 /sys/class/net/eth0/master pointing to /sys/class/net/bond0.
1316 This means that you can tell quickly whether or not an
1317 interface is enslaved by looking for the master symlink. Thus:
1318 # echo -eth0 > /sys/class/net/eth0/master/bonding/slaves
1319 will free eth0 from whatever bond it is enslaved to, regardless of
1320 the name of the bond interface.
1322 Changing a Bond's Configuration
1323 -------------------------------
1324 Each bond may be configured individually by manipulating the
1325 files located in /sys/class/net/<bond name>/bonding
1327 The names of these files correspond directly with the command-
1328 line parameters described elsewhere in this file, and, with the
1329 exception of arp_ip_target, they accept the same values. To see the
1330 current setting, simply cat the appropriate file.
1332 A few examples will be given here; for specific usage
1333 guidelines for each parameter, see the appropriate section in this
1336 To configure bond0 for balance-alb mode:
1337 # ifconfig bond0 down
1338 # echo 6 > /sys/class/net/bond0/bonding/mode
1340 # echo balance-alb > /sys/class/net/bond0/bonding/mode
1341 NOTE: The bond interface must be down before the mode can be
1344 To enable MII monitoring on bond0 with a 1 second interval:
1345 # echo 1000 > /sys/class/net/bond0/bonding/miimon
1346 NOTE: If ARP monitoring is enabled, it will disabled when MII
1347 monitoring is enabled, and vice-versa.
1350 # echo +192.168.0.100 > /sys/class/net/bond0/bonding/arp_ip_target
1351 # echo +192.168.0.101 > /sys/class/net/bond0/bonding/arp_ip_target
1352 NOTE: up to 16 target addresses may be specified.
1354 To remove an ARP target:
1355 # echo -192.168.0.100 > /sys/class/net/bond0/bonding/arp_ip_target
1357 Example Configuration
1358 ---------------------
1359 We begin with the same example that is shown in section 3.3,
1360 executed with sysfs, and without using ifenslave.
1362 To make a simple bond of two e100 devices (presumed to be eth0
1363 and eth1), and have it persist across reboots, edit the appropriate
1364 file (/etc/init.d/boot.local or /etc/rc.d/rc.local), and add the
1369 echo balance-alb > /sys/class/net/bond0/bonding/mode
1370 ifconfig bond0 192.168.1.1 netmask 255.255.255.0 up
1371 echo 100 > /sys/class/net/bond0/bonding/miimon
1372 echo +eth0 > /sys/class/net/bond0/bonding/slaves
1373 echo +eth1 > /sys/class/net/bond0/bonding/slaves
1375 To add a second bond, with two e1000 interfaces in
1376 active-backup mode, using ARP monitoring, add the following lines to
1380 echo +bond1 > /sys/class/net/bonding_masters
1381 echo active-backup > /sys/class/net/bond1/bonding/mode
1382 ifconfig bond1 192.168.2.1 netmask 255.255.255.0 up
1383 echo +192.168.2.100 /sys/class/net/bond1/bonding/arp_ip_target
1384 echo 2000 > /sys/class/net/bond1/bonding/arp_interval
1385 echo +eth2 > /sys/class/net/bond1/bonding/slaves
1386 echo +eth3 > /sys/class/net/bond1/bonding/slaves
1388 3.5 Configuration with Interfaces Support
1389 -----------------------------------------
1391 This section applies to distros which use /etc/network/interfaces file
1392 to describe network interface configuration, most notably Debian and it's
1395 The ifup and ifdown commands on Debian don't support bonding out of
1396 the box. The ifenslave-2.6 package should be installed to provide bonding
1397 support. Once installed, this package will provide bond-* options to be used
1398 into /etc/network/interfaces.
1400 Note that ifenslave-2.6 package will load the bonding module and use
1401 the ifenslave command when appropriate.
1403 Example Configurations
1404 ----------------------
1406 In /etc/network/interfaces, the following stanza will configure bond0, in
1407 active-backup mode, with eth0 and eth1 as slaves.
1410 iface bond0 inet dhcp
1411 bond-slaves eth0 eth1
1412 bond-mode active-backup
1414 bond-primary eth0 eth1
1416 If the above configuration doesn't work, you might have a system using
1417 upstart for system startup. This is most notably true for recent
1418 Ubuntu versions. The following stanza in /etc/network/interfaces will
1419 produce the same result on those systems.
1422 iface bond0 inet dhcp
1424 bond-mode active-backup
1428 iface eth0 inet manual
1430 bond-primary eth0 eth1
1433 iface eth1 inet manual
1435 bond-primary eth0 eth1
1437 For a full list of bond-* supported options in /etc/network/interfaces and some
1438 more advanced examples tailored to you particular distros, see the files in
1439 /usr/share/doc/ifenslave-2.6.
1441 3.6 Overriding Configuration for Special Cases
1442 ----------------------------------------------
1444 When using the bonding driver, the physical port which transmits a frame is
1445 typically selected by the bonding driver, and is not relevant to the user or
1446 system administrator. The output port is simply selected using the policies of
1447 the selected bonding mode. On occasion however, it is helpful to direct certain
1448 classes of traffic to certain physical interfaces on output to implement
1449 slightly more complex policies. For example, to reach a web server over a
1450 bonded interface in which eth0 connects to a private network, while eth1
1451 connects via a public network, it may be desirous to bias the bond to send said
1452 traffic over eth0 first, using eth1 only as a fall back, while all other traffic
1453 can safely be sent over either interface. Such configurations may be achieved
1454 using the traffic control utilities inherent in linux.
1456 By default the bonding driver is multiqueue aware and 16 queues are created
1457 when the driver initializes (see Documentation/networking/multiqueue.txt
1458 for details). If more or less queues are desired the module parameter
1459 tx_queues can be used to change this value. There is no sysfs parameter
1460 available as the allocation is done at module init time.
1462 The output of the file /proc/net/bonding/bondX has changed so the output Queue
1463 ID is now printed for each slave:
1465 Bonding Mode: fault-tolerance (active-backup)
1467 Currently Active Slave: eth0
1469 MII Polling Interval (ms): 0
1473 Slave Interface: eth0
1475 Link Failure Count: 0
1476 Permanent HW addr: 00:1a:a0:12:8f:cb
1479 Slave Interface: eth1
1481 Link Failure Count: 0
1482 Permanent HW addr: 00:1a:a0:12:8f:cc
1485 The queue_id for a slave can be set using the command:
1487 # echo "eth1:2" > /sys/class/net/bond0/bonding/queue_id
1489 Any interface that needs a queue_id set should set it with multiple calls
1490 like the one above until proper priorities are set for all interfaces. On
1491 distributions that allow configuration via initscripts, multiple 'queue_id'
1492 arguments can be added to BONDING_OPTS to set all needed slave queues.
1494 These queue id's can be used in conjunction with the tc utility to configure
1495 a multiqueue qdisc and filters to bias certain traffic to transmit on certain
1496 slave devices. For instance, say we wanted, in the above configuration to
1497 force all traffic bound to 192.168.1.100 to use eth1 in the bond as its output
1498 device. The following commands would accomplish this:
1500 # tc qdisc add dev bond0 handle 1 root multiq
1502 # tc filter add dev bond0 protocol ip parent 1: prio 1 u32 match ip dst \
1503 192.168.1.100 action skbedit queue_mapping 2
1505 These commands tell the kernel to attach a multiqueue queue discipline to the
1506 bond0 interface and filter traffic enqueued to it, such that packets with a dst
1507 ip of 192.168.1.100 have their output queue mapping value overwritten to 2.
1508 This value is then passed into the driver, causing the normal output path
1509 selection policy to be overridden, selecting instead qid 2, which maps to eth1.
1511 Note that qid values begin at 1. Qid 0 is reserved to initiate to the driver
1512 that normal output policy selection should take place. One benefit to simply
1513 leaving the qid for a slave to 0 is the multiqueue awareness in the bonding
1514 driver that is now present. This awareness allows tc filters to be placed on
1515 slave devices as well as bond devices and the bonding driver will simply act as
1516 a pass-through for selecting output queues on the slave device rather than
1517 output port selection.
1519 This feature first appeared in bonding driver version 3.7.0 and support for
1520 output slave selection was limited to round-robin and active-backup modes.
1522 4 Querying Bonding Configuration
1523 =================================
1525 4.1 Bonding Configuration
1526 -------------------------
1528 Each bonding device has a read-only file residing in the
1529 /proc/net/bonding directory. The file contents include information
1530 about the bonding configuration, options and state of each slave.
1532 For example, the contents of /proc/net/bonding/bond0 after the
1533 driver is loaded with parameters of mode=0 and miimon=1000 is
1534 generally as follows:
1536 Ethernet Channel Bonding Driver: 2.6.1 (October 29, 2004)
1537 Bonding Mode: load balancing (round-robin)
1538 Currently Active Slave: eth0
1540 MII Polling Interval (ms): 1000
1544 Slave Interface: eth1
1546 Link Failure Count: 1
1548 Slave Interface: eth0
1550 Link Failure Count: 1
1552 The precise format and contents will change depending upon the
1553 bonding configuration, state, and version of the bonding driver.
1555 4.2 Network configuration
1556 -------------------------
1558 The network configuration can be inspected using the ifconfig
1559 command. Bonding devices will have the MASTER flag set; Bonding slave
1560 devices will have the SLAVE flag set. The ifconfig output does not
1561 contain information on which slaves are associated with which masters.
1563 In the example below, the bond0 interface is the master
1564 (MASTER) while eth0 and eth1 are slaves (SLAVE). Notice all slaves of
1565 bond0 have the same MAC address (HWaddr) as bond0 for all modes except
1566 TLB and ALB that require a unique MAC address for each slave.
1569 bond0 Link encap:Ethernet HWaddr 00:C0:F0:1F:37:B4
1570 inet addr:XXX.XXX.XXX.YYY Bcast:XXX.XXX.XXX.255 Mask:255.255.252.0
1571 UP BROADCAST RUNNING MASTER MULTICAST MTU:1500 Metric:1
1572 RX packets:7224794 errors:0 dropped:0 overruns:0 frame:0
1573 TX packets:3286647 errors:1 dropped:0 overruns:1 carrier:0
1574 collisions:0 txqueuelen:0
1576 eth0 Link encap:Ethernet HWaddr 00:C0:F0:1F:37:B4
1577 UP BROADCAST RUNNING SLAVE MULTICAST MTU:1500 Metric:1
1578 RX packets:3573025 errors:0 dropped:0 overruns:0 frame:0
1579 TX packets:1643167 errors:1 dropped:0 overruns:1 carrier:0
1580 collisions:0 txqueuelen:100
1581 Interrupt:10 Base address:0x1080
1583 eth1 Link encap:Ethernet HWaddr 00:C0:F0:1F:37:B4
1584 UP BROADCAST RUNNING SLAVE MULTICAST MTU:1500 Metric:1
1585 RX packets:3651769 errors:0 dropped:0 overruns:0 frame:0
1586 TX packets:1643480 errors:0 dropped:0 overruns:0 carrier:0
1587 collisions:0 txqueuelen:100
1588 Interrupt:9 Base address:0x1400
1590 5. Switch Configuration
1591 =======================
1593 For this section, "switch" refers to whatever system the
1594 bonded devices are directly connected to (i.e., where the other end of
1595 the cable plugs into). This may be an actual dedicated switch device,
1596 or it may be another regular system (e.g., another computer running
1599 The active-backup, balance-tlb and balance-alb modes do not
1600 require any specific configuration of the switch.
1602 The 802.3ad mode requires that the switch have the appropriate
1603 ports configured as an 802.3ad aggregation. The precise method used
1604 to configure this varies from switch to switch, but, for example, a
1605 Cisco 3550 series switch requires that the appropriate ports first be
1606 grouped together in a single etherchannel instance, then that
1607 etherchannel is set to mode "lacp" to enable 802.3ad (instead of
1608 standard EtherChannel).
1610 The balance-rr, balance-xor and broadcast modes generally
1611 require that the switch have the appropriate ports grouped together.
1612 The nomenclature for such a group differs between switches, it may be
1613 called an "etherchannel" (as in the Cisco example, above), a "trunk
1614 group" or some other similar variation. For these modes, each switch
1615 will also have its own configuration options for the switch's transmit
1616 policy to the bond. Typical choices include XOR of either the MAC or
1617 IP addresses. The transmit policy of the two peers does not need to
1618 match. For these three modes, the bonding mode really selects a
1619 transmit policy for an EtherChannel group; all three will interoperate
1620 with another EtherChannel group.
1623 6. 802.1q VLAN Support
1624 ======================
1626 It is possible to configure VLAN devices over a bond interface
1627 using the 8021q driver. However, only packets coming from the 8021q
1628 driver and passing through bonding will be tagged by default. Self
1629 generated packets, for example, bonding's learning packets or ARP
1630 packets generated by either ALB mode or the ARP monitor mechanism, are
1631 tagged internally by bonding itself. As a result, bonding must
1632 "learn" the VLAN IDs configured above it, and use those IDs to tag
1633 self generated packets.
1635 For reasons of simplicity, and to support the use of adapters
1636 that can do VLAN hardware acceleration offloading, the bonding
1637 interface declares itself as fully hardware offloading capable, it gets
1638 the add_vid/kill_vid notifications to gather the necessary
1639 information, and it propagates those actions to the slaves. In case
1640 of mixed adapter types, hardware accelerated tagged packets that
1641 should go through an adapter that is not offloading capable are
1642 "un-accelerated" by the bonding driver so the VLAN tag sits in the
1645 VLAN interfaces *must* be added on top of a bonding interface
1646 only after enslaving at least one slave. The bonding interface has a
1647 hardware address of 00:00:00:00:00:00 until the first slave is added.
1648 If the VLAN interface is created prior to the first enslavement, it
1649 would pick up the all-zeroes hardware address. Once the first slave
1650 is attached to the bond, the bond device itself will pick up the
1651 slave's hardware address, which is then available for the VLAN device.
1653 Also, be aware that a similar problem can occur if all slaves
1654 are released from a bond that still has one or more VLAN interfaces on
1655 top of it. When a new slave is added, the bonding interface will
1656 obtain its hardware address from the first slave, which might not
1657 match the hardware address of the VLAN interfaces (which was
1658 ultimately copied from an earlier slave).
1660 There are two methods to insure that the VLAN device operates
1661 with the correct hardware address if all slaves are removed from a
1664 1. Remove all VLAN interfaces then recreate them
1666 2. Set the bonding interface's hardware address so that it
1667 matches the hardware address of the VLAN interfaces.
1669 Note that changing a VLAN interface's HW address would set the
1670 underlying device -- i.e. the bonding interface -- to promiscuous
1671 mode, which might not be what you want.
1677 The bonding driver at present supports two schemes for
1678 monitoring a slave device's link state: the ARP monitor and the MII
1681 At the present time, due to implementation restrictions in the
1682 bonding driver itself, it is not possible to enable both ARP and MII
1683 monitoring simultaneously.
1685 7.1 ARP Monitor Operation
1686 -------------------------
1688 The ARP monitor operates as its name suggests: it sends ARP
1689 queries to one or more designated peer systems on the network, and
1690 uses the response as an indication that the link is operating. This
1691 gives some assurance that traffic is actually flowing to and from one
1692 or more peers on the local network.
1694 The ARP monitor relies on the device driver itself to verify
1695 that traffic is flowing. In particular, the driver must keep up to
1696 date the last receive time, dev->last_rx, and transmit start time,
1697 dev->trans_start. If these are not updated by the driver, then the
1698 ARP monitor will immediately fail any slaves using that driver, and
1699 those slaves will stay down. If networking monitoring (tcpdump, etc)
1700 shows the ARP requests and replies on the network, then it may be that
1701 your device driver is not updating last_rx and trans_start.
1703 7.2 Configuring Multiple ARP Targets
1704 ------------------------------------
1706 While ARP monitoring can be done with just one target, it can
1707 be useful in a High Availability setup to have several targets to
1708 monitor. In the case of just one target, the target itself may go
1709 down or have a problem making it unresponsive to ARP requests. Having
1710 an additional target (or several) increases the reliability of the ARP
1713 Multiple ARP targets must be separated by commas as follows:
1715 # example options for ARP monitoring with three targets
1717 options bond0 arp_interval=60 arp_ip_target=192.168.0.1,192.168.0.3,192.168.0.9
1719 For just a single target the options would resemble:
1721 # example options for ARP monitoring with one target
1723 options bond0 arp_interval=60 arp_ip_target=192.168.0.100
1726 7.3 MII Monitor Operation
1727 -------------------------
1729 The MII monitor monitors only the carrier state of the local
1730 network interface. It accomplishes this in one of three ways: by
1731 depending upon the device driver to maintain its carrier state, by
1732 querying the device's MII registers, or by making an ethtool query to
1735 If the use_carrier module parameter is 1 (the default value),
1736 then the MII monitor will rely on the driver for carrier state
1737 information (via the netif_carrier subsystem). As explained in the
1738 use_carrier parameter information, above, if the MII monitor fails to
1739 detect carrier loss on the device (e.g., when the cable is physically
1740 disconnected), it may be that the driver does not support
1743 If use_carrier is 0, then the MII monitor will first query the
1744 device's (via ioctl) MII registers and check the link state. If that
1745 request fails (not just that it returns carrier down), then the MII
1746 monitor will make an ethtool ETHOOL_GLINK request to attempt to obtain
1747 the same information. If both methods fail (i.e., the driver either
1748 does not support or had some error in processing both the MII register
1749 and ethtool requests), then the MII monitor will assume the link is
1752 8. Potential Sources of Trouble
1753 ===============================
1755 8.1 Adventures in Routing
1756 -------------------------
1758 When bonding is configured, it is important that the slave
1759 devices not have routes that supersede routes of the master (or,
1760 generally, not have routes at all). For example, suppose the bonding
1761 device bond0 has two slaves, eth0 and eth1, and the routing table is
1764 Kernel IP routing table
1765 Destination Gateway Genmask Flags MSS Window irtt Iface
1766 10.0.0.0 0.0.0.0 255.255.0.0 U 40 0 0 eth0
1767 10.0.0.0 0.0.0.0 255.255.0.0 U 40 0 0 eth1
1768 10.0.0.0 0.0.0.0 255.255.0.0 U 40 0 0 bond0
1769 127.0.0.0 0.0.0.0 255.0.0.0 U 40 0 0 lo
1771 This routing configuration will likely still update the
1772 receive/transmit times in the driver (needed by the ARP monitor), but
1773 may bypass the bonding driver (because outgoing traffic to, in this
1774 case, another host on network 10 would use eth0 or eth1 before bond0).
1776 The ARP monitor (and ARP itself) may become confused by this
1777 configuration, because ARP requests (generated by the ARP monitor)
1778 will be sent on one interface (bond0), but the corresponding reply
1779 will arrive on a different interface (eth0). This reply looks to ARP
1780 as an unsolicited ARP reply (because ARP matches replies on an
1781 interface basis), and is discarded. The MII monitor is not affected
1782 by the state of the routing table.
1784 The solution here is simply to insure that slaves do not have
1785 routes of their own, and if for some reason they must, those routes do
1786 not supersede routes of their master. This should generally be the
1787 case, but unusual configurations or errant manual or automatic static
1788 route additions may cause trouble.
1790 8.2 Ethernet Device Renaming
1791 ----------------------------
1793 On systems with network configuration scripts that do not
1794 associate physical devices directly with network interface names (so
1795 that the same physical device always has the same "ethX" name), it may
1796 be necessary to add some special logic to either /etc/modules.conf or
1797 /etc/modprobe.conf (depending upon which is installed on the system).
1799 For example, given a modules.conf containing the following:
1802 options bond0 mode=some-mode miimon=50
1808 If neither eth0 and eth1 are slaves to bond0, then when the
1809 bond0 interface comes up, the devices may end up reordered. This
1810 happens because bonding is loaded first, then its slave device's
1811 drivers are loaded next. Since no other drivers have been loaded,
1812 when the e1000 driver loads, it will receive eth0 and eth1 for its
1813 devices, but the bonding configuration tries to enslave eth2 and eth3
1814 (which may later be assigned to the tg3 devices).
1816 Adding the following:
1818 add above bonding e1000 tg3
1820 causes modprobe to load e1000 then tg3, in that order, when
1821 bonding is loaded. This command is fully documented in the
1822 modules.conf manual page.
1824 On systems utilizing modprobe.conf (or modprobe.conf.local),
1825 an equivalent problem can occur. In this case, the following can be
1826 added to modprobe.conf (or modprobe.conf.local, as appropriate), as
1827 follows (all on one line; it has been split here for clarity):
1829 install bonding /sbin/modprobe tg3; /sbin/modprobe e1000;
1830 /sbin/modprobe --ignore-install bonding
1832 This will, when loading the bonding module, rather than
1833 performing the normal action, instead execute the provided command.
1834 This command loads the device drivers in the order needed, then calls
1835 modprobe with --ignore-install to cause the normal action to then take
1836 place. Full documentation on this can be found in the modprobe.conf
1837 and modprobe manual pages.
1839 8.3. Painfully Slow Or No Failed Link Detection By Miimon
1840 ---------------------------------------------------------
1842 By default, bonding enables the use_carrier option, which
1843 instructs bonding to trust the driver to maintain carrier state.
1845 As discussed in the options section, above, some drivers do
1846 not support the netif_carrier_on/_off link state tracking system.
1847 With use_carrier enabled, bonding will always see these links as up,
1848 regardless of their actual state.
1850 Additionally, other drivers do support netif_carrier, but do
1851 not maintain it in real time, e.g., only polling the link state at
1852 some fixed interval. In this case, miimon will detect failures, but
1853 only after some long period of time has expired. If it appears that
1854 miimon is very slow in detecting link failures, try specifying
1855 use_carrier=0 to see if that improves the failure detection time. If
1856 it does, then it may be that the driver checks the carrier state at a
1857 fixed interval, but does not cache the MII register values (so the
1858 use_carrier=0 method of querying the registers directly works). If
1859 use_carrier=0 does not improve the failover, then the driver may cache
1860 the registers, or the problem may be elsewhere.
1862 Also, remember that miimon only checks for the device's
1863 carrier state. It has no way to determine the state of devices on or
1864 beyond other ports of a switch, or if a switch is refusing to pass
1865 traffic while still maintaining carrier on.
1870 If running SNMP agents, the bonding driver should be loaded
1871 before any network drivers participating in a bond. This requirement
1872 is due to the interface index (ipAdEntIfIndex) being associated to
1873 the first interface found with a given IP address. That is, there is
1874 only one ipAdEntIfIndex for each IP address. For example, if eth0 and
1875 eth1 are slaves of bond0 and the driver for eth0 is loaded before the
1876 bonding driver, the interface for the IP address will be associated
1877 with the eth0 interface. This configuration is shown below, the IP
1878 address 192.168.1.1 has an interface index of 2 which indexes to eth0
1879 in the ifDescr table (ifDescr.2).
1881 interfaces.ifTable.ifEntry.ifDescr.1 = lo
1882 interfaces.ifTable.ifEntry.ifDescr.2 = eth0
1883 interfaces.ifTable.ifEntry.ifDescr.3 = eth1
1884 interfaces.ifTable.ifEntry.ifDescr.4 = eth2
1885 interfaces.ifTable.ifEntry.ifDescr.5 = eth3
1886 interfaces.ifTable.ifEntry.ifDescr.6 = bond0
1887 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.10.10.10 = 5
1888 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.192.168.1.1 = 2
1889 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.74.20.94 = 4
1890 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.127.0.0.1 = 1
1892 This problem is avoided by loading the bonding driver before
1893 any network drivers participating in a bond. Below is an example of
1894 loading the bonding driver first, the IP address 192.168.1.1 is
1895 correctly associated with ifDescr.2.
1897 interfaces.ifTable.ifEntry.ifDescr.1 = lo
1898 interfaces.ifTable.ifEntry.ifDescr.2 = bond0
1899 interfaces.ifTable.ifEntry.ifDescr.3 = eth0
1900 interfaces.ifTable.ifEntry.ifDescr.4 = eth1
1901 interfaces.ifTable.ifEntry.ifDescr.5 = eth2
1902 interfaces.ifTable.ifEntry.ifDescr.6 = eth3
1903 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.10.10.10 = 6
1904 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.192.168.1.1 = 2
1905 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.74.20.94 = 5
1906 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.127.0.0.1 = 1
1908 While some distributions may not report the interface name in
1909 ifDescr, the association between the IP address and IfIndex remains
1910 and SNMP functions such as Interface_Scan_Next will report that
1913 10. Promiscuous mode
1914 ====================
1916 When running network monitoring tools, e.g., tcpdump, it is
1917 common to enable promiscuous mode on the device, so that all traffic
1918 is seen (instead of seeing only traffic destined for the local host).
1919 The bonding driver handles promiscuous mode changes to the bonding
1920 master device (e.g., bond0), and propagates the setting to the slave
1923 For the balance-rr, balance-xor, broadcast, and 802.3ad modes,
1924 the promiscuous mode setting is propagated to all slaves.
1926 For the active-backup, balance-tlb and balance-alb modes, the
1927 promiscuous mode setting is propagated only to the active slave.
1929 For balance-tlb mode, the active slave is the slave currently
1930 receiving inbound traffic.
1932 For balance-alb mode, the active slave is the slave used as a
1933 "primary." This slave is used for mode-specific control traffic, for
1934 sending to peers that are unassigned or if the load is unbalanced.
1936 For the active-backup, balance-tlb and balance-alb modes, when
1937 the active slave changes (e.g., due to a link failure), the
1938 promiscuous setting will be propagated to the new active slave.
1940 11. Configuring Bonding for High Availability
1941 =============================================
1943 High Availability refers to configurations that provide
1944 maximum network availability by having redundant or backup devices,
1945 links or switches between the host and the rest of the world. The
1946 goal is to provide the maximum availability of network connectivity
1947 (i.e., the network always works), even though other configurations
1948 could provide higher throughput.
1950 11.1 High Availability in a Single Switch Topology
1951 --------------------------------------------------
1953 If two hosts (or a host and a single switch) are directly
1954 connected via multiple physical links, then there is no availability
1955 penalty to optimizing for maximum bandwidth. In this case, there is
1956 only one switch (or peer), so if it fails, there is no alternative
1957 access to fail over to. Additionally, the bonding load balance modes
1958 support link monitoring of their members, so if individual links fail,
1959 the load will be rebalanced across the remaining devices.
1961 See Section 13, "Configuring Bonding for Maximum Throughput"
1962 for information on configuring bonding with one peer device.
1964 11.2 High Availability in a Multiple Switch Topology
1965 ----------------------------------------------------
1967 With multiple switches, the configuration of bonding and the
1968 network changes dramatically. In multiple switch topologies, there is
1969 a trade off between network availability and usable bandwidth.
1971 Below is a sample network, configured to maximize the
1972 availability of the network:
1976 +-----+----+ +-----+----+
1977 | |port2 ISL port2| |
1978 | switch A +--------------------------+ switch B |
1980 +-----+----+ +-----++---+
1983 +-------------+ host1 +---------------+
1986 In this configuration, there is a link between the two
1987 switches (ISL, or inter switch link), and multiple ports connecting to
1988 the outside world ("port3" on each switch). There is no technical
1989 reason that this could not be extended to a third switch.
1991 11.2.1 HA Bonding Mode Selection for Multiple Switch Topology
1992 -------------------------------------------------------------
1994 In a topology such as the example above, the active-backup and
1995 broadcast modes are the only useful bonding modes when optimizing for
1996 availability; the other modes require all links to terminate on the
1997 same peer for them to behave rationally.
1999 active-backup: This is generally the preferred mode, particularly if
2000 the switches have an ISL and play together well. If the
2001 network configuration is such that one switch is specifically
2002 a backup switch (e.g., has lower capacity, higher cost, etc),
2003 then the primary option can be used to insure that the
2004 preferred link is always used when it is available.
2006 broadcast: This mode is really a special purpose mode, and is suitable
2007 only for very specific needs. For example, if the two
2008 switches are not connected (no ISL), and the networks beyond
2009 them are totally independent. In this case, if it is
2010 necessary for some specific one-way traffic to reach both
2011 independent networks, then the broadcast mode may be suitable.
2013 11.2.2 HA Link Monitoring Selection for Multiple Switch Topology
2014 ----------------------------------------------------------------
2016 The choice of link monitoring ultimately depends upon your
2017 switch. If the switch can reliably fail ports in response to other
2018 failures, then either the MII or ARP monitors should work. For
2019 example, in the above example, if the "port3" link fails at the remote
2020 end, the MII monitor has no direct means to detect this. The ARP
2021 monitor could be configured with a target at the remote end of port3,
2022 thus detecting that failure without switch support.
2024 In general, however, in a multiple switch topology, the ARP
2025 monitor can provide a higher level of reliability in detecting end to
2026 end connectivity failures (which may be caused by the failure of any
2027 individual component to pass traffic for any reason). Additionally,
2028 the ARP monitor should be configured with multiple targets (at least
2029 one for each switch in the network). This will insure that,
2030 regardless of which switch is active, the ARP monitor has a suitable
2033 Note, also, that of late many switches now support a functionality
2034 generally referred to as "trunk failover." This is a feature of the
2035 switch that causes the link state of a particular switch port to be set
2036 down (or up) when the state of another switch port goes down (or up).
2037 Its purpose is to propagate link failures from logically "exterior" ports
2038 to the logically "interior" ports that bonding is able to monitor via
2039 miimon. Availability and configuration for trunk failover varies by
2040 switch, but this can be a viable alternative to the ARP monitor when using
2043 12. Configuring Bonding for Maximum Throughput
2044 ==============================================
2046 12.1 Maximizing Throughput in a Single Switch Topology
2047 ------------------------------------------------------
2049 In a single switch configuration, the best method to maximize
2050 throughput depends upon the application and network environment. The
2051 various load balancing modes each have strengths and weaknesses in
2052 different environments, as detailed below.
2054 For this discussion, we will break down the topologies into
2055 two categories. Depending upon the destination of most traffic, we
2056 categorize them into either "gatewayed" or "local" configurations.
2058 In a gatewayed configuration, the "switch" is acting primarily
2059 as a router, and the majority of traffic passes through this router to
2060 other networks. An example would be the following:
2063 +----------+ +----------+
2064 | |eth0 port1| | to other networks
2065 | Host A +---------------------+ router +------------------->
2066 | +---------------------+ | Hosts B and C are out
2067 | |eth1 port2| | here somewhere
2068 +----------+ +----------+
2070 The router may be a dedicated router device, or another host
2071 acting as a gateway. For our discussion, the important point is that
2072 the majority of traffic from Host A will pass through the router to
2073 some other network before reaching its final destination.
2075 In a gatewayed network configuration, although Host A may
2076 communicate with many other systems, all of its traffic will be sent
2077 and received via one other peer on the local network, the router.
2079 Note that the case of two systems connected directly via
2080 multiple physical links is, for purposes of configuring bonding, the
2081 same as a gatewayed configuration. In that case, it happens that all
2082 traffic is destined for the "gateway" itself, not some other network
2085 In a local configuration, the "switch" is acting primarily as
2086 a switch, and the majority of traffic passes through this switch to
2087 reach other stations on the same network. An example would be the
2090 +----------+ +----------+ +--------+
2091 | |eth0 port1| +-------+ Host B |
2092 | Host A +------------+ switch |port3 +--------+
2093 | +------------+ | +--------+
2094 | |eth1 port2| +------------------+ Host C |
2095 +----------+ +----------+port4 +--------+
2098 Again, the switch may be a dedicated switch device, or another
2099 host acting as a gateway. For our discussion, the important point is
2100 that the majority of traffic from Host A is destined for other hosts
2101 on the same local network (Hosts B and C in the above example).
2103 In summary, in a gatewayed configuration, traffic to and from
2104 the bonded device will be to the same MAC level peer on the network
2105 (the gateway itself, i.e., the router), regardless of its final
2106 destination. In a local configuration, traffic flows directly to and
2107 from the final destinations, thus, each destination (Host B, Host C)
2108 will be addressed directly by their individual MAC addresses.
2110 This distinction between a gatewayed and a local network
2111 configuration is important because many of the load balancing modes
2112 available use the MAC addresses of the local network source and
2113 destination to make load balancing decisions. The behavior of each
2114 mode is described below.
2117 12.1.1 MT Bonding Mode Selection for Single Switch Topology
2118 -----------------------------------------------------------
2120 This configuration is the easiest to set up and to understand,
2121 although you will have to decide which bonding mode best suits your
2122 needs. The trade offs for each mode are detailed below:
2124 balance-rr: This mode is the only mode that will permit a single
2125 TCP/IP connection to stripe traffic across multiple
2126 interfaces. It is therefore the only mode that will allow a
2127 single TCP/IP stream to utilize more than one interface's
2128 worth of throughput. This comes at a cost, however: the
2129 striping generally results in peer systems receiving packets out
2130 of order, causing TCP/IP's congestion control system to kick
2131 in, often by retransmitting segments.
2133 It is possible to adjust TCP/IP's congestion limits by
2134 altering the net.ipv4.tcp_reordering sysctl parameter. The
2135 usual default value is 3, and the maximum useful value is 127.
2136 For a four interface balance-rr bond, expect that a single
2137 TCP/IP stream will utilize no more than approximately 2.3
2138 interface's worth of throughput, even after adjusting
2141 Note that the fraction of packets that will be delivered out of
2142 order is highly variable, and is unlikely to be zero. The level
2143 of reordering depends upon a variety of factors, including the
2144 networking interfaces, the switch, and the topology of the
2145 configuration. Speaking in general terms, higher speed network
2146 cards produce more reordering (due to factors such as packet
2147 coalescing), and a "many to many" topology will reorder at a
2148 higher rate than a "many slow to one fast" configuration.
2150 Many switches do not support any modes that stripe traffic
2151 (instead choosing a port based upon IP or MAC level addresses);
2152 for those devices, traffic for a particular connection flowing
2153 through the switch to a balance-rr bond will not utilize greater
2154 than one interface's worth of bandwidth.
2156 If you are utilizing protocols other than TCP/IP, UDP for
2157 example, and your application can tolerate out of order
2158 delivery, then this mode can allow for single stream datagram
2159 performance that scales near linearly as interfaces are added
2162 This mode requires the switch to have the appropriate ports
2163 configured for "etherchannel" or "trunking."
2165 active-backup: There is not much advantage in this network topology to
2166 the active-backup mode, as the inactive backup devices are all
2167 connected to the same peer as the primary. In this case, a
2168 load balancing mode (with link monitoring) will provide the
2169 same level of network availability, but with increased
2170 available bandwidth. On the plus side, active-backup mode
2171 does not require any configuration of the switch, so it may
2172 have value if the hardware available does not support any of
2173 the load balance modes.
2175 balance-xor: This mode will limit traffic such that packets destined
2176 for specific peers will always be sent over the same
2177 interface. Since the destination is determined by the MAC
2178 addresses involved, this mode works best in a "local" network
2179 configuration (as described above), with destinations all on
2180 the same local network. This mode is likely to be suboptimal
2181 if all your traffic is passed through a single router (i.e., a
2182 "gatewayed" network configuration, as described above).
2184 As with balance-rr, the switch ports need to be configured for
2185 "etherchannel" or "trunking."
2187 broadcast: Like active-backup, there is not much advantage to this
2188 mode in this type of network topology.
2190 802.3ad: This mode can be a good choice for this type of network
2191 topology. The 802.3ad mode is an IEEE standard, so all peers
2192 that implement 802.3ad should interoperate well. The 802.3ad
2193 protocol includes automatic configuration of the aggregates,
2194 so minimal manual configuration of the switch is needed
2195 (typically only to designate that some set of devices is
2196 available for 802.3ad). The 802.3ad standard also mandates
2197 that frames be delivered in order (within certain limits), so
2198 in general single connections will not see misordering of
2199 packets. The 802.3ad mode does have some drawbacks: the
2200 standard mandates that all devices in the aggregate operate at
2201 the same speed and duplex. Also, as with all bonding load
2202 balance modes other than balance-rr, no single connection will
2203 be able to utilize more than a single interface's worth of
2206 Additionally, the linux bonding 802.3ad implementation
2207 distributes traffic by peer (using an XOR of MAC addresses),
2208 so in a "gatewayed" configuration, all outgoing traffic will
2209 generally use the same device. Incoming traffic may also end
2210 up on a single device, but that is dependent upon the
2211 balancing policy of the peer's 8023.ad implementation. In a
2212 "local" configuration, traffic will be distributed across the
2213 devices in the bond.
2215 Finally, the 802.3ad mode mandates the use of the MII monitor,
2216 therefore, the ARP monitor is not available in this mode.
2218 balance-tlb: The balance-tlb mode balances outgoing traffic by peer.
2219 Since the balancing is done according to MAC address, in a
2220 "gatewayed" configuration (as described above), this mode will
2221 send all traffic across a single device. However, in a
2222 "local" network configuration, this mode balances multiple
2223 local network peers across devices in a vaguely intelligent
2224 manner (not a simple XOR as in balance-xor or 802.3ad mode),
2225 so that mathematically unlucky MAC addresses (i.e., ones that
2226 XOR to the same value) will not all "bunch up" on a single
2229 Unlike 802.3ad, interfaces may be of differing speeds, and no
2230 special switch configuration is required. On the down side,
2231 in this mode all incoming traffic arrives over a single
2232 interface, this mode requires certain ethtool support in the
2233 network device driver of the slave interfaces, and the ARP
2234 monitor is not available.
2236 balance-alb: This mode is everything that balance-tlb is, and more.
2237 It has all of the features (and restrictions) of balance-tlb,
2238 and will also balance incoming traffic from local network
2239 peers (as described in the Bonding Module Options section,
2242 The only additional down side to this mode is that the network
2243 device driver must support changing the hardware address while
2246 12.1.2 MT Link Monitoring for Single Switch Topology
2247 ----------------------------------------------------
2249 The choice of link monitoring may largely depend upon which
2250 mode you choose to use. The more advanced load balancing modes do not
2251 support the use of the ARP monitor, and are thus restricted to using
2252 the MII monitor (which does not provide as high a level of end to end
2253 assurance as the ARP monitor).
2255 12.2 Maximum Throughput in a Multiple Switch Topology
2256 -----------------------------------------------------
2258 Multiple switches may be utilized to optimize for throughput
2259 when they are configured in parallel as part of an isolated network
2260 between two or more systems, for example:
2266 +--------+ | +---------+
2268 +------+---+ +-----+----+ +-----+----+
2269 | Switch A | | Switch B | | Switch C |
2270 +------+---+ +-----+----+ +-----+----+
2272 +--------+ | +---------+
2278 In this configuration, the switches are isolated from one
2279 another. One reason to employ a topology such as this is for an
2280 isolated network with many hosts (a cluster configured for high
2281 performance, for example), using multiple smaller switches can be more
2282 cost effective than a single larger switch, e.g., on a network with 24
2283 hosts, three 24 port switches can be significantly less expensive than
2284 a single 72 port switch.
2286 If access beyond the network is required, an individual host
2287 can be equipped with an additional network device connected to an
2288 external network; this host then additionally acts as a gateway.
2290 12.2.1 MT Bonding Mode Selection for Multiple Switch Topology
2291 -------------------------------------------------------------
2293 In actual practice, the bonding mode typically employed in
2294 configurations of this type is balance-rr. Historically, in this
2295 network configuration, the usual caveats about out of order packet
2296 delivery are mitigated by the use of network adapters that do not do
2297 any kind of packet coalescing (via the use of NAPI, or because the
2298 device itself does not generate interrupts until some number of
2299 packets has arrived). When employed in this fashion, the balance-rr
2300 mode allows individual connections between two hosts to effectively
2301 utilize greater than one interface's bandwidth.
2303 12.2.2 MT Link Monitoring for Multiple Switch Topology
2304 ------------------------------------------------------
2306 Again, in actual practice, the MII monitor is most often used
2307 in this configuration, as performance is given preference over
2308 availability. The ARP monitor will function in this topology, but its
2309 advantages over the MII monitor are mitigated by the volume of probes
2310 needed as the number of systems involved grows (remember that each
2311 host in the network is configured with bonding).
2313 13. Switch Behavior Issues
2314 ==========================
2316 13.1 Link Establishment and Failover Delays
2317 -------------------------------------------
2319 Some switches exhibit undesirable behavior with regard to the
2320 timing of link up and down reporting by the switch.
2322 First, when a link comes up, some switches may indicate that
2323 the link is up (carrier available), but not pass traffic over the
2324 interface for some period of time. This delay is typically due to
2325 some type of autonegotiation or routing protocol, but may also occur
2326 during switch initialization (e.g., during recovery after a switch
2327 failure). If you find this to be a problem, specify an appropriate
2328 value to the updelay bonding module option to delay the use of the
2329 relevant interface(s).
2331 Second, some switches may "bounce" the link state one or more
2332 times while a link is changing state. This occurs most commonly while
2333 the switch is initializing. Again, an appropriate updelay value may
2336 Note that when a bonding interface has no active links, the
2337 driver will immediately reuse the first link that goes up, even if the
2338 updelay parameter has been specified (the updelay is ignored in this
2339 case). If there are slave interfaces waiting for the updelay timeout
2340 to expire, the interface that first went into that state will be
2341 immediately reused. This reduces down time of the network if the
2342 value of updelay has been overestimated, and since this occurs only in
2343 cases with no connectivity, there is no additional penalty for
2344 ignoring the updelay.
2346 In addition to the concerns about switch timings, if your
2347 switches take a long time to go into backup mode, it may be desirable
2348 to not activate a backup interface immediately after a link goes down.
2349 Failover may be delayed via the downdelay bonding module option.
2351 13.2 Duplicated Incoming Packets
2352 --------------------------------
2354 NOTE: Starting with version 3.0.2, the bonding driver has logic to
2355 suppress duplicate packets, which should largely eliminate this problem.
2356 The following description is kept for reference.
2358 It is not uncommon to observe a short burst of duplicated
2359 traffic when the bonding device is first used, or after it has been
2360 idle for some period of time. This is most easily observed by issuing
2361 a "ping" to some other host on the network, and noticing that the
2362 output from ping flags duplicates (typically one per slave).
2364 For example, on a bond in active-backup mode with five slaves
2365 all connected to one switch, the output may appear as follows:
2368 PING 10.0.4.2 (10.0.4.2) from 10.0.3.10 : 56(84) bytes of data.
2369 64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.7 ms
2370 64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
2371 64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
2372 64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
2373 64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
2374 64 bytes from 10.0.4.2: icmp_seq=2 ttl=64 time=0.216 ms
2375 64 bytes from 10.0.4.2: icmp_seq=3 ttl=64 time=0.267 ms
2376 64 bytes from 10.0.4.2: icmp_seq=4 ttl=64 time=0.222 ms
2378 This is not due to an error in the bonding driver, rather, it
2379 is a side effect of how many switches update their MAC forwarding
2380 tables. Initially, the switch does not associate the MAC address in
2381 the packet with a particular switch port, and so it may send the
2382 traffic to all ports until its MAC forwarding table is updated. Since
2383 the interfaces attached to the bond may occupy multiple ports on a
2384 single switch, when the switch (temporarily) floods the traffic to all
2385 ports, the bond device receives multiple copies of the same packet
2386 (one per slave device).
2388 The duplicated packet behavior is switch dependent, some
2389 switches exhibit this, and some do not. On switches that display this
2390 behavior, it can be induced by clearing the MAC forwarding table (on
2391 most Cisco switches, the privileged command "clear mac address-table
2392 dynamic" will accomplish this).
2394 14. Hardware Specific Considerations
2395 ====================================
2397 This section contains additional information for configuring
2398 bonding on specific hardware platforms, or for interfacing bonding
2399 with particular switches or other devices.
2401 14.1 IBM BladeCenter
2402 --------------------
2404 This applies to the JS20 and similar systems.
2406 On the JS20 blades, the bonding driver supports only
2407 balance-rr, active-backup, balance-tlb and balance-alb modes. This is
2408 largely due to the network topology inside the BladeCenter, detailed
2411 JS20 network adapter information
2412 --------------------------------
2414 All JS20s come with two Broadcom Gigabit Ethernet ports
2415 integrated on the planar (that's "motherboard" in IBM-speak). In the
2416 BladeCenter chassis, the eth0 port of all JS20 blades is hard wired to
2417 I/O Module #1; similarly, all eth1 ports are wired to I/O Module #2.
2418 An add-on Broadcom daughter card can be installed on a JS20 to provide
2419 two more Gigabit Ethernet ports. These ports, eth2 and eth3, are
2420 wired to I/O Modules 3 and 4, respectively.
2422 Each I/O Module may contain either a switch or a passthrough
2423 module (which allows ports to be directly connected to an external
2424 switch). Some bonding modes require a specific BladeCenter internal
2425 network topology in order to function; these are detailed below.
2427 Additional BladeCenter-specific networking information can be
2428 found in two IBM Redbooks (www.ibm.com/redbooks):
2430 "IBM eServer BladeCenter Networking Options"
2431 "IBM eServer BladeCenter Layer 2-7 Network Switching"
2433 BladeCenter networking configuration
2434 ------------------------------------
2436 Because a BladeCenter can be configured in a very large number
2437 of ways, this discussion will be confined to describing basic
2440 Normally, Ethernet Switch Modules (ESMs) are used in I/O
2441 modules 1 and 2. In this configuration, the eth0 and eth1 ports of a
2442 JS20 will be connected to different internal switches (in the
2443 respective I/O modules).
2445 A passthrough module (OPM or CPM, optical or copper,
2446 passthrough module) connects the I/O module directly to an external
2447 switch. By using PMs in I/O module #1 and #2, the eth0 and eth1
2448 interfaces of a JS20 can be redirected to the outside world and
2449 connected to a common external switch.
2451 Depending upon the mix of ESMs and PMs, the network will
2452 appear to bonding as either a single switch topology (all PMs) or as a
2453 multiple switch topology (one or more ESMs, zero or more PMs). It is
2454 also possible to connect ESMs together, resulting in a configuration
2455 much like the example in "High Availability in a Multiple Switch
2458 Requirements for specific modes
2459 -------------------------------
2461 The balance-rr mode requires the use of passthrough modules
2462 for devices in the bond, all connected to an common external switch.
2463 That switch must be configured for "etherchannel" or "trunking" on the
2464 appropriate ports, as is usual for balance-rr.
2466 The balance-alb and balance-tlb modes will function with
2467 either switch modules or passthrough modules (or a mix). The only
2468 specific requirement for these modes is that all network interfaces
2469 must be able to reach all destinations for traffic sent over the
2470 bonding device (i.e., the network must converge at some point outside
2473 The active-backup mode has no additional requirements.
2475 Link monitoring issues
2476 ----------------------
2478 When an Ethernet Switch Module is in place, only the ARP
2479 monitor will reliably detect link loss to an external switch. This is
2480 nothing unusual, but examination of the BladeCenter cabinet would
2481 suggest that the "external" network ports are the ethernet ports for
2482 the system, when it fact there is a switch between these "external"
2483 ports and the devices on the JS20 system itself. The MII monitor is
2484 only able to detect link failures between the ESM and the JS20 system.
2486 When a passthrough module is in place, the MII monitor does
2487 detect failures to the "external" port, which is then directly
2488 connected to the JS20 system.
2493 The Serial Over LAN (SoL) link is established over the primary
2494 ethernet (eth0) only, therefore, any loss of link to eth0 will result
2495 in losing your SoL connection. It will not fail over with other
2496 network traffic, as the SoL system is beyond the control of the
2499 It may be desirable to disable spanning tree on the switch
2500 (either the internal Ethernet Switch Module, or an external switch) to
2501 avoid fail-over delay issues when using bonding.
2504 15. Frequently Asked Questions
2505 ==============================
2509 Yes. The old 2.0.xx channel bonding patch was not SMP safe.
2510 The new driver was designed to be SMP safe from the start.
2512 2. What type of cards will work with it?
2514 Any Ethernet type cards (you can even mix cards - a Intel
2515 EtherExpress PRO/100 and a 3com 3c905b, for example). For most modes,
2516 devices need not be of the same speed.
2518 Starting with version 3.2.1, bonding also supports Infiniband
2519 slaves in active-backup mode.
2521 3. How many bonding devices can I have?
2525 4. How many slaves can a bonding device have?
2527 This is limited only by the number of network interfaces Linux
2528 supports and/or the number of network cards you can place in your
2531 5. What happens when a slave link dies?
2533 If link monitoring is enabled, then the failing device will be
2534 disabled. The active-backup mode will fail over to a backup link, and
2535 other modes will ignore the failed link. The link will continue to be
2536 monitored, and should it recover, it will rejoin the bond (in whatever
2537 manner is appropriate for the mode). See the sections on High
2538 Availability and the documentation for each mode for additional
2541 Link monitoring can be enabled via either the miimon or
2542 arp_interval parameters (described in the module parameters section,
2543 above). In general, miimon monitors the carrier state as sensed by
2544 the underlying network device, and the arp monitor (arp_interval)
2545 monitors connectivity to another host on the local network.
2547 If no link monitoring is configured, the bonding driver will
2548 be unable to detect link failures, and will assume that all links are
2549 always available. This will likely result in lost packets, and a
2550 resulting degradation of performance. The precise performance loss
2551 depends upon the bonding mode and network configuration.
2553 6. Can bonding be used for High Availability?
2555 Yes. See the section on High Availability for details.
2557 7. Which switches/systems does it work with?
2559 The full answer to this depends upon the desired mode.
2561 In the basic balance modes (balance-rr and balance-xor), it
2562 works with any system that supports etherchannel (also called
2563 trunking). Most managed switches currently available have such
2564 support, and many unmanaged switches as well.
2566 The advanced balance modes (balance-tlb and balance-alb) do
2567 not have special switch requirements, but do need device drivers that
2568 support specific features (described in the appropriate section under
2569 module parameters, above).
2571 In 802.3ad mode, it works with systems that support IEEE
2572 802.3ad Dynamic Link Aggregation. Most managed and many unmanaged
2573 switches currently available support 802.3ad.
2575 The active-backup mode should work with any Layer-II switch.
2577 8. Where does a bonding device get its MAC address from?
2579 When using slave devices that have fixed MAC addresses, or when
2580 the fail_over_mac option is enabled, the bonding device's MAC address is
2581 the MAC address of the active slave.
2583 For other configurations, if not explicitly configured (with
2584 ifconfig or ip link), the MAC address of the bonding device is taken from
2585 its first slave device. This MAC address is then passed to all following
2586 slaves and remains persistent (even if the first slave is removed) until
2587 the bonding device is brought down or reconfigured.
2589 If you wish to change the MAC address, you can set it with
2590 ifconfig or ip link:
2592 # ifconfig bond0 hw ether 00:11:22:33:44:55
2594 # ip link set bond0 address 66:77:88:99:aa:bb
2596 The MAC address can be also changed by bringing down/up the
2597 device and then changing its slaves (or their order):
2599 # ifconfig bond0 down ; modprobe -r bonding
2600 # ifconfig bond0 .... up
2601 # ifenslave bond0 eth...
2603 This method will automatically take the address from the next
2604 slave that is added.
2606 To restore your slaves' MAC addresses, you need to detach them
2607 from the bond (`ifenslave -d bond0 eth0'). The bonding driver will
2608 then restore the MAC addresses that the slaves had before they were
2611 16. Resources and Links
2612 =======================
2614 The latest version of the bonding driver can be found in the latest
2615 version of the linux kernel, found on http://kernel.org
2617 The latest version of this document can be found in the latest kernel
2618 source (named Documentation/networking/bonding.txt).
2620 Discussions regarding the usage of the bonding driver take place on the
2621 bonding-devel mailing list, hosted at sourceforge.net. If you have questions or
2622 problems, post them to the list. The list address is:
2624 bonding-devel@lists.sourceforge.net
2626 The administrative interface (to subscribe or unsubscribe) can
2629 https://lists.sourceforge.net/lists/listinfo/bonding-devel
2631 Discussions regarding the developpement of the bonding driver take place
2632 on the main Linux network mailing list, hosted at vger.kernel.org. The list
2635 netdev@vger.kernel.org
2637 The administrative interface (to subscribe or unsubscribe) can
2640 http://vger.kernel.org/vger-lists.html#netdev
2642 Donald Becker's Ethernet Drivers and diag programs may be found at :
2643 - http://web.archive.org/web/*/http://www.scyld.com/network/
2645 You will also find a lot of information regarding Ethernet, NWay, MII,
2646 etc. at www.scyld.com.