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. If your distro does not, or you
108 have need to compile bonding from source (e.g., configuring and
109 installing a mainline kernel from kernel.org), you'll need to perform
112 1.1 Configure and build the kernel with bonding
113 -----------------------------------------------
115 The current version of the bonding driver is available in the
116 drivers/net/bonding subdirectory of the most recent kernel source
117 (which is available on http://kernel.org). Most users "rolling their
118 own" will want to use the most recent kernel from kernel.org.
120 Configure kernel with "make menuconfig" (or "make xconfig" or
121 "make config"), then select "Bonding driver support" in the "Network
122 device support" section. It is recommended that you configure the
123 driver as module since it is currently the only way to pass parameters
124 to the driver or configure more than one bonding device.
126 Build and install the new kernel and modules.
128 1.2 Bonding Control Utility
129 -------------------------------------
131 It is recommended to configure bonding via iproute2 (netlink)
132 or sysfs, the old ifenslave control utility is obsolete.
134 2. Bonding Driver Options
135 =========================
137 Options for the bonding driver are supplied as parameters to the
138 bonding module at load time, or are specified via sysfs.
140 Module options may be given as command line arguments to the
141 insmod or modprobe command, but are usually specified in either the
142 /etc/modrobe.d/*.conf configuration files, or in a distro-specific
143 configuration file (some of which are detailed in the next section).
145 Details on bonding support for sysfs is provided in the
146 "Configuring Bonding Manually via Sysfs" section, below.
148 The available bonding driver parameters are listed below. If a
149 parameter is not specified the default value is used. When initially
150 configuring a bond, it is recommended "tail -f /var/log/messages" be
151 run in a separate window to watch for bonding driver error messages.
153 It is critical that either the miimon or arp_interval and
154 arp_ip_target parameters be specified, otherwise serious network
155 degradation will occur during link failures. Very few devices do not
156 support at least miimon, so there is really no reason not to use it.
158 Options with textual values will accept either the text name
159 or, for backwards compatibility, the option value. E.g.,
160 "mode=802.3ad" and "mode=4" set the same mode.
162 The parameters are as follows:
166 Specifies the new active slave for modes that support it
167 (active-backup, balance-alb and balance-tlb). Possible values
168 are the name of any currently enslaved interface, or an empty
169 string. If a name is given, the slave and its link must be up in order
170 to be selected as the new active slave. If an empty string is
171 specified, the current active slave is cleared, and a new active
172 slave is selected automatically.
174 Note that this is only available through the sysfs interface. No module
175 parameter by this name exists.
177 The normal value of this option is the name of the currently
178 active slave, or the empty string if there is no active slave or
179 the current mode does not use an active slave.
183 Specifies the 802.3ad aggregation selection logic to use. The
184 possible values and their effects are:
188 The active aggregator is chosen by largest aggregate
191 Reselection of the active aggregator occurs only when all
192 slaves of the active aggregator are down or the active
193 aggregator has no slaves.
195 This is the default value.
199 The active aggregator is chosen by largest aggregate
200 bandwidth. Reselection occurs if:
202 - A slave is added to or removed from the bond
204 - Any slave's link state changes
206 - Any slave's 802.3ad association state changes
208 - The bond's administrative state changes to up
212 The active aggregator is chosen by the largest number of
213 ports (slaves). Reselection occurs as described under the
214 "bandwidth" setting, above.
216 The bandwidth and count selection policies permit failover of
217 802.3ad aggregations when partial failure of the active aggregator
218 occurs. This keeps the aggregator with the highest availability
219 (either in bandwidth or in number of ports) active at all times.
221 This option was added in bonding version 3.4.0.
225 Specifies that duplicate frames (received on inactive ports) should be
226 dropped (0) or delivered (1).
228 Normally, bonding will drop duplicate frames (received on inactive
229 ports), which is desirable for most users. But there are some times
230 it is nice to allow duplicate frames to be delivered.
232 The default value is 0 (drop duplicate frames received on inactive
237 Specifies the ARP link monitoring frequency in milliseconds.
239 The ARP monitor works by periodically checking the slave
240 devices to determine whether they have sent or received
241 traffic recently (the precise criteria depends upon the
242 bonding mode, and the state of the slave). Regular traffic is
243 generated via ARP probes issued for the addresses specified by
244 the arp_ip_target option.
246 This behavior can be modified by the arp_validate option,
249 If ARP monitoring is used in an etherchannel compatible mode
250 (modes 0 and 2), the switch should be configured in a mode
251 that evenly distributes packets across all links. If the
252 switch is configured to distribute the packets in an XOR
253 fashion, all replies from the ARP targets will be received on
254 the same link which could cause the other team members to
255 fail. ARP monitoring should not be used in conjunction with
256 miimon. A value of 0 disables ARP monitoring. The default
261 Specifies the IP addresses to use as ARP monitoring peers when
262 arp_interval is > 0. These are the targets of the ARP request
263 sent to determine the health of the link to the targets.
264 Specify these values in ddd.ddd.ddd.ddd format. Multiple IP
265 addresses must be separated by a comma. At least one IP
266 address must be given for ARP monitoring to function. The
267 maximum number of targets that can be specified is 16. The
268 default value is no IP addresses.
272 Specifies whether or not ARP probes and replies should be
273 validated in the active-backup mode. This causes the ARP
274 monitor to examine the incoming ARP requests and replies, and
275 only consider a slave to be up if it is receiving the
276 appropriate ARP traffic.
282 No validation is performed. This is the default.
286 Validation is performed only for the active slave.
290 Validation is performed only for backup slaves.
294 Validation is performed for all slaves.
296 For the active slave, the validation checks ARP replies to
297 confirm that they were generated by an arp_ip_target. Since
298 backup slaves do not typically receive these replies, the
299 validation performed for backup slaves is on the ARP request
300 sent out via the active slave. It is possible that some
301 switch or network configurations may result in situations
302 wherein the backup slaves do not receive the ARP requests; in
303 such a situation, validation of backup slaves must be
306 This option is useful in network configurations in which
307 multiple bonding hosts are concurrently issuing ARPs to one or
308 more targets beyond a common switch. Should the link between
309 the switch and target fail (but not the switch itself), the
310 probe traffic generated by the multiple bonding instances will
311 fool the standard ARP monitor into considering the links as
312 still up. Use of the arp_validate option can resolve this, as
313 the ARP monitor will only consider ARP requests and replies
314 associated with its own instance of bonding.
316 This option was added in bonding version 3.1.0.
320 Specifies the time, in milliseconds, to wait before disabling
321 a slave after a link failure has been detected. This option
322 is only valid for the miimon link monitor. The downdelay
323 value should be a multiple of the miimon value; if not, it
324 will be rounded down to the nearest multiple. The default
329 Specifies whether active-backup mode should set all slaves to
330 the same MAC address at enslavement (the traditional
331 behavior), or, when enabled, perform special handling of the
332 bond's MAC address in accordance with the selected policy.
338 This setting disables fail_over_mac, and causes
339 bonding to set all slaves of an active-backup bond to
340 the same MAC address at enslavement time. This is the
345 The "active" fail_over_mac policy indicates that the
346 MAC address of the bond should always be the MAC
347 address of the currently active slave. The MAC
348 address of the slaves is not changed; instead, the MAC
349 address of the bond changes during a failover.
351 This policy is useful for devices that cannot ever
352 alter their MAC address, or for devices that refuse
353 incoming broadcasts with their own source MAC (which
354 interferes with the ARP monitor).
356 The down side of this policy is that every device on
357 the network must be updated via gratuitous ARP,
358 vs. just updating a switch or set of switches (which
359 often takes place for any traffic, not just ARP
360 traffic, if the switch snoops incoming traffic to
361 update its tables) for the traditional method. If the
362 gratuitous ARP is lost, communication may be
365 When this policy is used in conjunction with the mii
366 monitor, devices which assert link up prior to being
367 able to actually transmit and receive are particularly
368 susceptible to loss of the gratuitous ARP, and an
369 appropriate updelay setting may be required.
373 The "follow" fail_over_mac policy causes the MAC
374 address of the bond to be selected normally (normally
375 the MAC address of the first slave added to the bond).
376 However, the second and subsequent slaves are not set
377 to this MAC address while they are in a backup role; a
378 slave is programmed with the bond's MAC address at
379 failover time (and the formerly active slave receives
380 the newly active slave's MAC address).
382 This policy is useful for multiport devices that
383 either become confused or incur a performance penalty
384 when multiple ports are programmed with the same MAC
388 The default policy is none, unless the first slave cannot
389 change its MAC address, in which case the active policy is
392 This option may be modified via sysfs only when no slaves are
395 This option was added in bonding version 3.2.0. The "follow"
396 policy was added in bonding version 3.3.0.
400 Option specifying the rate in which we'll ask our link partner
401 to transmit LACPDU packets in 802.3ad mode. Possible values
405 Request partner to transmit LACPDUs every 30 seconds
408 Request partner to transmit LACPDUs every 1 second
414 Specifies the number of bonding devices to create for this
415 instance of the bonding driver. E.g., if max_bonds is 3, and
416 the bonding driver is not already loaded, then bond0, bond1
417 and bond2 will be created. The default value is 1. Specifying
418 a value of 0 will load bonding, but will not create any devices.
422 Specifies the MII link monitoring frequency in milliseconds.
423 This determines how often the link state of each slave is
424 inspected for link failures. A value of zero disables MII
425 link monitoring. A value of 100 is a good starting point.
426 The use_carrier option, below, affects how the link state is
427 determined. See the High Availability section for additional
428 information. The default value is 0.
432 Specifies the minimum number of links that must be active before
433 asserting carrier. It is similar to the Cisco EtherChannel min-links
434 feature. This allows setting the minimum number of member ports that
435 must be up (link-up state) before marking the bond device as up
436 (carrier on). This is useful for situations where higher level services
437 such as clustering want to ensure a minimum number of low bandwidth
438 links are active before switchover. This option only affect 802.3ad
441 The default value is 0. This will cause carrier to be asserted (for
442 802.3ad mode) whenever there is an active aggregator, regardless of the
443 number of available links in that aggregator. Note that, because an
444 aggregator cannot be active without at least one available link,
445 setting this option to 0 or to 1 has the exact same effect.
449 Specifies one of the bonding policies. The default is
450 balance-rr (round robin). Possible values are:
454 Round-robin policy: Transmit packets in sequential
455 order from the first available slave through the
456 last. This mode provides load balancing and fault
461 Active-backup policy: Only one slave in the bond is
462 active. A different slave becomes active if, and only
463 if, the active slave fails. The bond's MAC address is
464 externally visible on only one port (network adapter)
465 to avoid confusing the switch.
467 In bonding version 2.6.2 or later, when a failover
468 occurs in active-backup mode, bonding will issue one
469 or more gratuitous ARPs on the newly active slave.
470 One gratuitous ARP is issued for the bonding master
471 interface and each VLAN interfaces configured above
472 it, provided that the interface has at least one IP
473 address configured. Gratuitous ARPs issued for VLAN
474 interfaces are tagged with the appropriate VLAN id.
476 This mode provides fault tolerance. The primary
477 option, documented below, affects the behavior of this
482 XOR policy: Transmit based on the selected transmit
483 hash policy. The default policy is a simple [(source
484 MAC address XOR'd with destination MAC address) modulo
485 slave count]. Alternate transmit policies may be
486 selected via the xmit_hash_policy option, described
489 This mode provides load balancing and fault tolerance.
493 Broadcast policy: transmits everything on all slave
494 interfaces. This mode provides fault tolerance.
498 IEEE 802.3ad Dynamic link aggregation. Creates
499 aggregation groups that share the same speed and
500 duplex settings. Utilizes all slaves in the active
501 aggregator according to the 802.3ad specification.
503 Slave selection for outgoing traffic is done according
504 to the transmit hash policy, which may be changed from
505 the default simple XOR policy via the xmit_hash_policy
506 option, documented below. Note that not all transmit
507 policies may be 802.3ad compliant, particularly in
508 regards to the packet mis-ordering requirements of
509 section 43.2.4 of the 802.3ad standard. Differing
510 peer implementations will have varying tolerances for
515 1. Ethtool support in the base drivers for retrieving
516 the speed and duplex of each slave.
518 2. A switch that supports IEEE 802.3ad Dynamic link
521 Most switches will require some type of configuration
522 to enable 802.3ad mode.
526 Adaptive transmit load balancing: channel bonding that
527 does not require any special switch support. The
528 outgoing traffic is distributed according to the
529 current load (computed relative to the speed) on each
530 slave. Incoming traffic is received by the current
531 slave. If the receiving slave fails, another slave
532 takes over the MAC address of the failed receiving
537 Ethtool support in the base drivers for retrieving the
542 Adaptive load balancing: includes balance-tlb plus
543 receive load balancing (rlb) for IPV4 traffic, and
544 does not require any special switch support. The
545 receive load balancing is achieved by ARP negotiation.
546 The bonding driver intercepts the ARP Replies sent by
547 the local system on their way out and overwrites the
548 source hardware address with the unique hardware
549 address of one of the slaves in the bond such that
550 different peers use different hardware addresses for
553 Receive traffic from connections created by the server
554 is also balanced. When the local system sends an ARP
555 Request the bonding driver copies and saves the peer's
556 IP information from the ARP packet. When the ARP
557 Reply arrives from the peer, its hardware address is
558 retrieved and the bonding driver initiates an ARP
559 reply to this peer assigning it to one of the slaves
560 in the bond. A problematic outcome of using ARP
561 negotiation for balancing is that each time that an
562 ARP request is broadcast it uses the hardware address
563 of the bond. Hence, peers learn the hardware address
564 of the bond and the balancing of receive traffic
565 collapses to the current slave. This is handled by
566 sending updates (ARP Replies) to all the peers with
567 their individually assigned hardware address such that
568 the traffic is redistributed. Receive traffic is also
569 redistributed when a new slave is added to the bond
570 and when an inactive slave is re-activated. The
571 receive load is distributed sequentially (round robin)
572 among the group of highest speed slaves in the bond.
574 When a link is reconnected or a new slave joins the
575 bond the receive traffic is redistributed among all
576 active slaves in the bond by initiating ARP Replies
577 with the selected MAC address to each of the
578 clients. The updelay parameter (detailed below) must
579 be set to a value equal or greater than the switch's
580 forwarding delay so that the ARP Replies sent to the
581 peers will not be blocked by the switch.
585 1. Ethtool support in the base drivers for retrieving
586 the speed of each slave.
588 2. Base driver support for setting the hardware
589 address of a device while it is open. This is
590 required so that there will always be one slave in the
591 team using the bond hardware address (the
592 curr_active_slave) while having a unique hardware
593 address for each slave in the bond. If the
594 curr_active_slave fails its hardware address is
595 swapped with the new curr_active_slave that was
601 Specify the number of peer notifications (gratuitous ARPs and
602 unsolicited IPv6 Neighbor Advertisements) to be issued after a
603 failover event. As soon as the link is up on the new slave
604 (possibly immediately) a peer notification is sent on the
605 bonding device and each VLAN sub-device. This is repeated at
606 each link monitor interval (arp_interval or miimon, whichever
607 is active) if the number is greater than 1.
609 The valid range is 0 - 255; the default value is 1. These options
610 affect only the active-backup mode. These options were added for
611 bonding versions 3.3.0 and 3.4.0 respectively.
613 From Linux 3.0 and bonding version 3.7.1, these notifications
614 are generated by the ipv4 and ipv6 code and the numbers of
615 repetitions cannot be set independently.
619 A string (eth0, eth2, etc) specifying which slave is the
620 primary device. The specified device will always be the
621 active slave while it is available. Only when the primary is
622 off-line will alternate devices be used. This is useful when
623 one slave is preferred over another, e.g., when one slave has
624 higher throughput than another.
626 The primary option is only valid for active-backup mode.
630 Specifies the reselection policy for the primary slave. This
631 affects how the primary slave is chosen to become the active slave
632 when failure of the active slave or recovery of the primary slave
633 occurs. This option is designed to prevent flip-flopping between
634 the primary slave and other slaves. Possible values are:
636 always or 0 (default)
638 The primary slave becomes the active slave whenever it
643 The primary slave becomes the active slave when it comes
644 back up, if the speed and duplex of the primary slave is
645 better than the speed and duplex of the current active
650 The primary slave becomes the active slave only if the
651 current active slave fails and the primary slave is up.
653 The primary_reselect setting is ignored in two cases:
655 If no slaves are active, the first slave to recover is
656 made the active slave.
658 When initially enslaved, the primary slave is always made
661 Changing the primary_reselect policy via sysfs will cause an
662 immediate selection of the best active slave according to the new
663 policy. This may or may not result in a change of the active
664 slave, depending upon the circumstances.
666 This option was added for bonding version 3.6.0.
670 Specifies the time, in milliseconds, to wait before enabling a
671 slave after a link recovery has been detected. This option is
672 only valid for the miimon link monitor. The updelay value
673 should be a multiple of the miimon value; if not, it will be
674 rounded down to the nearest multiple. The default value is 0.
678 Specifies whether or not miimon should use MII or ETHTOOL
679 ioctls vs. netif_carrier_ok() to determine the link
680 status. The MII or ETHTOOL ioctls are less efficient and
681 utilize a deprecated calling sequence within the kernel. The
682 netif_carrier_ok() relies on the device driver to maintain its
683 state with netif_carrier_on/off; at this writing, most, but
684 not all, device drivers support this facility.
686 If bonding insists that the link is up when it should not be,
687 it may be that your network device driver does not support
688 netif_carrier_on/off. The default state for netif_carrier is
689 "carrier on," so if a driver does not support netif_carrier,
690 it will appear as if the link is always up. In this case,
691 setting use_carrier to 0 will cause bonding to revert to the
692 MII / ETHTOOL ioctl method to determine the link state.
694 A value of 1 enables the use of netif_carrier_ok(), a value of
695 0 will use the deprecated MII / ETHTOOL ioctls. The default
700 Selects the transmit hash policy to use for slave selection in
701 balance-xor and 802.3ad modes. Possible values are:
705 Uses XOR of hardware MAC addresses to generate the
708 (source MAC XOR destination MAC) modulo slave count
710 This algorithm will place all traffic to a particular
711 network peer on the same slave.
713 This algorithm is 802.3ad compliant.
717 This policy uses a combination of layer2 and layer3
718 protocol information to generate the hash.
720 Uses XOR of hardware MAC addresses and IP addresses to
721 generate the hash. The IPv4 formula is
723 (((source IP XOR dest IP) AND 0xffff) XOR
724 ( source MAC XOR destination MAC ))
729 hash = (source ip quad 2 XOR dest IP quad 2) XOR
730 (source ip quad 3 XOR dest IP quad 3) XOR
731 (source ip quad 4 XOR dest IP quad 4)
733 (((hash >> 24) XOR (hash >> 16) XOR (hash >> 8) XOR hash)
734 XOR (source MAC XOR destination MAC))
737 This algorithm will place all traffic to a particular
738 network peer on the same slave. For non-IP traffic,
739 the formula is the same as for the layer2 transmit
742 This policy is intended to provide a more balanced
743 distribution of traffic than layer2 alone, especially
744 in environments where a layer3 gateway device is
745 required to reach most destinations.
747 This algorithm is 802.3ad compliant.
751 This policy uses upper layer protocol information,
752 when available, to generate the hash. This allows for
753 traffic to a particular network peer to span multiple
754 slaves, although a single connection will not span
757 The formula for unfragmented IPv4 TCP and UDP packets is
759 ((source port XOR dest port) XOR
760 ((source IP XOR dest IP) AND 0xffff)
763 The formula for unfragmented IPv6 TCP and UDP packets is
765 hash = (source port XOR dest port) XOR
766 ((source ip quad 2 XOR dest IP quad 2) XOR
767 (source ip quad 3 XOR dest IP quad 3) XOR
768 (source ip quad 4 XOR dest IP quad 4))
770 ((hash >> 24) XOR (hash >> 16) XOR (hash >> 8) XOR hash)
773 For fragmented TCP or UDP packets and all other IPv4 and
774 IPv6 protocol traffic, the source and destination port
775 information is omitted. For non-IP traffic, the
776 formula is the same as for the layer2 transmit hash
779 The IPv4 policy is intended to mimic the behavior of
780 certain switches, notably Cisco switches with PFC2 as
781 well as some Foundry and IBM products.
783 This algorithm is not fully 802.3ad compliant. A
784 single TCP or UDP conversation containing both
785 fragmented and unfragmented packets will see packets
786 striped across two interfaces. This may result in out
787 of order delivery. Most traffic types will not meet
788 this criteria, as TCP rarely fragments traffic, and
789 most UDP traffic is not involved in extended
790 conversations. Other implementations of 802.3ad may
791 or may not tolerate this noncompliance.
793 The default value is layer2. This option was added in bonding
794 version 2.6.3. In earlier versions of bonding, this parameter
795 does not exist, and the layer2 policy is the only policy. The
796 layer2+3 value was added for bonding version 3.2.2.
800 Specifies the number of IGMP membership reports to be issued after
801 a failover event. One membership report is issued immediately after
802 the failover, subsequent packets are sent in each 200ms interval.
804 The valid range is 0 - 255; the default value is 1. A value of 0
805 prevents the IGMP membership report from being issued in response
806 to the failover event.
808 This option is useful for bonding modes balance-rr (0), active-backup
809 (1), balance-tlb (5) and balance-alb (6), in which a failover can
810 switch the IGMP traffic from one slave to another. Therefore a fresh
811 IGMP report must be issued to cause the switch to forward the incoming
812 IGMP traffic over the newly selected slave.
814 This option was added for bonding version 3.7.0.
816 3. Configuring Bonding Devices
817 ==============================
819 You can configure bonding using either your distro's network
820 initialization scripts, or manually using either iproute2 or the
821 sysfs interface. Distros generally use one of three packages for the
822 network initialization scripts: initscripts, sysconfig or interfaces.
823 Recent versions of these packages have support for bonding, while older
826 We will first describe the options for configuring bonding for
827 distros using versions of initscripts, sysconfig and interfaces with full
828 or partial support for bonding, then provide information on enabling
829 bonding without support from the network initialization scripts (i.e.,
830 older versions of initscripts or sysconfig).
832 If you're unsure whether your distro uses sysconfig,
833 initscripts or interfaces, or don't know if it's new enough, have no fear.
834 Determining this is fairly straightforward.
836 First, look for a file called interfaces in /etc/network directory.
837 If this file is present in your system, then your system use interfaces. See
838 Configuration with Interfaces Support.
840 Else, issue the command:
844 It will respond with a line of text starting with either
845 "initscripts" or "sysconfig," followed by some numbers. This is the
846 package that provides your network initialization scripts.
848 Next, to determine if your installation supports bonding,
851 $ grep ifenslave /sbin/ifup
853 If this returns any matches, then your initscripts or
854 sysconfig has support for bonding.
856 3.1 Configuration with Sysconfig Support
857 ----------------------------------------
859 This section applies to distros using a version of sysconfig
860 with bonding support, for example, SuSE Linux Enterprise Server 9.
862 SuSE SLES 9's networking configuration system does support
863 bonding, however, at this writing, the YaST system configuration
864 front end does not provide any means to work with bonding devices.
865 Bonding devices can be managed by hand, however, as follows.
867 First, if they have not already been configured, configure the
868 slave devices. On SLES 9, this is most easily done by running the
869 yast2 sysconfig configuration utility. The goal is for to create an
870 ifcfg-id file for each slave device. The simplest way to accomplish
871 this is to configure the devices for DHCP (this is only to get the
872 file ifcfg-id file created; see below for some issues with DHCP). The
873 name of the configuration file for each device will be of the form:
875 ifcfg-id-xx:xx:xx:xx:xx:xx
877 Where the "xx" portion will be replaced with the digits from
878 the device's permanent MAC address.
880 Once the set of ifcfg-id-xx:xx:xx:xx:xx:xx files has been
881 created, it is necessary to edit the configuration files for the slave
882 devices (the MAC addresses correspond to those of the slave devices).
883 Before editing, the file will contain multiple lines, and will look
889 UNIQUE='XNzu.WeZGOGF+4wE'
890 _nm_name='bus-pci-0001:61:01.0'
892 Change the BOOTPROTO and STARTMODE lines to the following:
897 Do not alter the UNIQUE or _nm_name lines. Remove any other
898 lines (USERCTL, etc).
900 Once the ifcfg-id-xx:xx:xx:xx:xx:xx files have been modified,
901 it's time to create the configuration file for the bonding device
902 itself. This file is named ifcfg-bondX, where X is the number of the
903 bonding device to create, starting at 0. The first such file is
904 ifcfg-bond0, the second is ifcfg-bond1, and so on. The sysconfig
905 network configuration system will correctly start multiple instances
908 The contents of the ifcfg-bondX file is as follows:
911 BROADCAST="10.0.2.255"
913 NETMASK="255.255.0.0"
918 BONDING_MODULE_OPTS="mode=active-backup miimon=100"
919 BONDING_SLAVE0="eth0"
920 BONDING_SLAVE1="bus-pci-0000:06:08.1"
922 Replace the sample BROADCAST, IPADDR, NETMASK and NETWORK
923 values with the appropriate values for your network.
925 The STARTMODE specifies when the device is brought online.
926 The possible values are:
928 onboot: The device is started at boot time. If you're not
929 sure, this is probably what you want.
931 manual: The device is started only when ifup is called
932 manually. Bonding devices may be configured this
933 way if you do not wish them to start automatically
934 at boot for some reason.
936 hotplug: The device is started by a hotplug event. This is not
937 a valid choice for a bonding device.
939 off or ignore: The device configuration is ignored.
941 The line BONDING_MASTER='yes' indicates that the device is a
942 bonding master device. The only useful value is "yes."
944 The contents of BONDING_MODULE_OPTS are supplied to the
945 instance of the bonding module for this device. Specify the options
946 for the bonding mode, link monitoring, and so on here. Do not include
947 the max_bonds bonding parameter; this will confuse the configuration
948 system if you have multiple bonding devices.
950 Finally, supply one BONDING_SLAVEn="slave device" for each
951 slave. where "n" is an increasing value, one for each slave. The
952 "slave device" is either an interface name, e.g., "eth0", or a device
953 specifier for the network device. The interface name is easier to
954 find, but the ethN names are subject to change at boot time if, e.g.,
955 a device early in the sequence has failed. The device specifiers
956 (bus-pci-0000:06:08.1 in the example above) specify the physical
957 network device, and will not change unless the device's bus location
958 changes (for example, it is moved from one PCI slot to another). The
959 example above uses one of each type for demonstration purposes; most
960 configurations will choose one or the other for all slave devices.
962 When all configuration files have been modified or created,
963 networking must be restarted for the configuration changes to take
964 effect. This can be accomplished via the following:
966 # /etc/init.d/network restart
968 Note that the network control script (/sbin/ifdown) will
969 remove the bonding module as part of the network shutdown processing,
970 so it is not necessary to remove the module by hand if, e.g., the
971 module parameters have changed.
973 Also, at this writing, YaST/YaST2 will not manage bonding
974 devices (they do not show bonding interfaces on its list of network
975 devices). It is necessary to edit the configuration file by hand to
976 change the bonding configuration.
978 Additional general options and details of the ifcfg file
979 format can be found in an example ifcfg template file:
981 /etc/sysconfig/network/ifcfg.template
983 Note that the template does not document the various BONDING_
984 settings described above, but does describe many of the other options.
986 3.1.1 Using DHCP with Sysconfig
987 -------------------------------
989 Under sysconfig, configuring a device with BOOTPROTO='dhcp'
990 will cause it to query DHCP for its IP address information. At this
991 writing, this does not function for bonding devices; the scripts
992 attempt to obtain the device address from DHCP prior to adding any of
993 the slave devices. Without active slaves, the DHCP requests are not
996 3.1.2 Configuring Multiple Bonds with Sysconfig
997 -----------------------------------------------
999 The sysconfig network initialization system is capable of
1000 handling multiple bonding devices. All that is necessary is for each
1001 bonding instance to have an appropriately configured ifcfg-bondX file
1002 (as described above). Do not specify the "max_bonds" parameter to any
1003 instance of bonding, as this will confuse sysconfig. If you require
1004 multiple bonding devices with identical parameters, create multiple
1007 Because the sysconfig scripts supply the bonding module
1008 options in the ifcfg-bondX file, it is not necessary to add them to
1009 the system /etc/modules.d/*.conf configuration files.
1011 3.2 Configuration with Initscripts Support
1012 ------------------------------------------
1014 This section applies to distros using a recent version of
1015 initscripts with bonding support, for example, Red Hat Enterprise Linux
1016 version 3 or later, Fedora, etc. On these systems, the network
1017 initialization scripts have knowledge of bonding, and can be configured to
1018 control bonding devices. Note that older versions of the initscripts
1019 package have lower levels of support for bonding; this will be noted where
1022 These distros will not automatically load the network adapter
1023 driver unless the ethX device is configured with an IP address.
1024 Because of this constraint, users must manually configure a
1025 network-script file for all physical adapters that will be members of
1026 a bondX link. Network script files are located in the directory:
1028 /etc/sysconfig/network-scripts
1030 The file name must be prefixed with "ifcfg-eth" and suffixed
1031 with the adapter's physical adapter number. For example, the script
1032 for eth0 would be named /etc/sysconfig/network-scripts/ifcfg-eth0.
1033 Place the following text in the file:
1042 The DEVICE= line will be different for every ethX device and
1043 must correspond with the name of the file, i.e., ifcfg-eth1 must have
1044 a device line of DEVICE=eth1. The setting of the MASTER= line will
1045 also depend on the final bonding interface name chosen for your bond.
1046 As with other network devices, these typically start at 0, and go up
1047 one for each device, i.e., the first bonding instance is bond0, the
1048 second is bond1, and so on.
1050 Next, create a bond network script. The file name for this
1051 script will be /etc/sysconfig/network-scripts/ifcfg-bondX where X is
1052 the number of the bond. For bond0 the file is named "ifcfg-bond0",
1053 for bond1 it is named "ifcfg-bond1", and so on. Within that file,
1054 place the following text:
1058 NETMASK=255.255.255.0
1060 BROADCAST=192.168.1.255
1065 Be sure to change the networking specific lines (IPADDR,
1066 NETMASK, NETWORK and BROADCAST) to match your network configuration.
1068 For later versions of initscripts, such as that found with Fedora
1069 7 (or later) and Red Hat Enterprise Linux version 5 (or later), it is possible,
1070 and, indeed, preferable, to specify the bonding options in the ifcfg-bond0
1071 file, e.g. a line of the format:
1073 BONDING_OPTS="mode=active-backup arp_interval=60 arp_ip_target=192.168.1.254"
1075 will configure the bond with the specified options. The options
1076 specified in BONDING_OPTS are identical to the bonding module parameters
1077 except for the arp_ip_target field when using versions of initscripts older
1078 than and 8.57 (Fedora 8) and 8.45.19 (Red Hat Enterprise Linux 5.2). When
1079 using older versions each target should be included as a separate option and
1080 should be preceded by a '+' to indicate it should be added to the list of
1081 queried targets, e.g.,
1083 arp_ip_target=+192.168.1.1 arp_ip_target=+192.168.1.2
1085 is the proper syntax to specify multiple targets. When specifying
1086 options via BONDING_OPTS, it is not necessary to edit /etc/modprobe.d/*.conf.
1088 For even older versions of initscripts that do not support
1089 BONDING_OPTS, it is necessary to edit /etc/modprobe.d/*.conf, depending upon
1090 your distro) to load the bonding module with your desired options when the
1091 bond0 interface is brought up. The following lines in /etc/modprobe.d/*.conf
1092 will load the bonding module, and select its options:
1095 options bond0 mode=balance-alb miimon=100
1097 Replace the sample parameters with the appropriate set of
1098 options for your configuration.
1100 Finally run "/etc/rc.d/init.d/network restart" as root. This
1101 will restart the networking subsystem and your bond link should be now
1104 3.2.1 Using DHCP with Initscripts
1105 ---------------------------------
1107 Recent versions of initscripts (the versions supplied with Fedora
1108 Core 3 and Red Hat Enterprise Linux 4, or later versions, are reported to
1109 work) have support for assigning IP information to bonding devices via
1112 To configure bonding for DHCP, configure it as described
1113 above, except replace the line "BOOTPROTO=none" with "BOOTPROTO=dhcp"
1114 and add a line consisting of "TYPE=Bonding". Note that the TYPE value
1117 3.2.2 Configuring Multiple Bonds with Initscripts
1118 -------------------------------------------------
1120 Initscripts packages that are included with Fedora 7 and Red Hat
1121 Enterprise Linux 5 support multiple bonding interfaces by simply
1122 specifying the appropriate BONDING_OPTS= in ifcfg-bondX where X is the
1123 number of the bond. This support requires sysfs support in the kernel,
1124 and a bonding driver of version 3.0.0 or later. Other configurations may
1125 not support this method for specifying multiple bonding interfaces; for
1126 those instances, see the "Configuring Multiple Bonds Manually" section,
1129 3.3 Configuring Bonding Manually with iproute2
1130 -----------------------------------------------
1132 This section applies to distros whose network initialization
1133 scripts (the sysconfig or initscripts package) do not have specific
1134 knowledge of bonding. One such distro is SuSE Linux Enterprise Server
1137 The general method for these systems is to place the bonding
1138 module parameters into a config file in /etc/modprobe.d/ (as
1139 appropriate for the installed distro), then add modprobe and/or
1140 `ip link` commands to the system's global init script. The name of
1141 the global init script differs; for sysconfig, it is
1142 /etc/init.d/boot.local and for initscripts it is /etc/rc.d/rc.local.
1144 For example, if you wanted to make a simple bond of two e100
1145 devices (presumed to be eth0 and eth1), and have it persist across
1146 reboots, edit the appropriate file (/etc/init.d/boot.local or
1147 /etc/rc.d/rc.local), and add the following:
1149 modprobe bonding mode=balance-alb miimon=100
1151 ifconfig bond0 192.168.1.1 netmask 255.255.255.0 up
1152 ip link set eth0 master bond0
1153 ip link set eth1 master bond0
1155 Replace the example bonding module parameters and bond0
1156 network configuration (IP address, netmask, etc) with the appropriate
1157 values for your configuration.
1159 Unfortunately, this method will not provide support for the
1160 ifup and ifdown scripts on the bond devices. To reload the bonding
1161 configuration, it is necessary to run the initialization script, e.g.,
1163 # /etc/init.d/boot.local
1167 # /etc/rc.d/rc.local
1169 It may be desirable in such a case to create a separate script
1170 which only initializes the bonding configuration, then call that
1171 separate script from within boot.local. This allows for bonding to be
1172 enabled without re-running the entire global init script.
1174 To shut down the bonding devices, it is necessary to first
1175 mark the bonding device itself as being down, then remove the
1176 appropriate device driver modules. For our example above, you can do
1179 # ifconfig bond0 down
1183 Again, for convenience, it may be desirable to create a script
1184 with these commands.
1187 3.3.1 Configuring Multiple Bonds Manually
1188 -----------------------------------------
1190 This section contains information on configuring multiple
1191 bonding devices with differing options for those systems whose network
1192 initialization scripts lack support for configuring multiple bonds.
1194 If you require multiple bonding devices, but all with the same
1195 options, you may wish to use the "max_bonds" module parameter,
1198 To create multiple bonding devices with differing options, it is
1199 preferable to use bonding parameters exported by sysfs, documented in the
1202 For versions of bonding without sysfs support, the only means to
1203 provide multiple instances of bonding with differing options is to load
1204 the bonding driver multiple times. Note that current versions of the
1205 sysconfig network initialization scripts handle this automatically; if
1206 your distro uses these scripts, no special action is needed. See the
1207 section Configuring Bonding Devices, above, if you're not sure about your
1208 network initialization scripts.
1210 To load multiple instances of the module, it is necessary to
1211 specify a different name for each instance (the module loading system
1212 requires that every loaded module, even multiple instances of the same
1213 module, have a unique name). This is accomplished by supplying multiple
1214 sets of bonding options in /etc/modprobe.d/*.conf, for example:
1217 options bond0 -o bond0 mode=balance-rr miimon=100
1220 options bond1 -o bond1 mode=balance-alb miimon=50
1222 will load the bonding module two times. The first instance is
1223 named "bond0" and creates the bond0 device in balance-rr mode with an
1224 miimon of 100. The second instance is named "bond1" and creates the
1225 bond1 device in balance-alb mode with an miimon of 50.
1227 In some circumstances (typically with older distributions),
1228 the above does not work, and the second bonding instance never sees
1229 its options. In that case, the second options line can be substituted
1232 install bond1 /sbin/modprobe --ignore-install bonding -o bond1 \
1233 mode=balance-alb miimon=50
1235 This may be repeated any number of times, specifying a new and
1236 unique name in place of bond1 for each subsequent instance.
1238 It has been observed that some Red Hat supplied kernels are unable
1239 to rename modules at load time (the "-o bond1" part). Attempts to pass
1240 that option to modprobe will produce an "Operation not permitted" error.
1241 This has been reported on some Fedora Core kernels, and has been seen on
1242 RHEL 4 as well. On kernels exhibiting this problem, it will be impossible
1243 to configure multiple bonds with differing parameters (as they are older
1244 kernels, and also lack sysfs support).
1246 3.4 Configuring Bonding Manually via Sysfs
1247 ------------------------------------------
1249 Starting with version 3.0.0, Channel Bonding may be configured
1250 via the sysfs interface. This interface allows dynamic configuration
1251 of all bonds in the system without unloading the module. It also
1252 allows for adding and removing bonds at runtime. Ifenslave is no
1253 longer required, though it is still supported.
1255 Use of the sysfs interface allows you to use multiple bonds
1256 with different configurations without having to reload the module.
1257 It also allows you to use multiple, differently configured bonds when
1258 bonding is compiled into the kernel.
1260 You must have the sysfs filesystem mounted to configure
1261 bonding this way. The examples in this document assume that you
1262 are using the standard mount point for sysfs, e.g. /sys. If your
1263 sysfs filesystem is mounted elsewhere, you will need to adjust the
1264 example paths accordingly.
1266 Creating and Destroying Bonds
1267 -----------------------------
1268 To add a new bond foo:
1269 # echo +foo > /sys/class/net/bonding_masters
1271 To remove an existing bond bar:
1272 # echo -bar > /sys/class/net/bonding_masters
1274 To show all existing bonds:
1275 # cat /sys/class/net/bonding_masters
1277 NOTE: due to 4K size limitation of sysfs files, this list may be
1278 truncated if you have more than a few hundred bonds. This is unlikely
1279 to occur under normal operating conditions.
1281 Adding and Removing Slaves
1282 --------------------------
1283 Interfaces may be enslaved to a bond using the file
1284 /sys/class/net/<bond>/bonding/slaves. The semantics for this file
1285 are the same as for the bonding_masters file.
1287 To enslave interface eth0 to bond bond0:
1289 # echo +eth0 > /sys/class/net/bond0/bonding/slaves
1291 To free slave eth0 from bond bond0:
1292 # echo -eth0 > /sys/class/net/bond0/bonding/slaves
1294 When an interface is enslaved to a bond, symlinks between the
1295 two are created in the sysfs filesystem. In this case, you would get
1296 /sys/class/net/bond0/slave_eth0 pointing to /sys/class/net/eth0, and
1297 /sys/class/net/eth0/master pointing to /sys/class/net/bond0.
1299 This means that you can tell quickly whether or not an
1300 interface is enslaved by looking for the master symlink. Thus:
1301 # echo -eth0 > /sys/class/net/eth0/master/bonding/slaves
1302 will free eth0 from whatever bond it is enslaved to, regardless of
1303 the name of the bond interface.
1305 Changing a Bond's Configuration
1306 -------------------------------
1307 Each bond may be configured individually by manipulating the
1308 files located in /sys/class/net/<bond name>/bonding
1310 The names of these files correspond directly with the command-
1311 line parameters described elsewhere in this file, and, with the
1312 exception of arp_ip_target, they accept the same values. To see the
1313 current setting, simply cat the appropriate file.
1315 A few examples will be given here; for specific usage
1316 guidelines for each parameter, see the appropriate section in this
1319 To configure bond0 for balance-alb mode:
1320 # ifconfig bond0 down
1321 # echo 6 > /sys/class/net/bond0/bonding/mode
1323 # echo balance-alb > /sys/class/net/bond0/bonding/mode
1324 NOTE: The bond interface must be down before the mode can be
1327 To enable MII monitoring on bond0 with a 1 second interval:
1328 # echo 1000 > /sys/class/net/bond0/bonding/miimon
1329 NOTE: If ARP monitoring is enabled, it will disabled when MII
1330 monitoring is enabled, and vice-versa.
1333 # echo +192.168.0.100 > /sys/class/net/bond0/bonding/arp_ip_target
1334 # echo +192.168.0.101 > /sys/class/net/bond0/bonding/arp_ip_target
1335 NOTE: up to 16 target addresses may be specified.
1337 To remove an ARP target:
1338 # echo -192.168.0.100 > /sys/class/net/bond0/bonding/arp_ip_target
1340 Example Configuration
1341 ---------------------
1342 We begin with the same example that is shown in section 3.3,
1343 executed with sysfs, and without using ifenslave.
1345 To make a simple bond of two e100 devices (presumed to be eth0
1346 and eth1), and have it persist across reboots, edit the appropriate
1347 file (/etc/init.d/boot.local or /etc/rc.d/rc.local), and add the
1352 echo balance-alb > /sys/class/net/bond0/bonding/mode
1353 ifconfig bond0 192.168.1.1 netmask 255.255.255.0 up
1354 echo 100 > /sys/class/net/bond0/bonding/miimon
1355 echo +eth0 > /sys/class/net/bond0/bonding/slaves
1356 echo +eth1 > /sys/class/net/bond0/bonding/slaves
1358 To add a second bond, with two e1000 interfaces in
1359 active-backup mode, using ARP monitoring, add the following lines to
1363 echo +bond1 > /sys/class/net/bonding_masters
1364 echo active-backup > /sys/class/net/bond1/bonding/mode
1365 ifconfig bond1 192.168.2.1 netmask 255.255.255.0 up
1366 echo +192.168.2.100 /sys/class/net/bond1/bonding/arp_ip_target
1367 echo 2000 > /sys/class/net/bond1/bonding/arp_interval
1368 echo +eth2 > /sys/class/net/bond1/bonding/slaves
1369 echo +eth3 > /sys/class/net/bond1/bonding/slaves
1371 3.5 Configuration with Interfaces Support
1372 -----------------------------------------
1374 This section applies to distros which use /etc/network/interfaces file
1375 to describe network interface configuration, most notably Debian and it's
1378 The ifup and ifdown commands on Debian don't support bonding out of
1379 the box. The ifenslave-2.6 package should be installed to provide bonding
1380 support. Once installed, this package will provide bond-* options to be used
1381 into /etc/network/interfaces.
1383 Note that ifenslave-2.6 package will load the bonding module and use
1384 the ifenslave command when appropriate.
1386 Example Configurations
1387 ----------------------
1389 In /etc/network/interfaces, the following stanza will configure bond0, in
1390 active-backup mode, with eth0 and eth1 as slaves.
1393 iface bond0 inet dhcp
1394 bond-slaves eth0 eth1
1395 bond-mode active-backup
1397 bond-primary eth0 eth1
1399 If the above configuration doesn't work, you might have a system using
1400 upstart for system startup. This is most notably true for recent
1401 Ubuntu versions. The following stanza in /etc/network/interfaces will
1402 produce the same result on those systems.
1405 iface bond0 inet dhcp
1407 bond-mode active-backup
1411 iface eth0 inet manual
1413 bond-primary eth0 eth1
1416 iface eth1 inet manual
1418 bond-primary eth0 eth1
1420 For a full list of bond-* supported options in /etc/network/interfaces and some
1421 more advanced examples tailored to you particular distros, see the files in
1422 /usr/share/doc/ifenslave-2.6.
1424 3.6 Overriding Configuration for Special Cases
1425 ----------------------------------------------
1427 When using the bonding driver, the physical port which transmits a frame is
1428 typically selected by the bonding driver, and is not relevant to the user or
1429 system administrator. The output port is simply selected using the policies of
1430 the selected bonding mode. On occasion however, it is helpful to direct certain
1431 classes of traffic to certain physical interfaces on output to implement
1432 slightly more complex policies. For example, to reach a web server over a
1433 bonded interface in which eth0 connects to a private network, while eth1
1434 connects via a public network, it may be desirous to bias the bond to send said
1435 traffic over eth0 first, using eth1 only as a fall back, while all other traffic
1436 can safely be sent over either interface. Such configurations may be achieved
1437 using the traffic control utilities inherent in linux.
1439 By default the bonding driver is multiqueue aware and 16 queues are created
1440 when the driver initializes (see Documentation/networking/multiqueue.txt
1441 for details). If more or less queues are desired the module parameter
1442 tx_queues can be used to change this value. There is no sysfs parameter
1443 available as the allocation is done at module init time.
1445 The output of the file /proc/net/bonding/bondX has changed so the output Queue
1446 ID is now printed for each slave:
1448 Bonding Mode: fault-tolerance (active-backup)
1450 Currently Active Slave: eth0
1452 MII Polling Interval (ms): 0
1456 Slave Interface: eth0
1458 Link Failure Count: 0
1459 Permanent HW addr: 00:1a:a0:12:8f:cb
1462 Slave Interface: eth1
1464 Link Failure Count: 0
1465 Permanent HW addr: 00:1a:a0:12:8f:cc
1468 The queue_id for a slave can be set using the command:
1470 # echo "eth1:2" > /sys/class/net/bond0/bonding/queue_id
1472 Any interface that needs a queue_id set should set it with multiple calls
1473 like the one above until proper priorities are set for all interfaces. On
1474 distributions that allow configuration via initscripts, multiple 'queue_id'
1475 arguments can be added to BONDING_OPTS to set all needed slave queues.
1477 These queue id's can be used in conjunction with the tc utility to configure
1478 a multiqueue qdisc and filters to bias certain traffic to transmit on certain
1479 slave devices. For instance, say we wanted, in the above configuration to
1480 force all traffic bound to 192.168.1.100 to use eth1 in the bond as its output
1481 device. The following commands would accomplish this:
1483 # tc qdisc add dev bond0 handle 1 root multiq
1485 # tc filter add dev bond0 protocol ip parent 1: prio 1 u32 match ip dst \
1486 192.168.1.100 action skbedit queue_mapping 2
1488 These commands tell the kernel to attach a multiqueue queue discipline to the
1489 bond0 interface and filter traffic enqueued to it, such that packets with a dst
1490 ip of 192.168.1.100 have their output queue mapping value overwritten to 2.
1491 This value is then passed into the driver, causing the normal output path
1492 selection policy to be overridden, selecting instead qid 2, which maps to eth1.
1494 Note that qid values begin at 1. Qid 0 is reserved to initiate to the driver
1495 that normal output policy selection should take place. One benefit to simply
1496 leaving the qid for a slave to 0 is the multiqueue awareness in the bonding
1497 driver that is now present. This awareness allows tc filters to be placed on
1498 slave devices as well as bond devices and the bonding driver will simply act as
1499 a pass-through for selecting output queues on the slave device rather than
1500 output port selection.
1502 This feature first appeared in bonding driver version 3.7.0 and support for
1503 output slave selection was limited to round-robin and active-backup modes.
1505 4 Querying Bonding Configuration
1506 =================================
1508 4.1 Bonding Configuration
1509 -------------------------
1511 Each bonding device has a read-only file residing in the
1512 /proc/net/bonding directory. The file contents include information
1513 about the bonding configuration, options and state of each slave.
1515 For example, the contents of /proc/net/bonding/bond0 after the
1516 driver is loaded with parameters of mode=0 and miimon=1000 is
1517 generally as follows:
1519 Ethernet Channel Bonding Driver: 2.6.1 (October 29, 2004)
1520 Bonding Mode: load balancing (round-robin)
1521 Currently Active Slave: eth0
1523 MII Polling Interval (ms): 1000
1527 Slave Interface: eth1
1529 Link Failure Count: 1
1531 Slave Interface: eth0
1533 Link Failure Count: 1
1535 The precise format and contents will change depending upon the
1536 bonding configuration, state, and version of the bonding driver.
1538 4.2 Network configuration
1539 -------------------------
1541 The network configuration can be inspected using the ifconfig
1542 command. Bonding devices will have the MASTER flag set; Bonding slave
1543 devices will have the SLAVE flag set. The ifconfig output does not
1544 contain information on which slaves are associated with which masters.
1546 In the example below, the bond0 interface is the master
1547 (MASTER) while eth0 and eth1 are slaves (SLAVE). Notice all slaves of
1548 bond0 have the same MAC address (HWaddr) as bond0 for all modes except
1549 TLB and ALB that require a unique MAC address for each slave.
1552 bond0 Link encap:Ethernet HWaddr 00:C0:F0:1F:37:B4
1553 inet addr:XXX.XXX.XXX.YYY Bcast:XXX.XXX.XXX.255 Mask:255.255.252.0
1554 UP BROADCAST RUNNING MASTER MULTICAST MTU:1500 Metric:1
1555 RX packets:7224794 errors:0 dropped:0 overruns:0 frame:0
1556 TX packets:3286647 errors:1 dropped:0 overruns:1 carrier:0
1557 collisions:0 txqueuelen:0
1559 eth0 Link encap:Ethernet HWaddr 00:C0:F0:1F:37:B4
1560 UP BROADCAST RUNNING SLAVE MULTICAST MTU:1500 Metric:1
1561 RX packets:3573025 errors:0 dropped:0 overruns:0 frame:0
1562 TX packets:1643167 errors:1 dropped:0 overruns:1 carrier:0
1563 collisions:0 txqueuelen:100
1564 Interrupt:10 Base address:0x1080
1566 eth1 Link encap:Ethernet HWaddr 00:C0:F0:1F:37:B4
1567 UP BROADCAST RUNNING SLAVE MULTICAST MTU:1500 Metric:1
1568 RX packets:3651769 errors:0 dropped:0 overruns:0 frame:0
1569 TX packets:1643480 errors:0 dropped:0 overruns:0 carrier:0
1570 collisions:0 txqueuelen:100
1571 Interrupt:9 Base address:0x1400
1573 5. Switch Configuration
1574 =======================
1576 For this section, "switch" refers to whatever system the
1577 bonded devices are directly connected to (i.e., where the other end of
1578 the cable plugs into). This may be an actual dedicated switch device,
1579 or it may be another regular system (e.g., another computer running
1582 The active-backup, balance-tlb and balance-alb modes do not
1583 require any specific configuration of the switch.
1585 The 802.3ad mode requires that the switch have the appropriate
1586 ports configured as an 802.3ad aggregation. The precise method used
1587 to configure this varies from switch to switch, but, for example, a
1588 Cisco 3550 series switch requires that the appropriate ports first be
1589 grouped together in a single etherchannel instance, then that
1590 etherchannel is set to mode "lacp" to enable 802.3ad (instead of
1591 standard EtherChannel).
1593 The balance-rr, balance-xor and broadcast modes generally
1594 require that the switch have the appropriate ports grouped together.
1595 The nomenclature for such a group differs between switches, it may be
1596 called an "etherchannel" (as in the Cisco example, above), a "trunk
1597 group" or some other similar variation. For these modes, each switch
1598 will also have its own configuration options for the switch's transmit
1599 policy to the bond. Typical choices include XOR of either the MAC or
1600 IP addresses. The transmit policy of the two peers does not need to
1601 match. For these three modes, the bonding mode really selects a
1602 transmit policy for an EtherChannel group; all three will interoperate
1603 with another EtherChannel group.
1606 6. 802.1q VLAN Support
1607 ======================
1609 It is possible to configure VLAN devices over a bond interface
1610 using the 8021q driver. However, only packets coming from the 8021q
1611 driver and passing through bonding will be tagged by default. Self
1612 generated packets, for example, bonding's learning packets or ARP
1613 packets generated by either ALB mode or the ARP monitor mechanism, are
1614 tagged internally by bonding itself. As a result, bonding must
1615 "learn" the VLAN IDs configured above it, and use those IDs to tag
1616 self generated packets.
1618 For reasons of simplicity, and to support the use of adapters
1619 that can do VLAN hardware acceleration offloading, the bonding
1620 interface declares itself as fully hardware offloading capable, it gets
1621 the add_vid/kill_vid notifications to gather the necessary
1622 information, and it propagates those actions to the slaves. In case
1623 of mixed adapter types, hardware accelerated tagged packets that
1624 should go through an adapter that is not offloading capable are
1625 "un-accelerated" by the bonding driver so the VLAN tag sits in the
1628 VLAN interfaces *must* be added on top of a bonding interface
1629 only after enslaving at least one slave. The bonding interface has a
1630 hardware address of 00:00:00:00:00:00 until the first slave is added.
1631 If the VLAN interface is created prior to the first enslavement, it
1632 would pick up the all-zeroes hardware address. Once the first slave
1633 is attached to the bond, the bond device itself will pick up the
1634 slave's hardware address, which is then available for the VLAN device.
1636 Also, be aware that a similar problem can occur if all slaves
1637 are released from a bond that still has one or more VLAN interfaces on
1638 top of it. When a new slave is added, the bonding interface will
1639 obtain its hardware address from the first slave, which might not
1640 match the hardware address of the VLAN interfaces (which was
1641 ultimately copied from an earlier slave).
1643 There are two methods to insure that the VLAN device operates
1644 with the correct hardware address if all slaves are removed from a
1647 1. Remove all VLAN interfaces then recreate them
1649 2. Set the bonding interface's hardware address so that it
1650 matches the hardware address of the VLAN interfaces.
1652 Note that changing a VLAN interface's HW address would set the
1653 underlying device -- i.e. the bonding interface -- to promiscuous
1654 mode, which might not be what you want.
1660 The bonding driver at present supports two schemes for
1661 monitoring a slave device's link state: the ARP monitor and the MII
1664 At the present time, due to implementation restrictions in the
1665 bonding driver itself, it is not possible to enable both ARP and MII
1666 monitoring simultaneously.
1668 7.1 ARP Monitor Operation
1669 -------------------------
1671 The ARP monitor operates as its name suggests: it sends ARP
1672 queries to one or more designated peer systems on the network, and
1673 uses the response as an indication that the link is operating. This
1674 gives some assurance that traffic is actually flowing to and from one
1675 or more peers on the local network.
1677 The ARP monitor relies on the device driver itself to verify
1678 that traffic is flowing. In particular, the driver must keep up to
1679 date the last receive time, dev->last_rx, and transmit start time,
1680 dev->trans_start. If these are not updated by the driver, then the
1681 ARP monitor will immediately fail any slaves using that driver, and
1682 those slaves will stay down. If networking monitoring (tcpdump, etc)
1683 shows the ARP requests and replies on the network, then it may be that
1684 your device driver is not updating last_rx and trans_start.
1686 7.2 Configuring Multiple ARP Targets
1687 ------------------------------------
1689 While ARP monitoring can be done with just one target, it can
1690 be useful in a High Availability setup to have several targets to
1691 monitor. In the case of just one target, the target itself may go
1692 down or have a problem making it unresponsive to ARP requests. Having
1693 an additional target (or several) increases the reliability of the ARP
1696 Multiple ARP targets must be separated by commas as follows:
1698 # example options for ARP monitoring with three targets
1700 options bond0 arp_interval=60 arp_ip_target=192.168.0.1,192.168.0.3,192.168.0.9
1702 For just a single target the options would resemble:
1704 # example options for ARP monitoring with one target
1706 options bond0 arp_interval=60 arp_ip_target=192.168.0.100
1709 7.3 MII Monitor Operation
1710 -------------------------
1712 The MII monitor monitors only the carrier state of the local
1713 network interface. It accomplishes this in one of three ways: by
1714 depending upon the device driver to maintain its carrier state, by
1715 querying the device's MII registers, or by making an ethtool query to
1718 If the use_carrier module parameter is 1 (the default value),
1719 then the MII monitor will rely on the driver for carrier state
1720 information (via the netif_carrier subsystem). As explained in the
1721 use_carrier parameter information, above, if the MII monitor fails to
1722 detect carrier loss on the device (e.g., when the cable is physically
1723 disconnected), it may be that the driver does not support
1726 If use_carrier is 0, then the MII monitor will first query the
1727 device's (via ioctl) MII registers and check the link state. If that
1728 request fails (not just that it returns carrier down), then the MII
1729 monitor will make an ethtool ETHOOL_GLINK request to attempt to obtain
1730 the same information. If both methods fail (i.e., the driver either
1731 does not support or had some error in processing both the MII register
1732 and ethtool requests), then the MII monitor will assume the link is
1735 8. Potential Sources of Trouble
1736 ===============================
1738 8.1 Adventures in Routing
1739 -------------------------
1741 When bonding is configured, it is important that the slave
1742 devices not have routes that supersede routes of the master (or,
1743 generally, not have routes at all). For example, suppose the bonding
1744 device bond0 has two slaves, eth0 and eth1, and the routing table is
1747 Kernel IP routing table
1748 Destination Gateway Genmask Flags MSS Window irtt Iface
1749 10.0.0.0 0.0.0.0 255.255.0.0 U 40 0 0 eth0
1750 10.0.0.0 0.0.0.0 255.255.0.0 U 40 0 0 eth1
1751 10.0.0.0 0.0.0.0 255.255.0.0 U 40 0 0 bond0
1752 127.0.0.0 0.0.0.0 255.0.0.0 U 40 0 0 lo
1754 This routing configuration will likely still update the
1755 receive/transmit times in the driver (needed by the ARP monitor), but
1756 may bypass the bonding driver (because outgoing traffic to, in this
1757 case, another host on network 10 would use eth0 or eth1 before bond0).
1759 The ARP monitor (and ARP itself) may become confused by this
1760 configuration, because ARP requests (generated by the ARP monitor)
1761 will be sent on one interface (bond0), but the corresponding reply
1762 will arrive on a different interface (eth0). This reply looks to ARP
1763 as an unsolicited ARP reply (because ARP matches replies on an
1764 interface basis), and is discarded. The MII monitor is not affected
1765 by the state of the routing table.
1767 The solution here is simply to insure that slaves do not have
1768 routes of their own, and if for some reason they must, those routes do
1769 not supersede routes of their master. This should generally be the
1770 case, but unusual configurations or errant manual or automatic static
1771 route additions may cause trouble.
1773 8.2 Ethernet Device Renaming
1774 ----------------------------
1776 On systems with network configuration scripts that do not
1777 associate physical devices directly with network interface names (so
1778 that the same physical device always has the same "ethX" name), it may
1779 be necessary to add some special logic to config files in
1782 For example, given a modules.conf containing the following:
1785 options bond0 mode=some-mode miimon=50
1791 If neither eth0 and eth1 are slaves to bond0, then when the
1792 bond0 interface comes up, the devices may end up reordered. This
1793 happens because bonding is loaded first, then its slave device's
1794 drivers are loaded next. Since no other drivers have been loaded,
1795 when the e1000 driver loads, it will receive eth0 and eth1 for its
1796 devices, but the bonding configuration tries to enslave eth2 and eth3
1797 (which may later be assigned to the tg3 devices).
1799 Adding the following:
1801 add above bonding e1000 tg3
1803 causes modprobe to load e1000 then tg3, in that order, when
1804 bonding is loaded. This command is fully documented in the
1805 modules.conf manual page.
1807 On systems utilizing modprobe an equivalent problem can occur.
1808 In this case, the following can be added to config files in
1809 /etc/modprobe.d/ as:
1811 softdep bonding pre: tg3 e1000
1813 This will load tg3 and e1000 modules before loading the bonding one.
1814 Full documentation on this can be found in the modprobe.d and modprobe
1817 8.3. Painfully Slow Or No Failed Link Detection By Miimon
1818 ---------------------------------------------------------
1820 By default, bonding enables the use_carrier option, which
1821 instructs bonding to trust the driver to maintain carrier state.
1823 As discussed in the options section, above, some drivers do
1824 not support the netif_carrier_on/_off link state tracking system.
1825 With use_carrier enabled, bonding will always see these links as up,
1826 regardless of their actual state.
1828 Additionally, other drivers do support netif_carrier, but do
1829 not maintain it in real time, e.g., only polling the link state at
1830 some fixed interval. In this case, miimon will detect failures, but
1831 only after some long period of time has expired. If it appears that
1832 miimon is very slow in detecting link failures, try specifying
1833 use_carrier=0 to see if that improves the failure detection time. If
1834 it does, then it may be that the driver checks the carrier state at a
1835 fixed interval, but does not cache the MII register values (so the
1836 use_carrier=0 method of querying the registers directly works). If
1837 use_carrier=0 does not improve the failover, then the driver may cache
1838 the registers, or the problem may be elsewhere.
1840 Also, remember that miimon only checks for the device's
1841 carrier state. It has no way to determine the state of devices on or
1842 beyond other ports of a switch, or if a switch is refusing to pass
1843 traffic while still maintaining carrier on.
1848 If running SNMP agents, the bonding driver should be loaded
1849 before any network drivers participating in a bond. This requirement
1850 is due to the interface index (ipAdEntIfIndex) being associated to
1851 the first interface found with a given IP address. That is, there is
1852 only one ipAdEntIfIndex for each IP address. For example, if eth0 and
1853 eth1 are slaves of bond0 and the driver for eth0 is loaded before the
1854 bonding driver, the interface for the IP address will be associated
1855 with the eth0 interface. This configuration is shown below, the IP
1856 address 192.168.1.1 has an interface index of 2 which indexes to eth0
1857 in the ifDescr table (ifDescr.2).
1859 interfaces.ifTable.ifEntry.ifDescr.1 = lo
1860 interfaces.ifTable.ifEntry.ifDescr.2 = eth0
1861 interfaces.ifTable.ifEntry.ifDescr.3 = eth1
1862 interfaces.ifTable.ifEntry.ifDescr.4 = eth2
1863 interfaces.ifTable.ifEntry.ifDescr.5 = eth3
1864 interfaces.ifTable.ifEntry.ifDescr.6 = bond0
1865 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.10.10.10 = 5
1866 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.192.168.1.1 = 2
1867 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.74.20.94 = 4
1868 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.127.0.0.1 = 1
1870 This problem is avoided by loading the bonding driver before
1871 any network drivers participating in a bond. Below is an example of
1872 loading the bonding driver first, the IP address 192.168.1.1 is
1873 correctly associated with ifDescr.2.
1875 interfaces.ifTable.ifEntry.ifDescr.1 = lo
1876 interfaces.ifTable.ifEntry.ifDescr.2 = bond0
1877 interfaces.ifTable.ifEntry.ifDescr.3 = eth0
1878 interfaces.ifTable.ifEntry.ifDescr.4 = eth1
1879 interfaces.ifTable.ifEntry.ifDescr.5 = eth2
1880 interfaces.ifTable.ifEntry.ifDescr.6 = eth3
1881 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.10.10.10 = 6
1882 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.192.168.1.1 = 2
1883 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.74.20.94 = 5
1884 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.127.0.0.1 = 1
1886 While some distributions may not report the interface name in
1887 ifDescr, the association between the IP address and IfIndex remains
1888 and SNMP functions such as Interface_Scan_Next will report that
1891 10. Promiscuous mode
1892 ====================
1894 When running network monitoring tools, e.g., tcpdump, it is
1895 common to enable promiscuous mode on the device, so that all traffic
1896 is seen (instead of seeing only traffic destined for the local host).
1897 The bonding driver handles promiscuous mode changes to the bonding
1898 master device (e.g., bond0), and propagates the setting to the slave
1901 For the balance-rr, balance-xor, broadcast, and 802.3ad modes,
1902 the promiscuous mode setting is propagated to all slaves.
1904 For the active-backup, balance-tlb and balance-alb modes, the
1905 promiscuous mode setting is propagated only to the active slave.
1907 For balance-tlb mode, the active slave is the slave currently
1908 receiving inbound traffic.
1910 For balance-alb mode, the active slave is the slave used as a
1911 "primary." This slave is used for mode-specific control traffic, for
1912 sending to peers that are unassigned or if the load is unbalanced.
1914 For the active-backup, balance-tlb and balance-alb modes, when
1915 the active slave changes (e.g., due to a link failure), the
1916 promiscuous setting will be propagated to the new active slave.
1918 11. Configuring Bonding for High Availability
1919 =============================================
1921 High Availability refers to configurations that provide
1922 maximum network availability by having redundant or backup devices,
1923 links or switches between the host and the rest of the world. The
1924 goal is to provide the maximum availability of network connectivity
1925 (i.e., the network always works), even though other configurations
1926 could provide higher throughput.
1928 11.1 High Availability in a Single Switch Topology
1929 --------------------------------------------------
1931 If two hosts (or a host and a single switch) are directly
1932 connected via multiple physical links, then there is no availability
1933 penalty to optimizing for maximum bandwidth. In this case, there is
1934 only one switch (or peer), so if it fails, there is no alternative
1935 access to fail over to. Additionally, the bonding load balance modes
1936 support link monitoring of their members, so if individual links fail,
1937 the load will be rebalanced across the remaining devices.
1939 See Section 12, "Configuring Bonding for Maximum Throughput"
1940 for information on configuring bonding with one peer device.
1942 11.2 High Availability in a Multiple Switch Topology
1943 ----------------------------------------------------
1945 With multiple switches, the configuration of bonding and the
1946 network changes dramatically. In multiple switch topologies, there is
1947 a trade off between network availability and usable bandwidth.
1949 Below is a sample network, configured to maximize the
1950 availability of the network:
1954 +-----+----+ +-----+----+
1955 | |port2 ISL port2| |
1956 | switch A +--------------------------+ switch B |
1958 +-----+----+ +-----++---+
1961 +-------------+ host1 +---------------+
1964 In this configuration, there is a link between the two
1965 switches (ISL, or inter switch link), and multiple ports connecting to
1966 the outside world ("port3" on each switch). There is no technical
1967 reason that this could not be extended to a third switch.
1969 11.2.1 HA Bonding Mode Selection for Multiple Switch Topology
1970 -------------------------------------------------------------
1972 In a topology such as the example above, the active-backup and
1973 broadcast modes are the only useful bonding modes when optimizing for
1974 availability; the other modes require all links to terminate on the
1975 same peer for them to behave rationally.
1977 active-backup: This is generally the preferred mode, particularly if
1978 the switches have an ISL and play together well. If the
1979 network configuration is such that one switch is specifically
1980 a backup switch (e.g., has lower capacity, higher cost, etc),
1981 then the primary option can be used to insure that the
1982 preferred link is always used when it is available.
1984 broadcast: This mode is really a special purpose mode, and is suitable
1985 only for very specific needs. For example, if the two
1986 switches are not connected (no ISL), and the networks beyond
1987 them are totally independent. In this case, if it is
1988 necessary for some specific one-way traffic to reach both
1989 independent networks, then the broadcast mode may be suitable.
1991 11.2.2 HA Link Monitoring Selection for Multiple Switch Topology
1992 ----------------------------------------------------------------
1994 The choice of link monitoring ultimately depends upon your
1995 switch. If the switch can reliably fail ports in response to other
1996 failures, then either the MII or ARP monitors should work. For
1997 example, in the above example, if the "port3" link fails at the remote
1998 end, the MII monitor has no direct means to detect this. The ARP
1999 monitor could be configured with a target at the remote end of port3,
2000 thus detecting that failure without switch support.
2002 In general, however, in a multiple switch topology, the ARP
2003 monitor can provide a higher level of reliability in detecting end to
2004 end connectivity failures (which may be caused by the failure of any
2005 individual component to pass traffic for any reason). Additionally,
2006 the ARP monitor should be configured with multiple targets (at least
2007 one for each switch in the network). This will insure that,
2008 regardless of which switch is active, the ARP monitor has a suitable
2011 Note, also, that of late many switches now support a functionality
2012 generally referred to as "trunk failover." This is a feature of the
2013 switch that causes the link state of a particular switch port to be set
2014 down (or up) when the state of another switch port goes down (or up).
2015 Its purpose is to propagate link failures from logically "exterior" ports
2016 to the logically "interior" ports that bonding is able to monitor via
2017 miimon. Availability and configuration for trunk failover varies by
2018 switch, but this can be a viable alternative to the ARP monitor when using
2021 12. Configuring Bonding for Maximum Throughput
2022 ==============================================
2024 12.1 Maximizing Throughput in a Single Switch Topology
2025 ------------------------------------------------------
2027 In a single switch configuration, the best method to maximize
2028 throughput depends upon the application and network environment. The
2029 various load balancing modes each have strengths and weaknesses in
2030 different environments, as detailed below.
2032 For this discussion, we will break down the topologies into
2033 two categories. Depending upon the destination of most traffic, we
2034 categorize them into either "gatewayed" or "local" configurations.
2036 In a gatewayed configuration, the "switch" is acting primarily
2037 as a router, and the majority of traffic passes through this router to
2038 other networks. An example would be the following:
2041 +----------+ +----------+
2042 | |eth0 port1| | to other networks
2043 | Host A +---------------------+ router +------------------->
2044 | +---------------------+ | Hosts B and C are out
2045 | |eth1 port2| | here somewhere
2046 +----------+ +----------+
2048 The router may be a dedicated router device, or another host
2049 acting as a gateway. For our discussion, the important point is that
2050 the majority of traffic from Host A will pass through the router to
2051 some other network before reaching its final destination.
2053 In a gatewayed network configuration, although Host A may
2054 communicate with many other systems, all of its traffic will be sent
2055 and received via one other peer on the local network, the router.
2057 Note that the case of two systems connected directly via
2058 multiple physical links is, for purposes of configuring bonding, the
2059 same as a gatewayed configuration. In that case, it happens that all
2060 traffic is destined for the "gateway" itself, not some other network
2063 In a local configuration, the "switch" is acting primarily as
2064 a switch, and the majority of traffic passes through this switch to
2065 reach other stations on the same network. An example would be the
2068 +----------+ +----------+ +--------+
2069 | |eth0 port1| +-------+ Host B |
2070 | Host A +------------+ switch |port3 +--------+
2071 | +------------+ | +--------+
2072 | |eth1 port2| +------------------+ Host C |
2073 +----------+ +----------+port4 +--------+
2076 Again, the switch may be a dedicated switch device, or another
2077 host acting as a gateway. For our discussion, the important point is
2078 that the majority of traffic from Host A is destined for other hosts
2079 on the same local network (Hosts B and C in the above example).
2081 In summary, in a gatewayed configuration, traffic to and from
2082 the bonded device will be to the same MAC level peer on the network
2083 (the gateway itself, i.e., the router), regardless of its final
2084 destination. In a local configuration, traffic flows directly to and
2085 from the final destinations, thus, each destination (Host B, Host C)
2086 will be addressed directly by their individual MAC addresses.
2088 This distinction between a gatewayed and a local network
2089 configuration is important because many of the load balancing modes
2090 available use the MAC addresses of the local network source and
2091 destination to make load balancing decisions. The behavior of each
2092 mode is described below.
2095 12.1.1 MT Bonding Mode Selection for Single Switch Topology
2096 -----------------------------------------------------------
2098 This configuration is the easiest to set up and to understand,
2099 although you will have to decide which bonding mode best suits your
2100 needs. The trade offs for each mode are detailed below:
2102 balance-rr: This mode is the only mode that will permit a single
2103 TCP/IP connection to stripe traffic across multiple
2104 interfaces. It is therefore the only mode that will allow a
2105 single TCP/IP stream to utilize more than one interface's
2106 worth of throughput. This comes at a cost, however: the
2107 striping generally results in peer systems receiving packets out
2108 of order, causing TCP/IP's congestion control system to kick
2109 in, often by retransmitting segments.
2111 It is possible to adjust TCP/IP's congestion limits by
2112 altering the net.ipv4.tcp_reordering sysctl parameter. The
2113 usual default value is 3, and the maximum useful value is 127.
2114 For a four interface balance-rr bond, expect that a single
2115 TCP/IP stream will utilize no more than approximately 2.3
2116 interface's worth of throughput, even after adjusting
2119 Note that the fraction of packets that will be delivered out of
2120 order is highly variable, and is unlikely to be zero. The level
2121 of reordering depends upon a variety of factors, including the
2122 networking interfaces, the switch, and the topology of the
2123 configuration. Speaking in general terms, higher speed network
2124 cards produce more reordering (due to factors such as packet
2125 coalescing), and a "many to many" topology will reorder at a
2126 higher rate than a "many slow to one fast" configuration.
2128 Many switches do not support any modes that stripe traffic
2129 (instead choosing a port based upon IP or MAC level addresses);
2130 for those devices, traffic for a particular connection flowing
2131 through the switch to a balance-rr bond will not utilize greater
2132 than one interface's worth of bandwidth.
2134 If you are utilizing protocols other than TCP/IP, UDP for
2135 example, and your application can tolerate out of order
2136 delivery, then this mode can allow for single stream datagram
2137 performance that scales near linearly as interfaces are added
2140 This mode requires the switch to have the appropriate ports
2141 configured for "etherchannel" or "trunking."
2143 active-backup: There is not much advantage in this network topology to
2144 the active-backup mode, as the inactive backup devices are all
2145 connected to the same peer as the primary. In this case, a
2146 load balancing mode (with link monitoring) will provide the
2147 same level of network availability, but with increased
2148 available bandwidth. On the plus side, active-backup mode
2149 does not require any configuration of the switch, so it may
2150 have value if the hardware available does not support any of
2151 the load balance modes.
2153 balance-xor: This mode will limit traffic such that packets destined
2154 for specific peers will always be sent over the same
2155 interface. Since the destination is determined by the MAC
2156 addresses involved, this mode works best in a "local" network
2157 configuration (as described above), with destinations all on
2158 the same local network. This mode is likely to be suboptimal
2159 if all your traffic is passed through a single router (i.e., a
2160 "gatewayed" network configuration, as described above).
2162 As with balance-rr, the switch ports need to be configured for
2163 "etherchannel" or "trunking."
2165 broadcast: Like active-backup, there is not much advantage to this
2166 mode in this type of network topology.
2168 802.3ad: This mode can be a good choice for this type of network
2169 topology. The 802.3ad mode is an IEEE standard, so all peers
2170 that implement 802.3ad should interoperate well. The 802.3ad
2171 protocol includes automatic configuration of the aggregates,
2172 so minimal manual configuration of the switch is needed
2173 (typically only to designate that some set of devices is
2174 available for 802.3ad). The 802.3ad standard also mandates
2175 that frames be delivered in order (within certain limits), so
2176 in general single connections will not see misordering of
2177 packets. The 802.3ad mode does have some drawbacks: the
2178 standard mandates that all devices in the aggregate operate at
2179 the same speed and duplex. Also, as with all bonding load
2180 balance modes other than balance-rr, no single connection will
2181 be able to utilize more than a single interface's worth of
2184 Additionally, the linux bonding 802.3ad implementation
2185 distributes traffic by peer (using an XOR of MAC addresses),
2186 so in a "gatewayed" configuration, all outgoing traffic will
2187 generally use the same device. Incoming traffic may also end
2188 up on a single device, but that is dependent upon the
2189 balancing policy of the peer's 8023.ad implementation. In a
2190 "local" configuration, traffic will be distributed across the
2191 devices in the bond.
2193 Finally, the 802.3ad mode mandates the use of the MII monitor,
2194 therefore, the ARP monitor is not available in this mode.
2196 balance-tlb: The balance-tlb mode balances outgoing traffic by peer.
2197 Since the balancing is done according to MAC address, in a
2198 "gatewayed" configuration (as described above), this mode will
2199 send all traffic across a single device. However, in a
2200 "local" network configuration, this mode balances multiple
2201 local network peers across devices in a vaguely intelligent
2202 manner (not a simple XOR as in balance-xor or 802.3ad mode),
2203 so that mathematically unlucky MAC addresses (i.e., ones that
2204 XOR to the same value) will not all "bunch up" on a single
2207 Unlike 802.3ad, interfaces may be of differing speeds, and no
2208 special switch configuration is required. On the down side,
2209 in this mode all incoming traffic arrives over a single
2210 interface, this mode requires certain ethtool support in the
2211 network device driver of the slave interfaces, and the ARP
2212 monitor is not available.
2214 balance-alb: This mode is everything that balance-tlb is, and more.
2215 It has all of the features (and restrictions) of balance-tlb,
2216 and will also balance incoming traffic from local network
2217 peers (as described in the Bonding Module Options section,
2220 The only additional down side to this mode is that the network
2221 device driver must support changing the hardware address while
2224 12.1.2 MT Link Monitoring for Single Switch Topology
2225 ----------------------------------------------------
2227 The choice of link monitoring may largely depend upon which
2228 mode you choose to use. The more advanced load balancing modes do not
2229 support the use of the ARP monitor, and are thus restricted to using
2230 the MII monitor (which does not provide as high a level of end to end
2231 assurance as the ARP monitor).
2233 12.2 Maximum Throughput in a Multiple Switch Topology
2234 -----------------------------------------------------
2236 Multiple switches may be utilized to optimize for throughput
2237 when they are configured in parallel as part of an isolated network
2238 between two or more systems, for example:
2244 +--------+ | +---------+
2246 +------+---+ +-----+----+ +-----+----+
2247 | Switch A | | Switch B | | Switch C |
2248 +------+---+ +-----+----+ +-----+----+
2250 +--------+ | +---------+
2256 In this configuration, the switches are isolated from one
2257 another. One reason to employ a topology such as this is for an
2258 isolated network with many hosts (a cluster configured for high
2259 performance, for example), using multiple smaller switches can be more
2260 cost effective than a single larger switch, e.g., on a network with 24
2261 hosts, three 24 port switches can be significantly less expensive than
2262 a single 72 port switch.
2264 If access beyond the network is required, an individual host
2265 can be equipped with an additional network device connected to an
2266 external network; this host then additionally acts as a gateway.
2268 12.2.1 MT Bonding Mode Selection for Multiple Switch Topology
2269 -------------------------------------------------------------
2271 In actual practice, the bonding mode typically employed in
2272 configurations of this type is balance-rr. Historically, in this
2273 network configuration, the usual caveats about out of order packet
2274 delivery are mitigated by the use of network adapters that do not do
2275 any kind of packet coalescing (via the use of NAPI, or because the
2276 device itself does not generate interrupts until some number of
2277 packets has arrived). When employed in this fashion, the balance-rr
2278 mode allows individual connections between two hosts to effectively
2279 utilize greater than one interface's bandwidth.
2281 12.2.2 MT Link Monitoring for Multiple Switch Topology
2282 ------------------------------------------------------
2284 Again, in actual practice, the MII monitor is most often used
2285 in this configuration, as performance is given preference over
2286 availability. The ARP monitor will function in this topology, but its
2287 advantages over the MII monitor are mitigated by the volume of probes
2288 needed as the number of systems involved grows (remember that each
2289 host in the network is configured with bonding).
2291 13. Switch Behavior Issues
2292 ==========================
2294 13.1 Link Establishment and Failover Delays
2295 -------------------------------------------
2297 Some switches exhibit undesirable behavior with regard to the
2298 timing of link up and down reporting by the switch.
2300 First, when a link comes up, some switches may indicate that
2301 the link is up (carrier available), but not pass traffic over the
2302 interface for some period of time. This delay is typically due to
2303 some type of autonegotiation or routing protocol, but may also occur
2304 during switch initialization (e.g., during recovery after a switch
2305 failure). If you find this to be a problem, specify an appropriate
2306 value to the updelay bonding module option to delay the use of the
2307 relevant interface(s).
2309 Second, some switches may "bounce" the link state one or more
2310 times while a link is changing state. This occurs most commonly while
2311 the switch is initializing. Again, an appropriate updelay value may
2314 Note that when a bonding interface has no active links, the
2315 driver will immediately reuse the first link that goes up, even if the
2316 updelay parameter has been specified (the updelay is ignored in this
2317 case). If there are slave interfaces waiting for the updelay timeout
2318 to expire, the interface that first went into that state will be
2319 immediately reused. This reduces down time of the network if the
2320 value of updelay has been overestimated, and since this occurs only in
2321 cases with no connectivity, there is no additional penalty for
2322 ignoring the updelay.
2324 In addition to the concerns about switch timings, if your
2325 switches take a long time to go into backup mode, it may be desirable
2326 to not activate a backup interface immediately after a link goes down.
2327 Failover may be delayed via the downdelay bonding module option.
2329 13.2 Duplicated Incoming Packets
2330 --------------------------------
2332 NOTE: Starting with version 3.0.2, the bonding driver has logic to
2333 suppress duplicate packets, which should largely eliminate this problem.
2334 The following description is kept for reference.
2336 It is not uncommon to observe a short burst of duplicated
2337 traffic when the bonding device is first used, or after it has been
2338 idle for some period of time. This is most easily observed by issuing
2339 a "ping" to some other host on the network, and noticing that the
2340 output from ping flags duplicates (typically one per slave).
2342 For example, on a bond in active-backup mode with five slaves
2343 all connected to one switch, the output may appear as follows:
2346 PING 10.0.4.2 (10.0.4.2) from 10.0.3.10 : 56(84) bytes of data.
2347 64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.7 ms
2348 64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
2349 64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
2350 64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
2351 64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
2352 64 bytes from 10.0.4.2: icmp_seq=2 ttl=64 time=0.216 ms
2353 64 bytes from 10.0.4.2: icmp_seq=3 ttl=64 time=0.267 ms
2354 64 bytes from 10.0.4.2: icmp_seq=4 ttl=64 time=0.222 ms
2356 This is not due to an error in the bonding driver, rather, it
2357 is a side effect of how many switches update their MAC forwarding
2358 tables. Initially, the switch does not associate the MAC address in
2359 the packet with a particular switch port, and so it may send the
2360 traffic to all ports until its MAC forwarding table is updated. Since
2361 the interfaces attached to the bond may occupy multiple ports on a
2362 single switch, when the switch (temporarily) floods the traffic to all
2363 ports, the bond device receives multiple copies of the same packet
2364 (one per slave device).
2366 The duplicated packet behavior is switch dependent, some
2367 switches exhibit this, and some do not. On switches that display this
2368 behavior, it can be induced by clearing the MAC forwarding table (on
2369 most Cisco switches, the privileged command "clear mac address-table
2370 dynamic" will accomplish this).
2372 14. Hardware Specific Considerations
2373 ====================================
2375 This section contains additional information for configuring
2376 bonding on specific hardware platforms, or for interfacing bonding
2377 with particular switches or other devices.
2379 14.1 IBM BladeCenter
2380 --------------------
2382 This applies to the JS20 and similar systems.
2384 On the JS20 blades, the bonding driver supports only
2385 balance-rr, active-backup, balance-tlb and balance-alb modes. This is
2386 largely due to the network topology inside the BladeCenter, detailed
2389 JS20 network adapter information
2390 --------------------------------
2392 All JS20s come with two Broadcom Gigabit Ethernet ports
2393 integrated on the planar (that's "motherboard" in IBM-speak). In the
2394 BladeCenter chassis, the eth0 port of all JS20 blades is hard wired to
2395 I/O Module #1; similarly, all eth1 ports are wired to I/O Module #2.
2396 An add-on Broadcom daughter card can be installed on a JS20 to provide
2397 two more Gigabit Ethernet ports. These ports, eth2 and eth3, are
2398 wired to I/O Modules 3 and 4, respectively.
2400 Each I/O Module may contain either a switch or a passthrough
2401 module (which allows ports to be directly connected to an external
2402 switch). Some bonding modes require a specific BladeCenter internal
2403 network topology in order to function; these are detailed below.
2405 Additional BladeCenter-specific networking information can be
2406 found in two IBM Redbooks (www.ibm.com/redbooks):
2408 "IBM eServer BladeCenter Networking Options"
2409 "IBM eServer BladeCenter Layer 2-7 Network Switching"
2411 BladeCenter networking configuration
2412 ------------------------------------
2414 Because a BladeCenter can be configured in a very large number
2415 of ways, this discussion will be confined to describing basic
2418 Normally, Ethernet Switch Modules (ESMs) are used in I/O
2419 modules 1 and 2. In this configuration, the eth0 and eth1 ports of a
2420 JS20 will be connected to different internal switches (in the
2421 respective I/O modules).
2423 A passthrough module (OPM or CPM, optical or copper,
2424 passthrough module) connects the I/O module directly to an external
2425 switch. By using PMs in I/O module #1 and #2, the eth0 and eth1
2426 interfaces of a JS20 can be redirected to the outside world and
2427 connected to a common external switch.
2429 Depending upon the mix of ESMs and PMs, the network will
2430 appear to bonding as either a single switch topology (all PMs) or as a
2431 multiple switch topology (one or more ESMs, zero or more PMs). It is
2432 also possible to connect ESMs together, resulting in a configuration
2433 much like the example in "High Availability in a Multiple Switch
2436 Requirements for specific modes
2437 -------------------------------
2439 The balance-rr mode requires the use of passthrough modules
2440 for devices in the bond, all connected to an common external switch.
2441 That switch must be configured for "etherchannel" or "trunking" on the
2442 appropriate ports, as is usual for balance-rr.
2444 The balance-alb and balance-tlb modes will function with
2445 either switch modules or passthrough modules (or a mix). The only
2446 specific requirement for these modes is that all network interfaces
2447 must be able to reach all destinations for traffic sent over the
2448 bonding device (i.e., the network must converge at some point outside
2451 The active-backup mode has no additional requirements.
2453 Link monitoring issues
2454 ----------------------
2456 When an Ethernet Switch Module is in place, only the ARP
2457 monitor will reliably detect link loss to an external switch. This is
2458 nothing unusual, but examination of the BladeCenter cabinet would
2459 suggest that the "external" network ports are the ethernet ports for
2460 the system, when it fact there is a switch between these "external"
2461 ports and the devices on the JS20 system itself. The MII monitor is
2462 only able to detect link failures between the ESM and the JS20 system.
2464 When a passthrough module is in place, the MII monitor does
2465 detect failures to the "external" port, which is then directly
2466 connected to the JS20 system.
2471 The Serial Over LAN (SoL) link is established over the primary
2472 ethernet (eth0) only, therefore, any loss of link to eth0 will result
2473 in losing your SoL connection. It will not fail over with other
2474 network traffic, as the SoL system is beyond the control of the
2477 It may be desirable to disable spanning tree on the switch
2478 (either the internal Ethernet Switch Module, or an external switch) to
2479 avoid fail-over delay issues when using bonding.
2482 15. Frequently Asked Questions
2483 ==============================
2487 Yes. The old 2.0.xx channel bonding patch was not SMP safe.
2488 The new driver was designed to be SMP safe from the start.
2490 2. What type of cards will work with it?
2492 Any Ethernet type cards (you can even mix cards - a Intel
2493 EtherExpress PRO/100 and a 3com 3c905b, for example). For most modes,
2494 devices need not be of the same speed.
2496 Starting with version 3.2.1, bonding also supports Infiniband
2497 slaves in active-backup mode.
2499 3. How many bonding devices can I have?
2503 4. How many slaves can a bonding device have?
2505 This is limited only by the number of network interfaces Linux
2506 supports and/or the number of network cards you can place in your
2509 5. What happens when a slave link dies?
2511 If link monitoring is enabled, then the failing device will be
2512 disabled. The active-backup mode will fail over to a backup link, and
2513 other modes will ignore the failed link. The link will continue to be
2514 monitored, and should it recover, it will rejoin the bond (in whatever
2515 manner is appropriate for the mode). See the sections on High
2516 Availability and the documentation for each mode for additional
2519 Link monitoring can be enabled via either the miimon or
2520 arp_interval parameters (described in the module parameters section,
2521 above). In general, miimon monitors the carrier state as sensed by
2522 the underlying network device, and the arp monitor (arp_interval)
2523 monitors connectivity to another host on the local network.
2525 If no link monitoring is configured, the bonding driver will
2526 be unable to detect link failures, and will assume that all links are
2527 always available. This will likely result in lost packets, and a
2528 resulting degradation of performance. The precise performance loss
2529 depends upon the bonding mode and network configuration.
2531 6. Can bonding be used for High Availability?
2533 Yes. See the section on High Availability for details.
2535 7. Which switches/systems does it work with?
2537 The full answer to this depends upon the desired mode.
2539 In the basic balance modes (balance-rr and balance-xor), it
2540 works with any system that supports etherchannel (also called
2541 trunking). Most managed switches currently available have such
2542 support, and many unmanaged switches as well.
2544 The advanced balance modes (balance-tlb and balance-alb) do
2545 not have special switch requirements, but do need device drivers that
2546 support specific features (described in the appropriate section under
2547 module parameters, above).
2549 In 802.3ad mode, it works with systems that support IEEE
2550 802.3ad Dynamic Link Aggregation. Most managed and many unmanaged
2551 switches currently available support 802.3ad.
2553 The active-backup mode should work with any Layer-II switch.
2555 8. Where does a bonding device get its MAC address from?
2557 When using slave devices that have fixed MAC addresses, or when
2558 the fail_over_mac option is enabled, the bonding device's MAC address is
2559 the MAC address of the active slave.
2561 For other configurations, if not explicitly configured (with
2562 ifconfig or ip link), the MAC address of the bonding device is taken from
2563 its first slave device. This MAC address is then passed to all following
2564 slaves and remains persistent (even if the first slave is removed) until
2565 the bonding device is brought down or reconfigured.
2567 If you wish to change the MAC address, you can set it with
2568 ifconfig or ip link:
2570 # ifconfig bond0 hw ether 00:11:22:33:44:55
2572 # ip link set bond0 address 66:77:88:99:aa:bb
2574 The MAC address can be also changed by bringing down/up the
2575 device and then changing its slaves (or their order):
2577 # ifconfig bond0 down ; modprobe -r bonding
2578 # ifconfig bond0 .... up
2579 # ifenslave bond0 eth...
2581 This method will automatically take the address from the next
2582 slave that is added.
2584 To restore your slaves' MAC addresses, you need to detach them
2585 from the bond (`ifenslave -d bond0 eth0'). The bonding driver will
2586 then restore the MAC addresses that the slaves had before they were
2589 16. Resources and Links
2590 =======================
2592 The latest version of the bonding driver can be found in the latest
2593 version of the linux kernel, found on http://kernel.org
2595 The latest version of this document can be found in the latest kernel
2596 source (named Documentation/networking/bonding.txt).
2598 Discussions regarding the usage of the bonding driver take place on the
2599 bonding-devel mailing list, hosted at sourceforge.net. If you have questions or
2600 problems, post them to the list. The list address is:
2602 bonding-devel@lists.sourceforge.net
2604 The administrative interface (to subscribe or unsubscribe) can
2607 https://lists.sourceforge.net/lists/listinfo/bonding-devel
2609 Discussions regarding the development of the bonding driver take place
2610 on the main Linux network mailing list, hosted at vger.kernel.org. The list
2613 netdev@vger.kernel.org
2615 The administrative interface (to subscribe or unsubscribe) can
2618 http://vger.kernel.org/vger-lists.html#netdev
2620 Donald Becker's Ethernet Drivers and diag programs may be found at :
2621 - http://web.archive.org/web/*/http://www.scyld.com/network/
2623 You will also find a lot of information regarding Ethernet, NWay, MII,
2624 etc. at www.scyld.com.