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 The validation of ARP requests on backup slaves is mainly
307 helping bonding to decide which slaves are more likely to
308 work in case of the active slave failure, it doesn't really
309 guarantee that the backup slave will work if it's selected
310 as the next active slave.
312 This option is useful in network configurations in which
313 multiple bonding hosts are concurrently issuing ARPs to one or
314 more targets beyond a common switch. Should the link between
315 the switch and target fail (but not the switch itself), the
316 probe traffic generated by the multiple bonding instances will
317 fool the standard ARP monitor into considering the links as
318 still up. Use of the arp_validate option can resolve this, as
319 the ARP monitor will only consider ARP requests and replies
320 associated with its own instance of bonding.
322 This option was added in bonding version 3.1.0.
326 Specifies the quantity of arp_ip_targets that must be reachable
327 in order for the ARP monitor to consider a slave as being up.
328 This option affects only active-backup mode for slaves with
329 arp_validation enabled.
335 consider the slave up only when any of the arp_ip_targets
340 consider the slave up only when all of the arp_ip_targets
345 Specifies the time, in milliseconds, to wait before disabling
346 a slave after a link failure has been detected. This option
347 is only valid for the miimon link monitor. The downdelay
348 value should be a multiple of the miimon value; if not, it
349 will be rounded down to the nearest multiple. The default
354 Specifies whether active-backup mode should set all slaves to
355 the same MAC address at enslavement (the traditional
356 behavior), or, when enabled, perform special handling of the
357 bond's MAC address in accordance with the selected policy.
363 This setting disables fail_over_mac, and causes
364 bonding to set all slaves of an active-backup bond to
365 the same MAC address at enslavement time. This is the
370 The "active" fail_over_mac policy indicates that the
371 MAC address of the bond should always be the MAC
372 address of the currently active slave. The MAC
373 address of the slaves is not changed; instead, the MAC
374 address of the bond changes during a failover.
376 This policy is useful for devices that cannot ever
377 alter their MAC address, or for devices that refuse
378 incoming broadcasts with their own source MAC (which
379 interferes with the ARP monitor).
381 The down side of this policy is that every device on
382 the network must be updated via gratuitous ARP,
383 vs. just updating a switch or set of switches (which
384 often takes place for any traffic, not just ARP
385 traffic, if the switch snoops incoming traffic to
386 update its tables) for the traditional method. If the
387 gratuitous ARP is lost, communication may be
390 When this policy is used in conjunction with the mii
391 monitor, devices which assert link up prior to being
392 able to actually transmit and receive are particularly
393 susceptible to loss of the gratuitous ARP, and an
394 appropriate updelay setting may be required.
398 The "follow" fail_over_mac policy causes the MAC
399 address of the bond to be selected normally (normally
400 the MAC address of the first slave added to the bond).
401 However, the second and subsequent slaves are not set
402 to this MAC address while they are in a backup role; a
403 slave is programmed with the bond's MAC address at
404 failover time (and the formerly active slave receives
405 the newly active slave's MAC address).
407 This policy is useful for multiport devices that
408 either become confused or incur a performance penalty
409 when multiple ports are programmed with the same MAC
413 The default policy is none, unless the first slave cannot
414 change its MAC address, in which case the active policy is
417 This option may be modified via sysfs only when no slaves are
420 This option was added in bonding version 3.2.0. The "follow"
421 policy was added in bonding version 3.3.0.
425 Option specifying the rate in which we'll ask our link partner
426 to transmit LACPDU packets in 802.3ad mode. Possible values
430 Request partner to transmit LACPDUs every 30 seconds
433 Request partner to transmit LACPDUs every 1 second
439 Specifies the number of bonding devices to create for this
440 instance of the bonding driver. E.g., if max_bonds is 3, and
441 the bonding driver is not already loaded, then bond0, bond1
442 and bond2 will be created. The default value is 1. Specifying
443 a value of 0 will load bonding, but will not create any devices.
447 Specifies the MII link monitoring frequency in milliseconds.
448 This determines how often the link state of each slave is
449 inspected for link failures. A value of zero disables MII
450 link monitoring. A value of 100 is a good starting point.
451 The use_carrier option, below, affects how the link state is
452 determined. See the High Availability section for additional
453 information. The default value is 0.
457 Specifies the minimum number of links that must be active before
458 asserting carrier. It is similar to the Cisco EtherChannel min-links
459 feature. This allows setting the minimum number of member ports that
460 must be up (link-up state) before marking the bond device as up
461 (carrier on). This is useful for situations where higher level services
462 such as clustering want to ensure a minimum number of low bandwidth
463 links are active before switchover. This option only affect 802.3ad
466 The default value is 0. This will cause carrier to be asserted (for
467 802.3ad mode) whenever there is an active aggregator, regardless of the
468 number of available links in that aggregator. Note that, because an
469 aggregator cannot be active without at least one available link,
470 setting this option to 0 or to 1 has the exact same effect.
474 Specifies one of the bonding policies. The default is
475 balance-rr (round robin). Possible values are:
479 Round-robin policy: Transmit packets in sequential
480 order from the first available slave through the
481 last. This mode provides load balancing and fault
486 Active-backup policy: Only one slave in the bond is
487 active. A different slave becomes active if, and only
488 if, the active slave fails. The bond's MAC address is
489 externally visible on only one port (network adapter)
490 to avoid confusing the switch.
492 In bonding version 2.6.2 or later, when a failover
493 occurs in active-backup mode, bonding will issue one
494 or more gratuitous ARPs on the newly active slave.
495 One gratuitous ARP is issued for the bonding master
496 interface and each VLAN interfaces configured above
497 it, provided that the interface has at least one IP
498 address configured. Gratuitous ARPs issued for VLAN
499 interfaces are tagged with the appropriate VLAN id.
501 This mode provides fault tolerance. The primary
502 option, documented below, affects the behavior of this
507 XOR policy: Transmit based on the selected transmit
508 hash policy. The default policy is a simple [(source
509 MAC address XOR'd with destination MAC address) modulo
510 slave count]. Alternate transmit policies may be
511 selected via the xmit_hash_policy option, described
514 This mode provides load balancing and fault tolerance.
518 Broadcast policy: transmits everything on all slave
519 interfaces. This mode provides fault tolerance.
523 IEEE 802.3ad Dynamic link aggregation. Creates
524 aggregation groups that share the same speed and
525 duplex settings. Utilizes all slaves in the active
526 aggregator according to the 802.3ad specification.
528 Slave selection for outgoing traffic is done according
529 to the transmit hash policy, which may be changed from
530 the default simple XOR policy via the xmit_hash_policy
531 option, documented below. Note that not all transmit
532 policies may be 802.3ad compliant, particularly in
533 regards to the packet mis-ordering requirements of
534 section 43.2.4 of the 802.3ad standard. Differing
535 peer implementations will have varying tolerances for
540 1. Ethtool support in the base drivers for retrieving
541 the speed and duplex of each slave.
543 2. A switch that supports IEEE 802.3ad Dynamic link
546 Most switches will require some type of configuration
547 to enable 802.3ad mode.
551 Adaptive transmit load balancing: channel bonding that
552 does not require any special switch support. The
553 outgoing traffic is distributed according to the
554 current load (computed relative to the speed) on each
555 slave. Incoming traffic is received by the current
556 slave. If the receiving slave fails, another slave
557 takes over the MAC address of the failed receiving
562 Ethtool support in the base drivers for retrieving the
567 Adaptive load balancing: includes balance-tlb plus
568 receive load balancing (rlb) for IPV4 traffic, and
569 does not require any special switch support. The
570 receive load balancing is achieved by ARP negotiation.
571 The bonding driver intercepts the ARP Replies sent by
572 the local system on their way out and overwrites the
573 source hardware address with the unique hardware
574 address of one of the slaves in the bond such that
575 different peers use different hardware addresses for
578 Receive traffic from connections created by the server
579 is also balanced. When the local system sends an ARP
580 Request the bonding driver copies and saves the peer's
581 IP information from the ARP packet. When the ARP
582 Reply arrives from the peer, its hardware address is
583 retrieved and the bonding driver initiates an ARP
584 reply to this peer assigning it to one of the slaves
585 in the bond. A problematic outcome of using ARP
586 negotiation for balancing is that each time that an
587 ARP request is broadcast it uses the hardware address
588 of the bond. Hence, peers learn the hardware address
589 of the bond and the balancing of receive traffic
590 collapses to the current slave. This is handled by
591 sending updates (ARP Replies) to all the peers with
592 their individually assigned hardware address such that
593 the traffic is redistributed. Receive traffic is also
594 redistributed when a new slave is added to the bond
595 and when an inactive slave is re-activated. The
596 receive load is distributed sequentially (round robin)
597 among the group of highest speed slaves in the bond.
599 When a link is reconnected or a new slave joins the
600 bond the receive traffic is redistributed among all
601 active slaves in the bond by initiating ARP Replies
602 with the selected MAC address to each of the
603 clients. The updelay parameter (detailed below) must
604 be set to a value equal or greater than the switch's
605 forwarding delay so that the ARP Replies sent to the
606 peers will not be blocked by the switch.
610 1. Ethtool support in the base drivers for retrieving
611 the speed of each slave.
613 2. Base driver support for setting the hardware
614 address of a device while it is open. This is
615 required so that there will always be one slave in the
616 team using the bond hardware address (the
617 curr_active_slave) while having a unique hardware
618 address for each slave in the bond. If the
619 curr_active_slave fails its hardware address is
620 swapped with the new curr_active_slave that was
626 Specify the number of peer notifications (gratuitous ARPs and
627 unsolicited IPv6 Neighbor Advertisements) to be issued after a
628 failover event. As soon as the link is up on the new slave
629 (possibly immediately) a peer notification is sent on the
630 bonding device and each VLAN sub-device. This is repeated at
631 each link monitor interval (arp_interval or miimon, whichever
632 is active) if the number is greater than 1.
634 The valid range is 0 - 255; the default value is 1. These options
635 affect only the active-backup mode. These options were added for
636 bonding versions 3.3.0 and 3.4.0 respectively.
638 From Linux 3.0 and bonding version 3.7.1, these notifications
639 are generated by the ipv4 and ipv6 code and the numbers of
640 repetitions cannot be set independently.
644 A string (eth0, eth2, etc) specifying which slave is the
645 primary device. The specified device will always be the
646 active slave while it is available. Only when the primary is
647 off-line will alternate devices be used. This is useful when
648 one slave is preferred over another, e.g., when one slave has
649 higher throughput than another.
651 The primary option is only valid for active-backup mode.
655 Specifies the reselection policy for the primary slave. This
656 affects how the primary slave is chosen to become the active slave
657 when failure of the active slave or recovery of the primary slave
658 occurs. This option is designed to prevent flip-flopping between
659 the primary slave and other slaves. Possible values are:
661 always or 0 (default)
663 The primary slave becomes the active slave whenever it
668 The primary slave becomes the active slave when it comes
669 back up, if the speed and duplex of the primary slave is
670 better than the speed and duplex of the current active
675 The primary slave becomes the active slave only if the
676 current active slave fails and the primary slave is up.
678 The primary_reselect setting is ignored in two cases:
680 If no slaves are active, the first slave to recover is
681 made the active slave.
683 When initially enslaved, the primary slave is always made
686 Changing the primary_reselect policy via sysfs will cause an
687 immediate selection of the best active slave according to the new
688 policy. This may or may not result in a change of the active
689 slave, depending upon the circumstances.
691 This option was added for bonding version 3.6.0.
695 Specifies the time, in milliseconds, to wait before enabling a
696 slave after a link recovery has been detected. This option is
697 only valid for the miimon link monitor. The updelay value
698 should be a multiple of the miimon value; if not, it will be
699 rounded down to the nearest multiple. The default value is 0.
703 Specifies whether or not miimon should use MII or ETHTOOL
704 ioctls vs. netif_carrier_ok() to determine the link
705 status. The MII or ETHTOOL ioctls are less efficient and
706 utilize a deprecated calling sequence within the kernel. The
707 netif_carrier_ok() relies on the device driver to maintain its
708 state with netif_carrier_on/off; at this writing, most, but
709 not all, device drivers support this facility.
711 If bonding insists that the link is up when it should not be,
712 it may be that your network device driver does not support
713 netif_carrier_on/off. The default state for netif_carrier is
714 "carrier on," so if a driver does not support netif_carrier,
715 it will appear as if the link is always up. In this case,
716 setting use_carrier to 0 will cause bonding to revert to the
717 MII / ETHTOOL ioctl method to determine the link state.
719 A value of 1 enables the use of netif_carrier_ok(), a value of
720 0 will use the deprecated MII / ETHTOOL ioctls. The default
725 Selects the transmit hash policy to use for slave selection in
726 balance-xor and 802.3ad modes. Possible values are:
730 Uses XOR of hardware MAC addresses to generate the
733 (source MAC XOR destination MAC) modulo slave count
735 This algorithm will place all traffic to a particular
736 network peer on the same slave.
738 This algorithm is 802.3ad compliant.
742 This policy uses a combination of layer2 and layer3
743 protocol information to generate the hash.
745 Uses XOR of hardware MAC addresses and IP addresses to
746 generate the hash. The formula is
748 hash = source MAC XOR destination MAC
749 hash = hash XOR source IP XOR destination IP
750 hash = hash XOR (hash RSHIFT 16)
751 hash = hash XOR (hash RSHIFT 8)
752 And then hash is reduced modulo slave count.
754 If the protocol is IPv6 then the source and destination
755 addresses are first hashed using ipv6_addr_hash.
757 This algorithm will place all traffic to a particular
758 network peer on the same slave. For non-IP traffic,
759 the formula is the same as for the layer2 transmit
762 This policy is intended to provide a more balanced
763 distribution of traffic than layer2 alone, especially
764 in environments where a layer3 gateway device is
765 required to reach most destinations.
767 This algorithm is 802.3ad compliant.
771 This policy uses upper layer protocol information,
772 when available, to generate the hash. This allows for
773 traffic to a particular network peer to span multiple
774 slaves, although a single connection will not span
777 The formula for unfragmented TCP and UDP packets is
779 hash = source port, destination port (as in the header)
780 hash = hash XOR source IP XOR destination IP
781 hash = hash XOR (hash RSHIFT 16)
782 hash = hash XOR (hash RSHIFT 8)
783 And then hash is reduced modulo slave count.
785 If the protocol is IPv6 then the source and destination
786 addresses are first hashed using ipv6_addr_hash.
788 For fragmented TCP or UDP packets and all other IPv4 and
789 IPv6 protocol traffic, the source and destination port
790 information is omitted. For non-IP traffic, the
791 formula is the same as for the layer2 transmit hash
794 This algorithm is not fully 802.3ad compliant. A
795 single TCP or UDP conversation containing both
796 fragmented and unfragmented packets will see packets
797 striped across two interfaces. This may result in out
798 of order delivery. Most traffic types will not meet
799 this criteria, as TCP rarely fragments traffic, and
800 most UDP traffic is not involved in extended
801 conversations. Other implementations of 802.3ad may
802 or may not tolerate this noncompliance.
806 This policy uses the same formula as layer2+3 but it
807 relies on skb_flow_dissect to obtain the header fields
808 which might result in the use of inner headers if an
809 encapsulation protocol is used. For example this will
810 improve the performance for tunnel users because the
811 packets will be distributed according to the encapsulated
816 This policy uses the same formula as layer3+4 but it
817 relies on skb_flow_dissect to obtain the header fields
818 which might result in the use of inner headers if an
819 encapsulation protocol is used. For example this will
820 improve the performance for tunnel users because the
821 packets will be distributed according to the encapsulated
824 The default value is layer2. This option was added in bonding
825 version 2.6.3. In earlier versions of bonding, this parameter
826 does not exist, and the layer2 policy is the only policy. The
827 layer2+3 value was added for bonding version 3.2.2.
831 Specifies the number of IGMP membership reports to be issued after
832 a failover event. One membership report is issued immediately after
833 the failover, subsequent packets are sent in each 200ms interval.
835 The valid range is 0 - 255; the default value is 1. A value of 0
836 prevents the IGMP membership report from being issued in response
837 to the failover event.
839 This option is useful for bonding modes balance-rr (0), active-backup
840 (1), balance-tlb (5) and balance-alb (6), in which a failover can
841 switch the IGMP traffic from one slave to another. Therefore a fresh
842 IGMP report must be issued to cause the switch to forward the incoming
843 IGMP traffic over the newly selected slave.
845 This option was added for bonding version 3.7.0.
847 3. Configuring Bonding Devices
848 ==============================
850 You can configure bonding using either your distro's network
851 initialization scripts, or manually using either iproute2 or the
852 sysfs interface. Distros generally use one of three packages for the
853 network initialization scripts: initscripts, sysconfig or interfaces.
854 Recent versions of these packages have support for bonding, while older
857 We will first describe the options for configuring bonding for
858 distros using versions of initscripts, sysconfig and interfaces with full
859 or partial support for bonding, then provide information on enabling
860 bonding without support from the network initialization scripts (i.e.,
861 older versions of initscripts or sysconfig).
863 If you're unsure whether your distro uses sysconfig,
864 initscripts or interfaces, or don't know if it's new enough, have no fear.
865 Determining this is fairly straightforward.
867 First, look for a file called interfaces in /etc/network directory.
868 If this file is present in your system, then your system use interfaces. See
869 Configuration with Interfaces Support.
871 Else, issue the command:
875 It will respond with a line of text starting with either
876 "initscripts" or "sysconfig," followed by some numbers. This is the
877 package that provides your network initialization scripts.
879 Next, to determine if your installation supports bonding,
882 $ grep ifenslave /sbin/ifup
884 If this returns any matches, then your initscripts or
885 sysconfig has support for bonding.
887 3.1 Configuration with Sysconfig Support
888 ----------------------------------------
890 This section applies to distros using a version of sysconfig
891 with bonding support, for example, SuSE Linux Enterprise Server 9.
893 SuSE SLES 9's networking configuration system does support
894 bonding, however, at this writing, the YaST system configuration
895 front end does not provide any means to work with bonding devices.
896 Bonding devices can be managed by hand, however, as follows.
898 First, if they have not already been configured, configure the
899 slave devices. On SLES 9, this is most easily done by running the
900 yast2 sysconfig configuration utility. The goal is for to create an
901 ifcfg-id file for each slave device. The simplest way to accomplish
902 this is to configure the devices for DHCP (this is only to get the
903 file ifcfg-id file created; see below for some issues with DHCP). The
904 name of the configuration file for each device will be of the form:
906 ifcfg-id-xx:xx:xx:xx:xx:xx
908 Where the "xx" portion will be replaced with the digits from
909 the device's permanent MAC address.
911 Once the set of ifcfg-id-xx:xx:xx:xx:xx:xx files has been
912 created, it is necessary to edit the configuration files for the slave
913 devices (the MAC addresses correspond to those of the slave devices).
914 Before editing, the file will contain multiple lines, and will look
920 UNIQUE='XNzu.WeZGOGF+4wE'
921 _nm_name='bus-pci-0001:61:01.0'
923 Change the BOOTPROTO and STARTMODE lines to the following:
928 Do not alter the UNIQUE or _nm_name lines. Remove any other
929 lines (USERCTL, etc).
931 Once the ifcfg-id-xx:xx:xx:xx:xx:xx files have been modified,
932 it's time to create the configuration file for the bonding device
933 itself. This file is named ifcfg-bondX, where X is the number of the
934 bonding device to create, starting at 0. The first such file is
935 ifcfg-bond0, the second is ifcfg-bond1, and so on. The sysconfig
936 network configuration system will correctly start multiple instances
939 The contents of the ifcfg-bondX file is as follows:
942 BROADCAST="10.0.2.255"
944 NETMASK="255.255.0.0"
949 BONDING_MODULE_OPTS="mode=active-backup miimon=100"
950 BONDING_SLAVE0="eth0"
951 BONDING_SLAVE1="bus-pci-0000:06:08.1"
953 Replace the sample BROADCAST, IPADDR, NETMASK and NETWORK
954 values with the appropriate values for your network.
956 The STARTMODE specifies when the device is brought online.
957 The possible values are:
959 onboot: The device is started at boot time. If you're not
960 sure, this is probably what you want.
962 manual: The device is started only when ifup is called
963 manually. Bonding devices may be configured this
964 way if you do not wish them to start automatically
965 at boot for some reason.
967 hotplug: The device is started by a hotplug event. This is not
968 a valid choice for a bonding device.
970 off or ignore: The device configuration is ignored.
972 The line BONDING_MASTER='yes' indicates that the device is a
973 bonding master device. The only useful value is "yes."
975 The contents of BONDING_MODULE_OPTS are supplied to the
976 instance of the bonding module for this device. Specify the options
977 for the bonding mode, link monitoring, and so on here. Do not include
978 the max_bonds bonding parameter; this will confuse the configuration
979 system if you have multiple bonding devices.
981 Finally, supply one BONDING_SLAVEn="slave device" for each
982 slave. where "n" is an increasing value, one for each slave. The
983 "slave device" is either an interface name, e.g., "eth0", or a device
984 specifier for the network device. The interface name is easier to
985 find, but the ethN names are subject to change at boot time if, e.g.,
986 a device early in the sequence has failed. The device specifiers
987 (bus-pci-0000:06:08.1 in the example above) specify the physical
988 network device, and will not change unless the device's bus location
989 changes (for example, it is moved from one PCI slot to another). The
990 example above uses one of each type for demonstration purposes; most
991 configurations will choose one or the other for all slave devices.
993 When all configuration files have been modified or created,
994 networking must be restarted for the configuration changes to take
995 effect. This can be accomplished via the following:
997 # /etc/init.d/network restart
999 Note that the network control script (/sbin/ifdown) will
1000 remove the bonding module as part of the network shutdown processing,
1001 so it is not necessary to remove the module by hand if, e.g., the
1002 module parameters have changed.
1004 Also, at this writing, YaST/YaST2 will not manage bonding
1005 devices (they do not show bonding interfaces on its list of network
1006 devices). It is necessary to edit the configuration file by hand to
1007 change the bonding configuration.
1009 Additional general options and details of the ifcfg file
1010 format can be found in an example ifcfg template file:
1012 /etc/sysconfig/network/ifcfg.template
1014 Note that the template does not document the various BONDING_
1015 settings described above, but does describe many of the other options.
1017 3.1.1 Using DHCP with Sysconfig
1018 -------------------------------
1020 Under sysconfig, configuring a device with BOOTPROTO='dhcp'
1021 will cause it to query DHCP for its IP address information. At this
1022 writing, this does not function for bonding devices; the scripts
1023 attempt to obtain the device address from DHCP prior to adding any of
1024 the slave devices. Without active slaves, the DHCP requests are not
1025 sent to the network.
1027 3.1.2 Configuring Multiple Bonds with Sysconfig
1028 -----------------------------------------------
1030 The sysconfig network initialization system is capable of
1031 handling multiple bonding devices. All that is necessary is for each
1032 bonding instance to have an appropriately configured ifcfg-bondX file
1033 (as described above). Do not specify the "max_bonds" parameter to any
1034 instance of bonding, as this will confuse sysconfig. If you require
1035 multiple bonding devices with identical parameters, create multiple
1038 Because the sysconfig scripts supply the bonding module
1039 options in the ifcfg-bondX file, it is not necessary to add them to
1040 the system /etc/modules.d/*.conf configuration files.
1042 3.2 Configuration with Initscripts Support
1043 ------------------------------------------
1045 This section applies to distros using a recent version of
1046 initscripts with bonding support, for example, Red Hat Enterprise Linux
1047 version 3 or later, Fedora, etc. On these systems, the network
1048 initialization scripts have knowledge of bonding, and can be configured to
1049 control bonding devices. Note that older versions of the initscripts
1050 package have lower levels of support for bonding; this will be noted where
1053 These distros will not automatically load the network adapter
1054 driver unless the ethX device is configured with an IP address.
1055 Because of this constraint, users must manually configure a
1056 network-script file for all physical adapters that will be members of
1057 a bondX link. Network script files are located in the directory:
1059 /etc/sysconfig/network-scripts
1061 The file name must be prefixed with "ifcfg-eth" and suffixed
1062 with the adapter's physical adapter number. For example, the script
1063 for eth0 would be named /etc/sysconfig/network-scripts/ifcfg-eth0.
1064 Place the following text in the file:
1073 The DEVICE= line will be different for every ethX device and
1074 must correspond with the name of the file, i.e., ifcfg-eth1 must have
1075 a device line of DEVICE=eth1. The setting of the MASTER= line will
1076 also depend on the final bonding interface name chosen for your bond.
1077 As with other network devices, these typically start at 0, and go up
1078 one for each device, i.e., the first bonding instance is bond0, the
1079 second is bond1, and so on.
1081 Next, create a bond network script. The file name for this
1082 script will be /etc/sysconfig/network-scripts/ifcfg-bondX where X is
1083 the number of the bond. For bond0 the file is named "ifcfg-bond0",
1084 for bond1 it is named "ifcfg-bond1", and so on. Within that file,
1085 place the following text:
1089 NETMASK=255.255.255.0
1091 BROADCAST=192.168.1.255
1096 Be sure to change the networking specific lines (IPADDR,
1097 NETMASK, NETWORK and BROADCAST) to match your network configuration.
1099 For later versions of initscripts, such as that found with Fedora
1100 7 (or later) and Red Hat Enterprise Linux version 5 (or later), it is possible,
1101 and, indeed, preferable, to specify the bonding options in the ifcfg-bond0
1102 file, e.g. a line of the format:
1104 BONDING_OPTS="mode=active-backup arp_interval=60 arp_ip_target=192.168.1.254"
1106 will configure the bond with the specified options. The options
1107 specified in BONDING_OPTS are identical to the bonding module parameters
1108 except for the arp_ip_target field when using versions of initscripts older
1109 than and 8.57 (Fedora 8) and 8.45.19 (Red Hat Enterprise Linux 5.2). When
1110 using older versions each target should be included as a separate option and
1111 should be preceded by a '+' to indicate it should be added to the list of
1112 queried targets, e.g.,
1114 arp_ip_target=+192.168.1.1 arp_ip_target=+192.168.1.2
1116 is the proper syntax to specify multiple targets. When specifying
1117 options via BONDING_OPTS, it is not necessary to edit /etc/modprobe.d/*.conf.
1119 For even older versions of initscripts that do not support
1120 BONDING_OPTS, it is necessary to edit /etc/modprobe.d/*.conf, depending upon
1121 your distro) to load the bonding module with your desired options when the
1122 bond0 interface is brought up. The following lines in /etc/modprobe.d/*.conf
1123 will load the bonding module, and select its options:
1126 options bond0 mode=balance-alb miimon=100
1128 Replace the sample parameters with the appropriate set of
1129 options for your configuration.
1131 Finally run "/etc/rc.d/init.d/network restart" as root. This
1132 will restart the networking subsystem and your bond link should be now
1135 3.2.1 Using DHCP with Initscripts
1136 ---------------------------------
1138 Recent versions of initscripts (the versions supplied with Fedora
1139 Core 3 and Red Hat Enterprise Linux 4, or later versions, are reported to
1140 work) have support for assigning IP information to bonding devices via
1143 To configure bonding for DHCP, configure it as described
1144 above, except replace the line "BOOTPROTO=none" with "BOOTPROTO=dhcp"
1145 and add a line consisting of "TYPE=Bonding". Note that the TYPE value
1148 3.2.2 Configuring Multiple Bonds with Initscripts
1149 -------------------------------------------------
1151 Initscripts packages that are included with Fedora 7 and Red Hat
1152 Enterprise Linux 5 support multiple bonding interfaces by simply
1153 specifying the appropriate BONDING_OPTS= in ifcfg-bondX where X is the
1154 number of the bond. This support requires sysfs support in the kernel,
1155 and a bonding driver of version 3.0.0 or later. Other configurations may
1156 not support this method for specifying multiple bonding interfaces; for
1157 those instances, see the "Configuring Multiple Bonds Manually" section,
1160 3.3 Configuring Bonding Manually with iproute2
1161 -----------------------------------------------
1163 This section applies to distros whose network initialization
1164 scripts (the sysconfig or initscripts package) do not have specific
1165 knowledge of bonding. One such distro is SuSE Linux Enterprise Server
1168 The general method for these systems is to place the bonding
1169 module parameters into a config file in /etc/modprobe.d/ (as
1170 appropriate for the installed distro), then add modprobe and/or
1171 `ip link` commands to the system's global init script. The name of
1172 the global init script differs; for sysconfig, it is
1173 /etc/init.d/boot.local and for initscripts it is /etc/rc.d/rc.local.
1175 For example, if you wanted to make a simple bond of two e100
1176 devices (presumed to be eth0 and eth1), and have it persist across
1177 reboots, edit the appropriate file (/etc/init.d/boot.local or
1178 /etc/rc.d/rc.local), and add the following:
1180 modprobe bonding mode=balance-alb miimon=100
1182 ifconfig bond0 192.168.1.1 netmask 255.255.255.0 up
1183 ip link set eth0 master bond0
1184 ip link set eth1 master bond0
1186 Replace the example bonding module parameters and bond0
1187 network configuration (IP address, netmask, etc) with the appropriate
1188 values for your configuration.
1190 Unfortunately, this method will not provide support for the
1191 ifup and ifdown scripts on the bond devices. To reload the bonding
1192 configuration, it is necessary to run the initialization script, e.g.,
1194 # /etc/init.d/boot.local
1198 # /etc/rc.d/rc.local
1200 It may be desirable in such a case to create a separate script
1201 which only initializes the bonding configuration, then call that
1202 separate script from within boot.local. This allows for bonding to be
1203 enabled without re-running the entire global init script.
1205 To shut down the bonding devices, it is necessary to first
1206 mark the bonding device itself as being down, then remove the
1207 appropriate device driver modules. For our example above, you can do
1210 # ifconfig bond0 down
1214 Again, for convenience, it may be desirable to create a script
1215 with these commands.
1218 3.3.1 Configuring Multiple Bonds Manually
1219 -----------------------------------------
1221 This section contains information on configuring multiple
1222 bonding devices with differing options for those systems whose network
1223 initialization scripts lack support for configuring multiple bonds.
1225 If you require multiple bonding devices, but all with the same
1226 options, you may wish to use the "max_bonds" module parameter,
1229 To create multiple bonding devices with differing options, it is
1230 preferable to use bonding parameters exported by sysfs, documented in the
1233 For versions of bonding without sysfs support, the only means to
1234 provide multiple instances of bonding with differing options is to load
1235 the bonding driver multiple times. Note that current versions of the
1236 sysconfig network initialization scripts handle this automatically; if
1237 your distro uses these scripts, no special action is needed. See the
1238 section Configuring Bonding Devices, above, if you're not sure about your
1239 network initialization scripts.
1241 To load multiple instances of the module, it is necessary to
1242 specify a different name for each instance (the module loading system
1243 requires that every loaded module, even multiple instances of the same
1244 module, have a unique name). This is accomplished by supplying multiple
1245 sets of bonding options in /etc/modprobe.d/*.conf, for example:
1248 options bond0 -o bond0 mode=balance-rr miimon=100
1251 options bond1 -o bond1 mode=balance-alb miimon=50
1253 will load the bonding module two times. The first instance is
1254 named "bond0" and creates the bond0 device in balance-rr mode with an
1255 miimon of 100. The second instance is named "bond1" and creates the
1256 bond1 device in balance-alb mode with an miimon of 50.
1258 In some circumstances (typically with older distributions),
1259 the above does not work, and the second bonding instance never sees
1260 its options. In that case, the second options line can be substituted
1263 install bond1 /sbin/modprobe --ignore-install bonding -o bond1 \
1264 mode=balance-alb miimon=50
1266 This may be repeated any number of times, specifying a new and
1267 unique name in place of bond1 for each subsequent instance.
1269 It has been observed that some Red Hat supplied kernels are unable
1270 to rename modules at load time (the "-o bond1" part). Attempts to pass
1271 that option to modprobe will produce an "Operation not permitted" error.
1272 This has been reported on some Fedora Core kernels, and has been seen on
1273 RHEL 4 as well. On kernels exhibiting this problem, it will be impossible
1274 to configure multiple bonds with differing parameters (as they are older
1275 kernels, and also lack sysfs support).
1277 3.4 Configuring Bonding Manually via Sysfs
1278 ------------------------------------------
1280 Starting with version 3.0.0, Channel Bonding may be configured
1281 via the sysfs interface. This interface allows dynamic configuration
1282 of all bonds in the system without unloading the module. It also
1283 allows for adding and removing bonds at runtime. Ifenslave is no
1284 longer required, though it is still supported.
1286 Use of the sysfs interface allows you to use multiple bonds
1287 with different configurations without having to reload the module.
1288 It also allows you to use multiple, differently configured bonds when
1289 bonding is compiled into the kernel.
1291 You must have the sysfs filesystem mounted to configure
1292 bonding this way. The examples in this document assume that you
1293 are using the standard mount point for sysfs, e.g. /sys. If your
1294 sysfs filesystem is mounted elsewhere, you will need to adjust the
1295 example paths accordingly.
1297 Creating and Destroying Bonds
1298 -----------------------------
1299 To add a new bond foo:
1300 # echo +foo > /sys/class/net/bonding_masters
1302 To remove an existing bond bar:
1303 # echo -bar > /sys/class/net/bonding_masters
1305 To show all existing bonds:
1306 # cat /sys/class/net/bonding_masters
1308 NOTE: due to 4K size limitation of sysfs files, this list may be
1309 truncated if you have more than a few hundred bonds. This is unlikely
1310 to occur under normal operating conditions.
1312 Adding and Removing Slaves
1313 --------------------------
1314 Interfaces may be enslaved to a bond using the file
1315 /sys/class/net/<bond>/bonding/slaves. The semantics for this file
1316 are the same as for the bonding_masters file.
1318 To enslave interface eth0 to bond bond0:
1320 # echo +eth0 > /sys/class/net/bond0/bonding/slaves
1322 To free slave eth0 from bond bond0:
1323 # echo -eth0 > /sys/class/net/bond0/bonding/slaves
1325 When an interface is enslaved to a bond, symlinks between the
1326 two are created in the sysfs filesystem. In this case, you would get
1327 /sys/class/net/bond0/slave_eth0 pointing to /sys/class/net/eth0, and
1328 /sys/class/net/eth0/master pointing to /sys/class/net/bond0.
1330 This means that you can tell quickly whether or not an
1331 interface is enslaved by looking for the master symlink. Thus:
1332 # echo -eth0 > /sys/class/net/eth0/master/bonding/slaves
1333 will free eth0 from whatever bond it is enslaved to, regardless of
1334 the name of the bond interface.
1336 Changing a Bond's Configuration
1337 -------------------------------
1338 Each bond may be configured individually by manipulating the
1339 files located in /sys/class/net/<bond name>/bonding
1341 The names of these files correspond directly with the command-
1342 line parameters described elsewhere in this file, and, with the
1343 exception of arp_ip_target, they accept the same values. To see the
1344 current setting, simply cat the appropriate file.
1346 A few examples will be given here; for specific usage
1347 guidelines for each parameter, see the appropriate section in this
1350 To configure bond0 for balance-alb mode:
1351 # ifconfig bond0 down
1352 # echo 6 > /sys/class/net/bond0/bonding/mode
1354 # echo balance-alb > /sys/class/net/bond0/bonding/mode
1355 NOTE: The bond interface must be down before the mode can be
1358 To enable MII monitoring on bond0 with a 1 second interval:
1359 # echo 1000 > /sys/class/net/bond0/bonding/miimon
1360 NOTE: If ARP monitoring is enabled, it will disabled when MII
1361 monitoring is enabled, and vice-versa.
1364 # echo +192.168.0.100 > /sys/class/net/bond0/bonding/arp_ip_target
1365 # echo +192.168.0.101 > /sys/class/net/bond0/bonding/arp_ip_target
1366 NOTE: up to 16 target addresses may be specified.
1368 To remove an ARP target:
1369 # echo -192.168.0.100 > /sys/class/net/bond0/bonding/arp_ip_target
1371 To configure the interval between learning packet transmits:
1372 # echo 12 > /sys/class/net/bond0/bonding/lp_interval
1373 NOTE: the lp_inteval is the number of seconds between instances where
1374 the bonding driver sends learning packets to each slaves peer switch. The
1375 default interval is 1 second.
1377 Example Configuration
1378 ---------------------
1379 We begin with the same example that is shown in section 3.3,
1380 executed with sysfs, and without using ifenslave.
1382 To make a simple bond of two e100 devices (presumed to be eth0
1383 and eth1), and have it persist across reboots, edit the appropriate
1384 file (/etc/init.d/boot.local or /etc/rc.d/rc.local), and add the
1389 echo balance-alb > /sys/class/net/bond0/bonding/mode
1390 ifconfig bond0 192.168.1.1 netmask 255.255.255.0 up
1391 echo 100 > /sys/class/net/bond0/bonding/miimon
1392 echo +eth0 > /sys/class/net/bond0/bonding/slaves
1393 echo +eth1 > /sys/class/net/bond0/bonding/slaves
1395 To add a second bond, with two e1000 interfaces in
1396 active-backup mode, using ARP monitoring, add the following lines to
1400 echo +bond1 > /sys/class/net/bonding_masters
1401 echo active-backup > /sys/class/net/bond1/bonding/mode
1402 ifconfig bond1 192.168.2.1 netmask 255.255.255.0 up
1403 echo +192.168.2.100 /sys/class/net/bond1/bonding/arp_ip_target
1404 echo 2000 > /sys/class/net/bond1/bonding/arp_interval
1405 echo +eth2 > /sys/class/net/bond1/bonding/slaves
1406 echo +eth3 > /sys/class/net/bond1/bonding/slaves
1408 3.5 Configuration with Interfaces Support
1409 -----------------------------------------
1411 This section applies to distros which use /etc/network/interfaces file
1412 to describe network interface configuration, most notably Debian and it's
1415 The ifup and ifdown commands on Debian don't support bonding out of
1416 the box. The ifenslave-2.6 package should be installed to provide bonding
1417 support. Once installed, this package will provide bond-* options to be used
1418 into /etc/network/interfaces.
1420 Note that ifenslave-2.6 package will load the bonding module and use
1421 the ifenslave command when appropriate.
1423 Example Configurations
1424 ----------------------
1426 In /etc/network/interfaces, the following stanza will configure bond0, in
1427 active-backup mode, with eth0 and eth1 as slaves.
1430 iface bond0 inet dhcp
1431 bond-slaves eth0 eth1
1432 bond-mode active-backup
1434 bond-primary eth0 eth1
1436 If the above configuration doesn't work, you might have a system using
1437 upstart for system startup. This is most notably true for recent
1438 Ubuntu versions. The following stanza in /etc/network/interfaces will
1439 produce the same result on those systems.
1442 iface bond0 inet dhcp
1444 bond-mode active-backup
1448 iface eth0 inet manual
1450 bond-primary eth0 eth1
1453 iface eth1 inet manual
1455 bond-primary eth0 eth1
1457 For a full list of bond-* supported options in /etc/network/interfaces and some
1458 more advanced examples tailored to you particular distros, see the files in
1459 /usr/share/doc/ifenslave-2.6.
1461 3.6 Overriding Configuration for Special Cases
1462 ----------------------------------------------
1464 When using the bonding driver, the physical port which transmits a frame is
1465 typically selected by the bonding driver, and is not relevant to the user or
1466 system administrator. The output port is simply selected using the policies of
1467 the selected bonding mode. On occasion however, it is helpful to direct certain
1468 classes of traffic to certain physical interfaces on output to implement
1469 slightly more complex policies. For example, to reach a web server over a
1470 bonded interface in which eth0 connects to a private network, while eth1
1471 connects via a public network, it may be desirous to bias the bond to send said
1472 traffic over eth0 first, using eth1 only as a fall back, while all other traffic
1473 can safely be sent over either interface. Such configurations may be achieved
1474 using the traffic control utilities inherent in linux.
1476 By default the bonding driver is multiqueue aware and 16 queues are created
1477 when the driver initializes (see Documentation/networking/multiqueue.txt
1478 for details). If more or less queues are desired the module parameter
1479 tx_queues can be used to change this value. There is no sysfs parameter
1480 available as the allocation is done at module init time.
1482 The output of the file /proc/net/bonding/bondX has changed so the output Queue
1483 ID is now printed for each slave:
1485 Bonding Mode: fault-tolerance (active-backup)
1487 Currently Active Slave: eth0
1489 MII Polling Interval (ms): 0
1493 Slave Interface: eth0
1495 Link Failure Count: 0
1496 Permanent HW addr: 00:1a:a0:12:8f:cb
1499 Slave Interface: eth1
1501 Link Failure Count: 0
1502 Permanent HW addr: 00:1a:a0:12:8f:cc
1505 The queue_id for a slave can be set using the command:
1507 # echo "eth1:2" > /sys/class/net/bond0/bonding/queue_id
1509 Any interface that needs a queue_id set should set it with multiple calls
1510 like the one above until proper priorities are set for all interfaces. On
1511 distributions that allow configuration via initscripts, multiple 'queue_id'
1512 arguments can be added to BONDING_OPTS to set all needed slave queues.
1514 These queue id's can be used in conjunction with the tc utility to configure
1515 a multiqueue qdisc and filters to bias certain traffic to transmit on certain
1516 slave devices. For instance, say we wanted, in the above configuration to
1517 force all traffic bound to 192.168.1.100 to use eth1 in the bond as its output
1518 device. The following commands would accomplish this:
1520 # tc qdisc add dev bond0 handle 1 root multiq
1522 # tc filter add dev bond0 protocol ip parent 1: prio 1 u32 match ip dst \
1523 192.168.1.100 action skbedit queue_mapping 2
1525 These commands tell the kernel to attach a multiqueue queue discipline to the
1526 bond0 interface and filter traffic enqueued to it, such that packets with a dst
1527 ip of 192.168.1.100 have their output queue mapping value overwritten to 2.
1528 This value is then passed into the driver, causing the normal output path
1529 selection policy to be overridden, selecting instead qid 2, which maps to eth1.
1531 Note that qid values begin at 1. Qid 0 is reserved to initiate to the driver
1532 that normal output policy selection should take place. One benefit to simply
1533 leaving the qid for a slave to 0 is the multiqueue awareness in the bonding
1534 driver that is now present. This awareness allows tc filters to be placed on
1535 slave devices as well as bond devices and the bonding driver will simply act as
1536 a pass-through for selecting output queues on the slave device rather than
1537 output port selection.
1539 This feature first appeared in bonding driver version 3.7.0 and support for
1540 output slave selection was limited to round-robin and active-backup modes.
1542 4 Querying Bonding Configuration
1543 =================================
1545 4.1 Bonding Configuration
1546 -------------------------
1548 Each bonding device has a read-only file residing in the
1549 /proc/net/bonding directory. The file contents include information
1550 about the bonding configuration, options and state of each slave.
1552 For example, the contents of /proc/net/bonding/bond0 after the
1553 driver is loaded with parameters of mode=0 and miimon=1000 is
1554 generally as follows:
1556 Ethernet Channel Bonding Driver: 2.6.1 (October 29, 2004)
1557 Bonding Mode: load balancing (round-robin)
1558 Currently Active Slave: eth0
1560 MII Polling Interval (ms): 1000
1564 Slave Interface: eth1
1566 Link Failure Count: 1
1568 Slave Interface: eth0
1570 Link Failure Count: 1
1572 The precise format and contents will change depending upon the
1573 bonding configuration, state, and version of the bonding driver.
1575 4.2 Network configuration
1576 -------------------------
1578 The network configuration can be inspected using the ifconfig
1579 command. Bonding devices will have the MASTER flag set; Bonding slave
1580 devices will have the SLAVE flag set. The ifconfig output does not
1581 contain information on which slaves are associated with which masters.
1583 In the example below, the bond0 interface is the master
1584 (MASTER) while eth0 and eth1 are slaves (SLAVE). Notice all slaves of
1585 bond0 have the same MAC address (HWaddr) as bond0 for all modes except
1586 TLB and ALB that require a unique MAC address for each slave.
1589 bond0 Link encap:Ethernet HWaddr 00:C0:F0:1F:37:B4
1590 inet addr:XXX.XXX.XXX.YYY Bcast:XXX.XXX.XXX.255 Mask:255.255.252.0
1591 UP BROADCAST RUNNING MASTER MULTICAST MTU:1500 Metric:1
1592 RX packets:7224794 errors:0 dropped:0 overruns:0 frame:0
1593 TX packets:3286647 errors:1 dropped:0 overruns:1 carrier:0
1594 collisions:0 txqueuelen:0
1596 eth0 Link encap:Ethernet HWaddr 00:C0:F0:1F:37:B4
1597 UP BROADCAST RUNNING SLAVE MULTICAST MTU:1500 Metric:1
1598 RX packets:3573025 errors:0 dropped:0 overruns:0 frame:0
1599 TX packets:1643167 errors:1 dropped:0 overruns:1 carrier:0
1600 collisions:0 txqueuelen:100
1601 Interrupt:10 Base address:0x1080
1603 eth1 Link encap:Ethernet HWaddr 00:C0:F0:1F:37:B4
1604 UP BROADCAST RUNNING SLAVE MULTICAST MTU:1500 Metric:1
1605 RX packets:3651769 errors:0 dropped:0 overruns:0 frame:0
1606 TX packets:1643480 errors:0 dropped:0 overruns:0 carrier:0
1607 collisions:0 txqueuelen:100
1608 Interrupt:9 Base address:0x1400
1610 5. Switch Configuration
1611 =======================
1613 For this section, "switch" refers to whatever system the
1614 bonded devices are directly connected to (i.e., where the other end of
1615 the cable plugs into). This may be an actual dedicated switch device,
1616 or it may be another regular system (e.g., another computer running
1619 The active-backup, balance-tlb and balance-alb modes do not
1620 require any specific configuration of the switch.
1622 The 802.3ad mode requires that the switch have the appropriate
1623 ports configured as an 802.3ad aggregation. The precise method used
1624 to configure this varies from switch to switch, but, for example, a
1625 Cisco 3550 series switch requires that the appropriate ports first be
1626 grouped together in a single etherchannel instance, then that
1627 etherchannel is set to mode "lacp" to enable 802.3ad (instead of
1628 standard EtherChannel).
1630 The balance-rr, balance-xor and broadcast modes generally
1631 require that the switch have the appropriate ports grouped together.
1632 The nomenclature for such a group differs between switches, it may be
1633 called an "etherchannel" (as in the Cisco example, above), a "trunk
1634 group" or some other similar variation. For these modes, each switch
1635 will also have its own configuration options for the switch's transmit
1636 policy to the bond. Typical choices include XOR of either the MAC or
1637 IP addresses. The transmit policy of the two peers does not need to
1638 match. For these three modes, the bonding mode really selects a
1639 transmit policy for an EtherChannel group; all three will interoperate
1640 with another EtherChannel group.
1643 6. 802.1q VLAN Support
1644 ======================
1646 It is possible to configure VLAN devices over a bond interface
1647 using the 8021q driver. However, only packets coming from the 8021q
1648 driver and passing through bonding will be tagged by default. Self
1649 generated packets, for example, bonding's learning packets or ARP
1650 packets generated by either ALB mode or the ARP monitor mechanism, are
1651 tagged internally by bonding itself. As a result, bonding must
1652 "learn" the VLAN IDs configured above it, and use those IDs to tag
1653 self generated packets.
1655 For reasons of simplicity, and to support the use of adapters
1656 that can do VLAN hardware acceleration offloading, the bonding
1657 interface declares itself as fully hardware offloading capable, it gets
1658 the add_vid/kill_vid notifications to gather the necessary
1659 information, and it propagates those actions to the slaves. In case
1660 of mixed adapter types, hardware accelerated tagged packets that
1661 should go through an adapter that is not offloading capable are
1662 "un-accelerated" by the bonding driver so the VLAN tag sits in the
1665 VLAN interfaces *must* be added on top of a bonding interface
1666 only after enslaving at least one slave. The bonding interface has a
1667 hardware address of 00:00:00:00:00:00 until the first slave is added.
1668 If the VLAN interface is created prior to the first enslavement, it
1669 would pick up the all-zeroes hardware address. Once the first slave
1670 is attached to the bond, the bond device itself will pick up the
1671 slave's hardware address, which is then available for the VLAN device.
1673 Also, be aware that a similar problem can occur if all slaves
1674 are released from a bond that still has one or more VLAN interfaces on
1675 top of it. When a new slave is added, the bonding interface will
1676 obtain its hardware address from the first slave, which might not
1677 match the hardware address of the VLAN interfaces (which was
1678 ultimately copied from an earlier slave).
1680 There are two methods to insure that the VLAN device operates
1681 with the correct hardware address if all slaves are removed from a
1684 1. Remove all VLAN interfaces then recreate them
1686 2. Set the bonding interface's hardware address so that it
1687 matches the hardware address of the VLAN interfaces.
1689 Note that changing a VLAN interface's HW address would set the
1690 underlying device -- i.e. the bonding interface -- to promiscuous
1691 mode, which might not be what you want.
1697 The bonding driver at present supports two schemes for
1698 monitoring a slave device's link state: the ARP monitor and the MII
1701 At the present time, due to implementation restrictions in the
1702 bonding driver itself, it is not possible to enable both ARP and MII
1703 monitoring simultaneously.
1705 7.1 ARP Monitor Operation
1706 -------------------------
1708 The ARP monitor operates as its name suggests: it sends ARP
1709 queries to one or more designated peer systems on the network, and
1710 uses the response as an indication that the link is operating. This
1711 gives some assurance that traffic is actually flowing to and from one
1712 or more peers on the local network.
1714 The ARP monitor relies on the device driver itself to verify
1715 that traffic is flowing. In particular, the driver must keep up to
1716 date the last receive time, dev->last_rx, and transmit start time,
1717 dev->trans_start. If these are not updated by the driver, then the
1718 ARP monitor will immediately fail any slaves using that driver, and
1719 those slaves will stay down. If networking monitoring (tcpdump, etc)
1720 shows the ARP requests and replies on the network, then it may be that
1721 your device driver is not updating last_rx and trans_start.
1723 7.2 Configuring Multiple ARP Targets
1724 ------------------------------------
1726 While ARP monitoring can be done with just one target, it can
1727 be useful in a High Availability setup to have several targets to
1728 monitor. In the case of just one target, the target itself may go
1729 down or have a problem making it unresponsive to ARP requests. Having
1730 an additional target (or several) increases the reliability of the ARP
1733 Multiple ARP targets must be separated by commas as follows:
1735 # example options for ARP monitoring with three targets
1737 options bond0 arp_interval=60 arp_ip_target=192.168.0.1,192.168.0.3,192.168.0.9
1739 For just a single target the options would resemble:
1741 # example options for ARP monitoring with one target
1743 options bond0 arp_interval=60 arp_ip_target=192.168.0.100
1746 7.3 MII Monitor Operation
1747 -------------------------
1749 The MII monitor monitors only the carrier state of the local
1750 network interface. It accomplishes this in one of three ways: by
1751 depending upon the device driver to maintain its carrier state, by
1752 querying the device's MII registers, or by making an ethtool query to
1755 If the use_carrier module parameter is 1 (the default value),
1756 then the MII monitor will rely on the driver for carrier state
1757 information (via the netif_carrier subsystem). As explained in the
1758 use_carrier parameter information, above, if the MII monitor fails to
1759 detect carrier loss on the device (e.g., when the cable is physically
1760 disconnected), it may be that the driver does not support
1763 If use_carrier is 0, then the MII monitor will first query the
1764 device's (via ioctl) MII registers and check the link state. If that
1765 request fails (not just that it returns carrier down), then the MII
1766 monitor will make an ethtool ETHOOL_GLINK request to attempt to obtain
1767 the same information. If both methods fail (i.e., the driver either
1768 does not support or had some error in processing both the MII register
1769 and ethtool requests), then the MII monitor will assume the link is
1772 8. Potential Sources of Trouble
1773 ===============================
1775 8.1 Adventures in Routing
1776 -------------------------
1778 When bonding is configured, it is important that the slave
1779 devices not have routes that supersede routes of the master (or,
1780 generally, not have routes at all). For example, suppose the bonding
1781 device bond0 has two slaves, eth0 and eth1, and the routing table is
1784 Kernel IP routing table
1785 Destination Gateway Genmask Flags MSS Window irtt Iface
1786 10.0.0.0 0.0.0.0 255.255.0.0 U 40 0 0 eth0
1787 10.0.0.0 0.0.0.0 255.255.0.0 U 40 0 0 eth1
1788 10.0.0.0 0.0.0.0 255.255.0.0 U 40 0 0 bond0
1789 127.0.0.0 0.0.0.0 255.0.0.0 U 40 0 0 lo
1791 This routing configuration will likely still update the
1792 receive/transmit times in the driver (needed by the ARP monitor), but
1793 may bypass the bonding driver (because outgoing traffic to, in this
1794 case, another host on network 10 would use eth0 or eth1 before bond0).
1796 The ARP monitor (and ARP itself) may become confused by this
1797 configuration, because ARP requests (generated by the ARP monitor)
1798 will be sent on one interface (bond0), but the corresponding reply
1799 will arrive on a different interface (eth0). This reply looks to ARP
1800 as an unsolicited ARP reply (because ARP matches replies on an
1801 interface basis), and is discarded. The MII monitor is not affected
1802 by the state of the routing table.
1804 The solution here is simply to insure that slaves do not have
1805 routes of their own, and if for some reason they must, those routes do
1806 not supersede routes of their master. This should generally be the
1807 case, but unusual configurations or errant manual or automatic static
1808 route additions may cause trouble.
1810 8.2 Ethernet Device Renaming
1811 ----------------------------
1813 On systems with network configuration scripts that do not
1814 associate physical devices directly with network interface names (so
1815 that the same physical device always has the same "ethX" name), it may
1816 be necessary to add some special logic to config files in
1819 For example, given a modules.conf containing the following:
1822 options bond0 mode=some-mode miimon=50
1828 If neither eth0 and eth1 are slaves to bond0, then when the
1829 bond0 interface comes up, the devices may end up reordered. This
1830 happens because bonding is loaded first, then its slave device's
1831 drivers are loaded next. Since no other drivers have been loaded,
1832 when the e1000 driver loads, it will receive eth0 and eth1 for its
1833 devices, but the bonding configuration tries to enslave eth2 and eth3
1834 (which may later be assigned to the tg3 devices).
1836 Adding the following:
1838 add above bonding e1000 tg3
1840 causes modprobe to load e1000 then tg3, in that order, when
1841 bonding is loaded. This command is fully documented in the
1842 modules.conf manual page.
1844 On systems utilizing modprobe an equivalent problem can occur.
1845 In this case, the following can be added to config files in
1846 /etc/modprobe.d/ as:
1848 softdep bonding pre: tg3 e1000
1850 This will load tg3 and e1000 modules before loading the bonding one.
1851 Full documentation on this can be found in the modprobe.d and modprobe
1854 8.3. Painfully Slow Or No Failed Link Detection By Miimon
1855 ---------------------------------------------------------
1857 By default, bonding enables the use_carrier option, which
1858 instructs bonding to trust the driver to maintain carrier state.
1860 As discussed in the options section, above, some drivers do
1861 not support the netif_carrier_on/_off link state tracking system.
1862 With use_carrier enabled, bonding will always see these links as up,
1863 regardless of their actual state.
1865 Additionally, other drivers do support netif_carrier, but do
1866 not maintain it in real time, e.g., only polling the link state at
1867 some fixed interval. In this case, miimon will detect failures, but
1868 only after some long period of time has expired. If it appears that
1869 miimon is very slow in detecting link failures, try specifying
1870 use_carrier=0 to see if that improves the failure detection time. If
1871 it does, then it may be that the driver checks the carrier state at a
1872 fixed interval, but does not cache the MII register values (so the
1873 use_carrier=0 method of querying the registers directly works). If
1874 use_carrier=0 does not improve the failover, then the driver may cache
1875 the registers, or the problem may be elsewhere.
1877 Also, remember that miimon only checks for the device's
1878 carrier state. It has no way to determine the state of devices on or
1879 beyond other ports of a switch, or if a switch is refusing to pass
1880 traffic while still maintaining carrier on.
1885 If running SNMP agents, the bonding driver should be loaded
1886 before any network drivers participating in a bond. This requirement
1887 is due to the interface index (ipAdEntIfIndex) being associated to
1888 the first interface found with a given IP address. That is, there is
1889 only one ipAdEntIfIndex for each IP address. For example, if eth0 and
1890 eth1 are slaves of bond0 and the driver for eth0 is loaded before the
1891 bonding driver, the interface for the IP address will be associated
1892 with the eth0 interface. This configuration is shown below, the IP
1893 address 192.168.1.1 has an interface index of 2 which indexes to eth0
1894 in the ifDescr table (ifDescr.2).
1896 interfaces.ifTable.ifEntry.ifDescr.1 = lo
1897 interfaces.ifTable.ifEntry.ifDescr.2 = eth0
1898 interfaces.ifTable.ifEntry.ifDescr.3 = eth1
1899 interfaces.ifTable.ifEntry.ifDescr.4 = eth2
1900 interfaces.ifTable.ifEntry.ifDescr.5 = eth3
1901 interfaces.ifTable.ifEntry.ifDescr.6 = bond0
1902 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.10.10.10 = 5
1903 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.192.168.1.1 = 2
1904 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.74.20.94 = 4
1905 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.127.0.0.1 = 1
1907 This problem is avoided by loading the bonding driver before
1908 any network drivers participating in a bond. Below is an example of
1909 loading the bonding driver first, the IP address 192.168.1.1 is
1910 correctly associated with ifDescr.2.
1912 interfaces.ifTable.ifEntry.ifDescr.1 = lo
1913 interfaces.ifTable.ifEntry.ifDescr.2 = bond0
1914 interfaces.ifTable.ifEntry.ifDescr.3 = eth0
1915 interfaces.ifTable.ifEntry.ifDescr.4 = eth1
1916 interfaces.ifTable.ifEntry.ifDescr.5 = eth2
1917 interfaces.ifTable.ifEntry.ifDescr.6 = eth3
1918 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.10.10.10 = 6
1919 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.192.168.1.1 = 2
1920 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.74.20.94 = 5
1921 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.127.0.0.1 = 1
1923 While some distributions may not report the interface name in
1924 ifDescr, the association between the IP address and IfIndex remains
1925 and SNMP functions such as Interface_Scan_Next will report that
1928 10. Promiscuous mode
1929 ====================
1931 When running network monitoring tools, e.g., tcpdump, it is
1932 common to enable promiscuous mode on the device, so that all traffic
1933 is seen (instead of seeing only traffic destined for the local host).
1934 The bonding driver handles promiscuous mode changes to the bonding
1935 master device (e.g., bond0), and propagates the setting to the slave
1938 For the balance-rr, balance-xor, broadcast, and 802.3ad modes,
1939 the promiscuous mode setting is propagated to all slaves.
1941 For the active-backup, balance-tlb and balance-alb modes, the
1942 promiscuous mode setting is propagated only to the active slave.
1944 For balance-tlb mode, the active slave is the slave currently
1945 receiving inbound traffic.
1947 For balance-alb mode, the active slave is the slave used as a
1948 "primary." This slave is used for mode-specific control traffic, for
1949 sending to peers that are unassigned or if the load is unbalanced.
1951 For the active-backup, balance-tlb and balance-alb modes, when
1952 the active slave changes (e.g., due to a link failure), the
1953 promiscuous setting will be propagated to the new active slave.
1955 11. Configuring Bonding for High Availability
1956 =============================================
1958 High Availability refers to configurations that provide
1959 maximum network availability by having redundant or backup devices,
1960 links or switches between the host and the rest of the world. The
1961 goal is to provide the maximum availability of network connectivity
1962 (i.e., the network always works), even though other configurations
1963 could provide higher throughput.
1965 11.1 High Availability in a Single Switch Topology
1966 --------------------------------------------------
1968 If two hosts (or a host and a single switch) are directly
1969 connected via multiple physical links, then there is no availability
1970 penalty to optimizing for maximum bandwidth. In this case, there is
1971 only one switch (or peer), so if it fails, there is no alternative
1972 access to fail over to. Additionally, the bonding load balance modes
1973 support link monitoring of their members, so if individual links fail,
1974 the load will be rebalanced across the remaining devices.
1976 See Section 12, "Configuring Bonding for Maximum Throughput"
1977 for information on configuring bonding with one peer device.
1979 11.2 High Availability in a Multiple Switch Topology
1980 ----------------------------------------------------
1982 With multiple switches, the configuration of bonding and the
1983 network changes dramatically. In multiple switch topologies, there is
1984 a trade off between network availability and usable bandwidth.
1986 Below is a sample network, configured to maximize the
1987 availability of the network:
1991 +-----+----+ +-----+----+
1992 | |port2 ISL port2| |
1993 | switch A +--------------------------+ switch B |
1995 +-----+----+ +-----++---+
1998 +-------------+ host1 +---------------+
2001 In this configuration, there is a link between the two
2002 switches (ISL, or inter switch link), and multiple ports connecting to
2003 the outside world ("port3" on each switch). There is no technical
2004 reason that this could not be extended to a third switch.
2006 11.2.1 HA Bonding Mode Selection for Multiple Switch Topology
2007 -------------------------------------------------------------
2009 In a topology such as the example above, the active-backup and
2010 broadcast modes are the only useful bonding modes when optimizing for
2011 availability; the other modes require all links to terminate on the
2012 same peer for them to behave rationally.
2014 active-backup: This is generally the preferred mode, particularly if
2015 the switches have an ISL and play together well. If the
2016 network configuration is such that one switch is specifically
2017 a backup switch (e.g., has lower capacity, higher cost, etc),
2018 then the primary option can be used to insure that the
2019 preferred link is always used when it is available.
2021 broadcast: This mode is really a special purpose mode, and is suitable
2022 only for very specific needs. For example, if the two
2023 switches are not connected (no ISL), and the networks beyond
2024 them are totally independent. In this case, if it is
2025 necessary for some specific one-way traffic to reach both
2026 independent networks, then the broadcast mode may be suitable.
2028 11.2.2 HA Link Monitoring Selection for Multiple Switch Topology
2029 ----------------------------------------------------------------
2031 The choice of link monitoring ultimately depends upon your
2032 switch. If the switch can reliably fail ports in response to other
2033 failures, then either the MII or ARP monitors should work. For
2034 example, in the above example, if the "port3" link fails at the remote
2035 end, the MII monitor has no direct means to detect this. The ARP
2036 monitor could be configured with a target at the remote end of port3,
2037 thus detecting that failure without switch support.
2039 In general, however, in a multiple switch topology, the ARP
2040 monitor can provide a higher level of reliability in detecting end to
2041 end connectivity failures (which may be caused by the failure of any
2042 individual component to pass traffic for any reason). Additionally,
2043 the ARP monitor should be configured with multiple targets (at least
2044 one for each switch in the network). This will insure that,
2045 regardless of which switch is active, the ARP monitor has a suitable
2048 Note, also, that of late many switches now support a functionality
2049 generally referred to as "trunk failover." This is a feature of the
2050 switch that causes the link state of a particular switch port to be set
2051 down (or up) when the state of another switch port goes down (or up).
2052 Its purpose is to propagate link failures from logically "exterior" ports
2053 to the logically "interior" ports that bonding is able to monitor via
2054 miimon. Availability and configuration for trunk failover varies by
2055 switch, but this can be a viable alternative to the ARP monitor when using
2058 12. Configuring Bonding for Maximum Throughput
2059 ==============================================
2061 12.1 Maximizing Throughput in a Single Switch Topology
2062 ------------------------------------------------------
2064 In a single switch configuration, the best method to maximize
2065 throughput depends upon the application and network environment. The
2066 various load balancing modes each have strengths and weaknesses in
2067 different environments, as detailed below.
2069 For this discussion, we will break down the topologies into
2070 two categories. Depending upon the destination of most traffic, we
2071 categorize them into either "gatewayed" or "local" configurations.
2073 In a gatewayed configuration, the "switch" is acting primarily
2074 as a router, and the majority of traffic passes through this router to
2075 other networks. An example would be the following:
2078 +----------+ +----------+
2079 | |eth0 port1| | to other networks
2080 | Host A +---------------------+ router +------------------->
2081 | +---------------------+ | Hosts B and C are out
2082 | |eth1 port2| | here somewhere
2083 +----------+ +----------+
2085 The router may be a dedicated router device, or another host
2086 acting as a gateway. For our discussion, the important point is that
2087 the majority of traffic from Host A will pass through the router to
2088 some other network before reaching its final destination.
2090 In a gatewayed network configuration, although Host A may
2091 communicate with many other systems, all of its traffic will be sent
2092 and received via one other peer on the local network, the router.
2094 Note that the case of two systems connected directly via
2095 multiple physical links is, for purposes of configuring bonding, the
2096 same as a gatewayed configuration. In that case, it happens that all
2097 traffic is destined for the "gateway" itself, not some other network
2100 In a local configuration, the "switch" is acting primarily as
2101 a switch, and the majority of traffic passes through this switch to
2102 reach other stations on the same network. An example would be the
2105 +----------+ +----------+ +--------+
2106 | |eth0 port1| +-------+ Host B |
2107 | Host A +------------+ switch |port3 +--------+
2108 | +------------+ | +--------+
2109 | |eth1 port2| +------------------+ Host C |
2110 +----------+ +----------+port4 +--------+
2113 Again, the switch may be a dedicated switch device, or another
2114 host acting as a gateway. For our discussion, the important point is
2115 that the majority of traffic from Host A is destined for other hosts
2116 on the same local network (Hosts B and C in the above example).
2118 In summary, in a gatewayed configuration, traffic to and from
2119 the bonded device will be to the same MAC level peer on the network
2120 (the gateway itself, i.e., the router), regardless of its final
2121 destination. In a local configuration, traffic flows directly to and
2122 from the final destinations, thus, each destination (Host B, Host C)
2123 will be addressed directly by their individual MAC addresses.
2125 This distinction between a gatewayed and a local network
2126 configuration is important because many of the load balancing modes
2127 available use the MAC addresses of the local network source and
2128 destination to make load balancing decisions. The behavior of each
2129 mode is described below.
2132 12.1.1 MT Bonding Mode Selection for Single Switch Topology
2133 -----------------------------------------------------------
2135 This configuration is the easiest to set up and to understand,
2136 although you will have to decide which bonding mode best suits your
2137 needs. The trade offs for each mode are detailed below:
2139 balance-rr: This mode is the only mode that will permit a single
2140 TCP/IP connection to stripe traffic across multiple
2141 interfaces. It is therefore the only mode that will allow a
2142 single TCP/IP stream to utilize more than one interface's
2143 worth of throughput. This comes at a cost, however: the
2144 striping generally results in peer systems receiving packets out
2145 of order, causing TCP/IP's congestion control system to kick
2146 in, often by retransmitting segments.
2148 It is possible to adjust TCP/IP's congestion limits by
2149 altering the net.ipv4.tcp_reordering sysctl parameter. The
2150 usual default value is 3, and the maximum useful value is 127.
2151 For a four interface balance-rr bond, expect that a single
2152 TCP/IP stream will utilize no more than approximately 2.3
2153 interface's worth of throughput, even after adjusting
2156 Note that the fraction of packets that will be delivered out of
2157 order is highly variable, and is unlikely to be zero. The level
2158 of reordering depends upon a variety of factors, including the
2159 networking interfaces, the switch, and the topology of the
2160 configuration. Speaking in general terms, higher speed network
2161 cards produce more reordering (due to factors such as packet
2162 coalescing), and a "many to many" topology will reorder at a
2163 higher rate than a "many slow to one fast" configuration.
2165 Many switches do not support any modes that stripe traffic
2166 (instead choosing a port based upon IP or MAC level addresses);
2167 for those devices, traffic for a particular connection flowing
2168 through the switch to a balance-rr bond will not utilize greater
2169 than one interface's worth of bandwidth.
2171 If you are utilizing protocols other than TCP/IP, UDP for
2172 example, and your application can tolerate out of order
2173 delivery, then this mode can allow for single stream datagram
2174 performance that scales near linearly as interfaces are added
2177 This mode requires the switch to have the appropriate ports
2178 configured for "etherchannel" or "trunking."
2180 active-backup: There is not much advantage in this network topology to
2181 the active-backup mode, as the inactive backup devices are all
2182 connected to the same peer as the primary. In this case, a
2183 load balancing mode (with link monitoring) will provide the
2184 same level of network availability, but with increased
2185 available bandwidth. On the plus side, active-backup mode
2186 does not require any configuration of the switch, so it may
2187 have value if the hardware available does not support any of
2188 the load balance modes.
2190 balance-xor: This mode will limit traffic such that packets destined
2191 for specific peers will always be sent over the same
2192 interface. Since the destination is determined by the MAC
2193 addresses involved, this mode works best in a "local" network
2194 configuration (as described above), with destinations all on
2195 the same local network. This mode is likely to be suboptimal
2196 if all your traffic is passed through a single router (i.e., a
2197 "gatewayed" network configuration, as described above).
2199 As with balance-rr, the switch ports need to be configured for
2200 "etherchannel" or "trunking."
2202 broadcast: Like active-backup, there is not much advantage to this
2203 mode in this type of network topology.
2205 802.3ad: This mode can be a good choice for this type of network
2206 topology. The 802.3ad mode is an IEEE standard, so all peers
2207 that implement 802.3ad should interoperate well. The 802.3ad
2208 protocol includes automatic configuration of the aggregates,
2209 so minimal manual configuration of the switch is needed
2210 (typically only to designate that some set of devices is
2211 available for 802.3ad). The 802.3ad standard also mandates
2212 that frames be delivered in order (within certain limits), so
2213 in general single connections will not see misordering of
2214 packets. The 802.3ad mode does have some drawbacks: the
2215 standard mandates that all devices in the aggregate operate at
2216 the same speed and duplex. Also, as with all bonding load
2217 balance modes other than balance-rr, no single connection will
2218 be able to utilize more than a single interface's worth of
2221 Additionally, the linux bonding 802.3ad implementation
2222 distributes traffic by peer (using an XOR of MAC addresses),
2223 so in a "gatewayed" configuration, all outgoing traffic will
2224 generally use the same device. Incoming traffic may also end
2225 up on a single device, but that is dependent upon the
2226 balancing policy of the peer's 8023.ad implementation. In a
2227 "local" configuration, traffic will be distributed across the
2228 devices in the bond.
2230 Finally, the 802.3ad mode mandates the use of the MII monitor,
2231 therefore, the ARP monitor is not available in this mode.
2233 balance-tlb: The balance-tlb mode balances outgoing traffic by peer.
2234 Since the balancing is done according to MAC address, in a
2235 "gatewayed" configuration (as described above), this mode will
2236 send all traffic across a single device. However, in a
2237 "local" network configuration, this mode balances multiple
2238 local network peers across devices in a vaguely intelligent
2239 manner (not a simple XOR as in balance-xor or 802.3ad mode),
2240 so that mathematically unlucky MAC addresses (i.e., ones that
2241 XOR to the same value) will not all "bunch up" on a single
2244 Unlike 802.3ad, interfaces may be of differing speeds, and no
2245 special switch configuration is required. On the down side,
2246 in this mode all incoming traffic arrives over a single
2247 interface, this mode requires certain ethtool support in the
2248 network device driver of the slave interfaces, and the ARP
2249 monitor is not available.
2251 balance-alb: This mode is everything that balance-tlb is, and more.
2252 It has all of the features (and restrictions) of balance-tlb,
2253 and will also balance incoming traffic from local network
2254 peers (as described in the Bonding Module Options section,
2257 The only additional down side to this mode is that the network
2258 device driver must support changing the hardware address while
2261 12.1.2 MT Link Monitoring for Single Switch Topology
2262 ----------------------------------------------------
2264 The choice of link monitoring may largely depend upon which
2265 mode you choose to use. The more advanced load balancing modes do not
2266 support the use of the ARP monitor, and are thus restricted to using
2267 the MII monitor (which does not provide as high a level of end to end
2268 assurance as the ARP monitor).
2270 12.2 Maximum Throughput in a Multiple Switch Topology
2271 -----------------------------------------------------
2273 Multiple switches may be utilized to optimize for throughput
2274 when they are configured in parallel as part of an isolated network
2275 between two or more systems, for example:
2281 +--------+ | +---------+
2283 +------+---+ +-----+----+ +-----+----+
2284 | Switch A | | Switch B | | Switch C |
2285 +------+---+ +-----+----+ +-----+----+
2287 +--------+ | +---------+
2293 In this configuration, the switches are isolated from one
2294 another. One reason to employ a topology such as this is for an
2295 isolated network with many hosts (a cluster configured for high
2296 performance, for example), using multiple smaller switches can be more
2297 cost effective than a single larger switch, e.g., on a network with 24
2298 hosts, three 24 port switches can be significantly less expensive than
2299 a single 72 port switch.
2301 If access beyond the network is required, an individual host
2302 can be equipped with an additional network device connected to an
2303 external network; this host then additionally acts as a gateway.
2305 12.2.1 MT Bonding Mode Selection for Multiple Switch Topology
2306 -------------------------------------------------------------
2308 In actual practice, the bonding mode typically employed in
2309 configurations of this type is balance-rr. Historically, in this
2310 network configuration, the usual caveats about out of order packet
2311 delivery are mitigated by the use of network adapters that do not do
2312 any kind of packet coalescing (via the use of NAPI, or because the
2313 device itself does not generate interrupts until some number of
2314 packets has arrived). When employed in this fashion, the balance-rr
2315 mode allows individual connections between two hosts to effectively
2316 utilize greater than one interface's bandwidth.
2318 12.2.2 MT Link Monitoring for Multiple Switch Topology
2319 ------------------------------------------------------
2321 Again, in actual practice, the MII monitor is most often used
2322 in this configuration, as performance is given preference over
2323 availability. The ARP monitor will function in this topology, but its
2324 advantages over the MII monitor are mitigated by the volume of probes
2325 needed as the number of systems involved grows (remember that each
2326 host in the network is configured with bonding).
2328 13. Switch Behavior Issues
2329 ==========================
2331 13.1 Link Establishment and Failover Delays
2332 -------------------------------------------
2334 Some switches exhibit undesirable behavior with regard to the
2335 timing of link up and down reporting by the switch.
2337 First, when a link comes up, some switches may indicate that
2338 the link is up (carrier available), but not pass traffic over the
2339 interface for some period of time. This delay is typically due to
2340 some type of autonegotiation or routing protocol, but may also occur
2341 during switch initialization (e.g., during recovery after a switch
2342 failure). If you find this to be a problem, specify an appropriate
2343 value to the updelay bonding module option to delay the use of the
2344 relevant interface(s).
2346 Second, some switches may "bounce" the link state one or more
2347 times while a link is changing state. This occurs most commonly while
2348 the switch is initializing. Again, an appropriate updelay value may
2351 Note that when a bonding interface has no active links, the
2352 driver will immediately reuse the first link that goes up, even if the
2353 updelay parameter has been specified (the updelay is ignored in this
2354 case). If there are slave interfaces waiting for the updelay timeout
2355 to expire, the interface that first went into that state will be
2356 immediately reused. This reduces down time of the network if the
2357 value of updelay has been overestimated, and since this occurs only in
2358 cases with no connectivity, there is no additional penalty for
2359 ignoring the updelay.
2361 In addition to the concerns about switch timings, if your
2362 switches take a long time to go into backup mode, it may be desirable
2363 to not activate a backup interface immediately after a link goes down.
2364 Failover may be delayed via the downdelay bonding module option.
2366 13.2 Duplicated Incoming Packets
2367 --------------------------------
2369 NOTE: Starting with version 3.0.2, the bonding driver has logic to
2370 suppress duplicate packets, which should largely eliminate this problem.
2371 The following description is kept for reference.
2373 It is not uncommon to observe a short burst of duplicated
2374 traffic when the bonding device is first used, or after it has been
2375 idle for some period of time. This is most easily observed by issuing
2376 a "ping" to some other host on the network, and noticing that the
2377 output from ping flags duplicates (typically one per slave).
2379 For example, on a bond in active-backup mode with five slaves
2380 all connected to one switch, the output may appear as follows:
2383 PING 10.0.4.2 (10.0.4.2) from 10.0.3.10 : 56(84) bytes of data.
2384 64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.7 ms
2385 64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
2386 64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
2387 64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
2388 64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
2389 64 bytes from 10.0.4.2: icmp_seq=2 ttl=64 time=0.216 ms
2390 64 bytes from 10.0.4.2: icmp_seq=3 ttl=64 time=0.267 ms
2391 64 bytes from 10.0.4.2: icmp_seq=4 ttl=64 time=0.222 ms
2393 This is not due to an error in the bonding driver, rather, it
2394 is a side effect of how many switches update their MAC forwarding
2395 tables. Initially, the switch does not associate the MAC address in
2396 the packet with a particular switch port, and so it may send the
2397 traffic to all ports until its MAC forwarding table is updated. Since
2398 the interfaces attached to the bond may occupy multiple ports on a
2399 single switch, when the switch (temporarily) floods the traffic to all
2400 ports, the bond device receives multiple copies of the same packet
2401 (one per slave device).
2403 The duplicated packet behavior is switch dependent, some
2404 switches exhibit this, and some do not. On switches that display this
2405 behavior, it can be induced by clearing the MAC forwarding table (on
2406 most Cisco switches, the privileged command "clear mac address-table
2407 dynamic" will accomplish this).
2409 14. Hardware Specific Considerations
2410 ====================================
2412 This section contains additional information for configuring
2413 bonding on specific hardware platforms, or for interfacing bonding
2414 with particular switches or other devices.
2416 14.1 IBM BladeCenter
2417 --------------------
2419 This applies to the JS20 and similar systems.
2421 On the JS20 blades, the bonding driver supports only
2422 balance-rr, active-backup, balance-tlb and balance-alb modes. This is
2423 largely due to the network topology inside the BladeCenter, detailed
2426 JS20 network adapter information
2427 --------------------------------
2429 All JS20s come with two Broadcom Gigabit Ethernet ports
2430 integrated on the planar (that's "motherboard" in IBM-speak). In the
2431 BladeCenter chassis, the eth0 port of all JS20 blades is hard wired to
2432 I/O Module #1; similarly, all eth1 ports are wired to I/O Module #2.
2433 An add-on Broadcom daughter card can be installed on a JS20 to provide
2434 two more Gigabit Ethernet ports. These ports, eth2 and eth3, are
2435 wired to I/O Modules 3 and 4, respectively.
2437 Each I/O Module may contain either a switch or a passthrough
2438 module (which allows ports to be directly connected to an external
2439 switch). Some bonding modes require a specific BladeCenter internal
2440 network topology in order to function; these are detailed below.
2442 Additional BladeCenter-specific networking information can be
2443 found in two IBM Redbooks (www.ibm.com/redbooks):
2445 "IBM eServer BladeCenter Networking Options"
2446 "IBM eServer BladeCenter Layer 2-7 Network Switching"
2448 BladeCenter networking configuration
2449 ------------------------------------
2451 Because a BladeCenter can be configured in a very large number
2452 of ways, this discussion will be confined to describing basic
2455 Normally, Ethernet Switch Modules (ESMs) are used in I/O
2456 modules 1 and 2. In this configuration, the eth0 and eth1 ports of a
2457 JS20 will be connected to different internal switches (in the
2458 respective I/O modules).
2460 A passthrough module (OPM or CPM, optical or copper,
2461 passthrough module) connects the I/O module directly to an external
2462 switch. By using PMs in I/O module #1 and #2, the eth0 and eth1
2463 interfaces of a JS20 can be redirected to the outside world and
2464 connected to a common external switch.
2466 Depending upon the mix of ESMs and PMs, the network will
2467 appear to bonding as either a single switch topology (all PMs) or as a
2468 multiple switch topology (one or more ESMs, zero or more PMs). It is
2469 also possible to connect ESMs together, resulting in a configuration
2470 much like the example in "High Availability in a Multiple Switch
2473 Requirements for specific modes
2474 -------------------------------
2476 The balance-rr mode requires the use of passthrough modules
2477 for devices in the bond, all connected to an common external switch.
2478 That switch must be configured for "etherchannel" or "trunking" on the
2479 appropriate ports, as is usual for balance-rr.
2481 The balance-alb and balance-tlb modes will function with
2482 either switch modules or passthrough modules (or a mix). The only
2483 specific requirement for these modes is that all network interfaces
2484 must be able to reach all destinations for traffic sent over the
2485 bonding device (i.e., the network must converge at some point outside
2488 The active-backup mode has no additional requirements.
2490 Link monitoring issues
2491 ----------------------
2493 When an Ethernet Switch Module is in place, only the ARP
2494 monitor will reliably detect link loss to an external switch. This is
2495 nothing unusual, but examination of the BladeCenter cabinet would
2496 suggest that the "external" network ports are the ethernet ports for
2497 the system, when it fact there is a switch between these "external"
2498 ports and the devices on the JS20 system itself. The MII monitor is
2499 only able to detect link failures between the ESM and the JS20 system.
2501 When a passthrough module is in place, the MII monitor does
2502 detect failures to the "external" port, which is then directly
2503 connected to the JS20 system.
2508 The Serial Over LAN (SoL) link is established over the primary
2509 ethernet (eth0) only, therefore, any loss of link to eth0 will result
2510 in losing your SoL connection. It will not fail over with other
2511 network traffic, as the SoL system is beyond the control of the
2514 It may be desirable to disable spanning tree on the switch
2515 (either the internal Ethernet Switch Module, or an external switch) to
2516 avoid fail-over delay issues when using bonding.
2519 15. Frequently Asked Questions
2520 ==============================
2524 Yes. The old 2.0.xx channel bonding patch was not SMP safe.
2525 The new driver was designed to be SMP safe from the start.
2527 2. What type of cards will work with it?
2529 Any Ethernet type cards (you can even mix cards - a Intel
2530 EtherExpress PRO/100 and a 3com 3c905b, for example). For most modes,
2531 devices need not be of the same speed.
2533 Starting with version 3.2.1, bonding also supports Infiniband
2534 slaves in active-backup mode.
2536 3. How many bonding devices can I have?
2540 4. How many slaves can a bonding device have?
2542 This is limited only by the number of network interfaces Linux
2543 supports and/or the number of network cards you can place in your
2546 5. What happens when a slave link dies?
2548 If link monitoring is enabled, then the failing device will be
2549 disabled. The active-backup mode will fail over to a backup link, and
2550 other modes will ignore the failed link. The link will continue to be
2551 monitored, and should it recover, it will rejoin the bond (in whatever
2552 manner is appropriate for the mode). See the sections on High
2553 Availability and the documentation for each mode for additional
2556 Link monitoring can be enabled via either the miimon or
2557 arp_interval parameters (described in the module parameters section,
2558 above). In general, miimon monitors the carrier state as sensed by
2559 the underlying network device, and the arp monitor (arp_interval)
2560 monitors connectivity to another host on the local network.
2562 If no link monitoring is configured, the bonding driver will
2563 be unable to detect link failures, and will assume that all links are
2564 always available. This will likely result in lost packets, and a
2565 resulting degradation of performance. The precise performance loss
2566 depends upon the bonding mode and network configuration.
2568 6. Can bonding be used for High Availability?
2570 Yes. See the section on High Availability for details.
2572 7. Which switches/systems does it work with?
2574 The full answer to this depends upon the desired mode.
2576 In the basic balance modes (balance-rr and balance-xor), it
2577 works with any system that supports etherchannel (also called
2578 trunking). Most managed switches currently available have such
2579 support, and many unmanaged switches as well.
2581 The advanced balance modes (balance-tlb and balance-alb) do
2582 not have special switch requirements, but do need device drivers that
2583 support specific features (described in the appropriate section under
2584 module parameters, above).
2586 In 802.3ad mode, it works with systems that support IEEE
2587 802.3ad Dynamic Link Aggregation. Most managed and many unmanaged
2588 switches currently available support 802.3ad.
2590 The active-backup mode should work with any Layer-II switch.
2592 8. Where does a bonding device get its MAC address from?
2594 When using slave devices that have fixed MAC addresses, or when
2595 the fail_over_mac option is enabled, the bonding device's MAC address is
2596 the MAC address of the active slave.
2598 For other configurations, if not explicitly configured (with
2599 ifconfig or ip link), the MAC address of the bonding device is taken from
2600 its first slave device. This MAC address is then passed to all following
2601 slaves and remains persistent (even if the first slave is removed) until
2602 the bonding device is brought down or reconfigured.
2604 If you wish to change the MAC address, you can set it with
2605 ifconfig or ip link:
2607 # ifconfig bond0 hw ether 00:11:22:33:44:55
2609 # ip link set bond0 address 66:77:88:99:aa:bb
2611 The MAC address can be also changed by bringing down/up the
2612 device and then changing its slaves (or their order):
2614 # ifconfig bond0 down ; modprobe -r bonding
2615 # ifconfig bond0 .... up
2616 # ifenslave bond0 eth...
2618 This method will automatically take the address from the next
2619 slave that is added.
2621 To restore your slaves' MAC addresses, you need to detach them
2622 from the bond (`ifenslave -d bond0 eth0'). The bonding driver will
2623 then restore the MAC addresses that the slaves had before they were
2626 16. Resources and Links
2627 =======================
2629 The latest version of the bonding driver can be found in the latest
2630 version of the linux kernel, found on http://kernel.org
2632 The latest version of this document can be found in the latest kernel
2633 source (named Documentation/networking/bonding.txt).
2635 Discussions regarding the usage of the bonding driver take place on the
2636 bonding-devel mailing list, hosted at sourceforge.net. If you have questions or
2637 problems, post them to the list. The list address is:
2639 bonding-devel@lists.sourceforge.net
2641 The administrative interface (to subscribe or unsubscribe) can
2644 https://lists.sourceforge.net/lists/listinfo/bonding-devel
2646 Discussions regarding the development of the bonding driver take place
2647 on the main Linux network mailing list, hosted at vger.kernel.org. The list
2650 netdev@vger.kernel.org
2652 The administrative interface (to subscribe or unsubscribe) can
2655 http://vger.kernel.org/vger-lists.html#netdev
2657 Donald Becker's Ethernet Drivers and diag programs may be found at :
2658 - http://web.archive.org/web/*/http://www.scyld.com/network/
2660 You will also find a lot of information regarding Ethernet, NWay, MII,
2661 etc. at www.scyld.com.