2 Linux Ethernet Bonding Driver HOWTO
4 Latest update: 23 September 2009
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 Overriding Configuration for Special Cases
54 4. Querying Bonding Configuration
55 4.1 Bonding Configuration
56 4.2 Network Configuration
58 5. Switch Configuration
60 6. 802.1q VLAN Support
63 7.1 ARP Monitor Operation
64 7.2 Configuring Multiple ARP Targets
65 7.3 MII Monitor Operation
67 8. Potential Trouble Sources
68 8.1 Adventures in Routing
69 8.2 Ethernet Device Renaming
70 8.3 Painfully Slow Or No Failed Link Detection By Miimon
76 11. Configuring Bonding for High Availability
77 11.1 High Availability in a Single Switch Topology
78 11.2 High Availability in a Multiple Switch Topology
79 11.2.1 HA Bonding Mode Selection for Multiple Switch Topology
80 11.2.2 HA Link Monitoring for Multiple Switch Topology
82 12. Configuring Bonding for Maximum Throughput
83 12.1 Maximum Throughput in a Single Switch Topology
84 12.1.1 MT Bonding Mode Selection for Single Switch Topology
85 12.1.2 MT Link Monitoring for Single Switch Topology
86 12.2 Maximum Throughput in a Multiple Switch Topology
87 12.2.1 MT Bonding Mode Selection for Multiple Switch Topology
88 12.2.2 MT Link Monitoring for Multiple Switch Topology
90 13. Switch Behavior Issues
91 13.1 Link Establishment and Failover Delays
92 13.2 Duplicated Incoming Packets
94 14. Hardware Specific Considerations
97 15. Frequently Asked Questions
99 16. Resources and Links
102 1. Bonding Driver Installation
103 ==============================
105 Most popular distro kernels ship with the bonding driver
106 already available as a module and the ifenslave user level control
107 program installed and ready for use. 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, then continue
127 below to install ifenslave.
129 1.2 Install ifenslave Control Utility
130 -------------------------------------
132 The ifenslave user level control program is included in the
133 kernel source tree, in the file Documentation/networking/ifenslave.c.
134 It is generally recommended that you use the ifenslave that
135 corresponds to the kernel that you are using (either from the same
136 source tree or supplied with the distro), however, ifenslave
137 executables from older kernels should function (but features newer
138 than the ifenslave release are not supported). Running an ifenslave
139 that is newer than the kernel is not supported, and may or may not
142 To install ifenslave, do the following:
144 # gcc -Wall -O -I/usr/src/linux/include ifenslave.c -o ifenslave
145 # cp ifenslave /sbin/ifenslave
147 If your kernel source is not in "/usr/src/linux," then replace
148 "/usr/src/linux/include" in the above with the location of your kernel
149 source include directory.
151 You may wish to back up any existing /sbin/ifenslave, or, for
152 testing or informal use, tag the ifenslave to the kernel version
153 (e.g., name the ifenslave executable /sbin/ifenslave-2.6.10).
157 If you omit the "-I" or specify an incorrect directory, you
158 may end up with an ifenslave that is incompatible with the kernel
159 you're trying to build it for. Some distros (e.g., Red Hat from 7.1
160 onwards) do not have /usr/include/linux symbolically linked to the
161 default kernel source include directory.
163 SECOND IMPORTANT NOTE:
164 If you plan to configure bonding using sysfs, you do not need
167 2. Bonding Driver Options
168 =========================
170 Options for the bonding driver are supplied as parameters to the
171 bonding module at load time, or are specified via sysfs.
173 Module options may be given as command line arguments to the
174 insmod or modprobe command, but are usually specified in either the
175 /etc/modules.conf or /etc/modprobe.conf configuration file, or in a
176 distro-specific configuration file (some of which are detailed in the next
179 Details on bonding support for sysfs is provided in the
180 "Configuring Bonding Manually via Sysfs" section, below.
182 The available bonding driver parameters are listed below. If a
183 parameter is not specified the default value is used. When initially
184 configuring a bond, it is recommended "tail -f /var/log/messages" be
185 run in a separate window to watch for bonding driver error messages.
187 It is critical that either the miimon or arp_interval and
188 arp_ip_target parameters be specified, otherwise serious network
189 degradation will occur during link failures. Very few devices do not
190 support at least miimon, so there is really no reason not to use it.
192 Options with textual values will accept either the text name
193 or, for backwards compatibility, the option value. E.g.,
194 "mode=802.3ad" and "mode=4" set the same mode.
196 The parameters are as follows:
200 Specifies the 802.3ad aggregation selection logic to use. The
201 possible values and their effects are:
205 The active aggregator is chosen by largest aggregate
208 Reselection of the active aggregator occurs only when all
209 slaves of the active aggregator are down or the active
210 aggregator has no slaves.
212 This is the default value.
216 The active aggregator is chosen by largest aggregate
217 bandwidth. Reselection occurs if:
219 - A slave is added to or removed from the bond
221 - Any slave's link state changes
223 - Any slave's 802.3ad association state changes
225 - The bond's administrative state changes to up
229 The active aggregator is chosen by the largest number of
230 ports (slaves). Reselection occurs as described under the
231 "bandwidth" setting, above.
233 The bandwidth and count selection policies permit failover of
234 802.3ad aggregations when partial failure of the active aggregator
235 occurs. This keeps the aggregator with the highest availability
236 (either in bandwidth or in number of ports) active at all times.
238 This option was added in bonding version 3.4.0.
242 Specifies the ARP link monitoring frequency in milliseconds.
244 The ARP monitor works by periodically checking the slave
245 devices to determine whether they have sent or received
246 traffic recently (the precise criteria depends upon the
247 bonding mode, and the state of the slave). Regular traffic is
248 generated via ARP probes issued for the addresses specified by
249 the arp_ip_target option.
251 This behavior can be modified by the arp_validate option,
254 If ARP monitoring is used in an etherchannel compatible mode
255 (modes 0 and 2), the switch should be configured in a mode
256 that evenly distributes packets across all links. If the
257 switch is configured to distribute the packets in an XOR
258 fashion, all replies from the ARP targets will be received on
259 the same link which could cause the other team members to
260 fail. ARP monitoring should not be used in conjunction with
261 miimon. A value of 0 disables ARP monitoring. The default
266 Specifies the IP addresses to use as ARP monitoring peers when
267 arp_interval is > 0. These are the targets of the ARP request
268 sent to determine the health of the link to the targets.
269 Specify these values in ddd.ddd.ddd.ddd format. Multiple IP
270 addresses must be separated by a comma. At least one IP
271 address must be given for ARP monitoring to function. The
272 maximum number of targets that can be specified is 16. The
273 default value is no IP addresses.
277 Specifies whether or not ARP probes and replies should be
278 validated in the active-backup mode. This causes the ARP
279 monitor to examine the incoming ARP requests and replies, and
280 only consider a slave to be up if it is receiving the
281 appropriate ARP traffic.
287 No validation is performed. This is the default.
291 Validation is performed only for the active slave.
295 Validation is performed only for backup slaves.
299 Validation is performed for all slaves.
301 For the active slave, the validation checks ARP replies to
302 confirm that they were generated by an arp_ip_target. Since
303 backup slaves do not typically receive these replies, the
304 validation performed for backup slaves is on the ARP request
305 sent out via the active slave. It is possible that some
306 switch or network configurations may result in situations
307 wherein the backup slaves do not receive the ARP requests; in
308 such a situation, validation of backup slaves must be
311 This option is useful in network configurations in which
312 multiple bonding hosts are concurrently issuing ARPs to one or
313 more targets beyond a common switch. Should the link between
314 the switch and target fail (but not the switch itself), the
315 probe traffic generated by the multiple bonding instances will
316 fool the standard ARP monitor into considering the links as
317 still up. Use of the arp_validate option can resolve this, as
318 the ARP monitor will only consider ARP requests and replies
319 associated with its own instance of bonding.
321 This option was added in bonding version 3.1.0.
325 Specifies the time, in milliseconds, to wait before disabling
326 a slave after a link failure has been detected. This option
327 is only valid for the miimon link monitor. The downdelay
328 value should be a multiple of the miimon value; if not, it
329 will be rounded down to the nearest multiple. The default
334 Specifies whether active-backup mode should set all slaves to
335 the same MAC address at enslavement (the traditional
336 behavior), or, when enabled, perform special handling of the
337 bond's MAC address in accordance with the selected policy.
343 This setting disables fail_over_mac, and causes
344 bonding to set all slaves of an active-backup bond to
345 the same MAC address at enslavement time. This is the
350 The "active" fail_over_mac policy indicates that the
351 MAC address of the bond should always be the MAC
352 address of the currently active slave. The MAC
353 address of the slaves is not changed; instead, the MAC
354 address of the bond changes during a failover.
356 This policy is useful for devices that cannot ever
357 alter their MAC address, or for devices that refuse
358 incoming broadcasts with their own source MAC (which
359 interferes with the ARP monitor).
361 The down side of this policy is that every device on
362 the network must be updated via gratuitous ARP,
363 vs. just updating a switch or set of switches (which
364 often takes place for any traffic, not just ARP
365 traffic, if the switch snoops incoming traffic to
366 update its tables) for the traditional method. If the
367 gratuitous ARP is lost, communication may be
370 When this policy is used in conjuction with the mii
371 monitor, devices which assert link up prior to being
372 able to actually transmit and receive are particularly
373 susceptible to loss of the gratuitous ARP, and an
374 appropriate updelay setting may be required.
378 The "follow" fail_over_mac policy causes the MAC
379 address of the bond to be selected normally (normally
380 the MAC address of the first slave added to the bond).
381 However, the second and subsequent slaves are not set
382 to this MAC address while they are in a backup role; a
383 slave is programmed with the bond's MAC address at
384 failover time (and the formerly active slave receives
385 the newly active slave's MAC address).
387 This policy is useful for multiport devices that
388 either become confused or incur a performance penalty
389 when multiple ports are programmed with the same MAC
393 The default policy is none, unless the first slave cannot
394 change its MAC address, in which case the active policy is
397 This option may be modified via sysfs only when no slaves are
400 This option was added in bonding version 3.2.0. The "follow"
401 policy was added in bonding version 3.3.0.
405 Option specifying the rate in which we'll ask our link partner
406 to transmit LACPDU packets in 802.3ad mode. Possible values
410 Request partner to transmit LACPDUs every 30 seconds
413 Request partner to transmit LACPDUs every 1 second
419 Specifies the number of bonding devices to create for this
420 instance of the bonding driver. E.g., if max_bonds is 3, and
421 the bonding driver is not already loaded, then bond0, bond1
422 and bond2 will be created. The default value is 1. Specifying
423 a value of 0 will load bonding, but will not create any devices.
427 Specifies the MII link monitoring frequency in milliseconds.
428 This determines how often the link state of each slave is
429 inspected for link failures. A value of zero disables MII
430 link monitoring. A value of 100 is a good starting point.
431 The use_carrier option, below, affects how the link state is
432 determined. See the High Availability section for additional
433 information. The default value is 0.
437 Specifies one of the bonding policies. The default is
438 balance-rr (round robin). Possible values are:
442 Round-robin policy: Transmit packets in sequential
443 order from the first available slave through the
444 last. This mode provides load balancing and fault
449 Active-backup policy: Only one slave in the bond is
450 active. A different slave becomes active if, and only
451 if, the active slave fails. The bond's MAC address is
452 externally visible on only one port (network adapter)
453 to avoid confusing the switch.
455 In bonding version 2.6.2 or later, when a failover
456 occurs in active-backup mode, bonding will issue one
457 or more gratuitous ARPs on the newly active slave.
458 One gratuitous ARP is issued for the bonding master
459 interface and each VLAN interfaces configured above
460 it, provided that the interface has at least one IP
461 address configured. Gratuitous ARPs issued for VLAN
462 interfaces are tagged with the appropriate VLAN id.
464 This mode provides fault tolerance. The primary
465 option, documented below, affects the behavior of this
470 XOR policy: Transmit based on the selected transmit
471 hash policy. The default policy is a simple [(source
472 MAC address XOR'd with destination MAC address) modulo
473 slave count]. Alternate transmit policies may be
474 selected via the xmit_hash_policy option, described
477 This mode provides load balancing and fault tolerance.
481 Broadcast policy: transmits everything on all slave
482 interfaces. This mode provides fault tolerance.
486 IEEE 802.3ad Dynamic link aggregation. Creates
487 aggregation groups that share the same speed and
488 duplex settings. Utilizes all slaves in the active
489 aggregator according to the 802.3ad specification.
491 Slave selection for outgoing traffic is done according
492 to the transmit hash policy, which may be changed from
493 the default simple XOR policy via the xmit_hash_policy
494 option, documented below. Note that not all transmit
495 policies may be 802.3ad compliant, particularly in
496 regards to the packet mis-ordering requirements of
497 section 43.2.4 of the 802.3ad standard. Differing
498 peer implementations will have varying tolerances for
503 1. Ethtool support in the base drivers for retrieving
504 the speed and duplex of each slave.
506 2. A switch that supports IEEE 802.3ad Dynamic link
509 Most switches will require some type of configuration
510 to enable 802.3ad mode.
514 Adaptive transmit load balancing: channel bonding that
515 does not require any special switch support. The
516 outgoing traffic is distributed according to the
517 current load (computed relative to the speed) on each
518 slave. Incoming traffic is received by the current
519 slave. If the receiving slave fails, another slave
520 takes over the MAC address of the failed receiving
525 Ethtool support in the base drivers for retrieving the
530 Adaptive load balancing: includes balance-tlb plus
531 receive load balancing (rlb) for IPV4 traffic, and
532 does not require any special switch support. The
533 receive load balancing is achieved by ARP negotiation.
534 The bonding driver intercepts the ARP Replies sent by
535 the local system on their way out and overwrites the
536 source hardware address with the unique hardware
537 address of one of the slaves in the bond such that
538 different peers use different hardware addresses for
541 Receive traffic from connections created by the server
542 is also balanced. When the local system sends an ARP
543 Request the bonding driver copies and saves the peer's
544 IP information from the ARP packet. When the ARP
545 Reply arrives from the peer, its hardware address is
546 retrieved and the bonding driver initiates an ARP
547 reply to this peer assigning it to one of the slaves
548 in the bond. A problematic outcome of using ARP
549 negotiation for balancing is that each time that an
550 ARP request is broadcast it uses the hardware address
551 of the bond. Hence, peers learn the hardware address
552 of the bond and the balancing of receive traffic
553 collapses to the current slave. This is handled by
554 sending updates (ARP Replies) to all the peers with
555 their individually assigned hardware address such that
556 the traffic is redistributed. Receive traffic is also
557 redistributed when a new slave is added to the bond
558 and when an inactive slave is re-activated. The
559 receive load is distributed sequentially (round robin)
560 among the group of highest speed slaves in the bond.
562 When a link is reconnected or a new slave joins the
563 bond the receive traffic is redistributed among all
564 active slaves in the bond by initiating ARP Replies
565 with the selected MAC address to each of the
566 clients. The updelay parameter (detailed below) must
567 be set to a value equal or greater than the switch's
568 forwarding delay so that the ARP Replies sent to the
569 peers will not be blocked by the switch.
573 1. Ethtool support in the base drivers for retrieving
574 the speed of each slave.
576 2. Base driver support for setting the hardware
577 address of a device while it is open. This is
578 required so that there will always be one slave in the
579 team using the bond hardware address (the
580 curr_active_slave) while having a unique hardware
581 address for each slave in the bond. If the
582 curr_active_slave fails its hardware address is
583 swapped with the new curr_active_slave that was
588 Specifies the number of gratuitous ARPs to be issued after a
589 failover event. One gratuitous ARP is issued immediately after
590 the failover, subsequent ARPs are sent at a rate of one per link
591 monitor interval (arp_interval or miimon, whichever is active).
593 The valid range is 0 - 255; the default value is 1. This option
594 affects only the active-backup mode. This option was added for
595 bonding version 3.3.0.
599 Specifies the number of unsolicited IPv6 Neighbor Advertisements
600 to be issued after a failover event. One unsolicited NA is issued
601 immediately after the failover.
603 The valid range is 0 - 255; the default value is 1. This option
604 affects only the active-backup mode. This option was added for
605 bonding version 3.4.0.
609 A string (eth0, eth2, etc) specifying which slave is the
610 primary device. The specified device will always be the
611 active slave while it is available. Only when the primary is
612 off-line will alternate devices be used. This is useful when
613 one slave is preferred over another, e.g., when one slave has
614 higher throughput than another.
616 The primary option is only valid for active-backup mode.
620 Specifies the reselection policy for the primary slave. This
621 affects how the primary slave is chosen to become the active slave
622 when failure of the active slave or recovery of the primary slave
623 occurs. This option is designed to prevent flip-flopping between
624 the primary slave and other slaves. Possible values are:
626 always or 0 (default)
628 The primary slave becomes the active slave whenever it
633 The primary slave becomes the active slave when it comes
634 back up, if the speed and duplex of the primary slave is
635 better than the speed and duplex of the current active
640 The primary slave becomes the active slave only if the
641 current active slave fails and the primary slave is up.
643 The primary_reselect setting is ignored in two cases:
645 If no slaves are active, the first slave to recover is
646 made the active slave.
648 When initially enslaved, the primary slave is always made
651 Changing the primary_reselect policy via sysfs will cause an
652 immediate selection of the best active slave according to the new
653 policy. This may or may not result in a change of the active
654 slave, depending upon the circumstances.
656 This option was added for bonding version 3.6.0.
660 Specifies the time, in milliseconds, to wait before enabling a
661 slave after a link recovery has been detected. This option is
662 only valid for the miimon link monitor. The updelay value
663 should be a multiple of the miimon value; if not, it will be
664 rounded down to the nearest multiple. The default value is 0.
668 Specifies whether or not miimon should use MII or ETHTOOL
669 ioctls vs. netif_carrier_ok() to determine the link
670 status. The MII or ETHTOOL ioctls are less efficient and
671 utilize a deprecated calling sequence within the kernel. The
672 netif_carrier_ok() relies on the device driver to maintain its
673 state with netif_carrier_on/off; at this writing, most, but
674 not all, device drivers support this facility.
676 If bonding insists that the link is up when it should not be,
677 it may be that your network device driver does not support
678 netif_carrier_on/off. The default state for netif_carrier is
679 "carrier on," so if a driver does not support netif_carrier,
680 it will appear as if the link is always up. In this case,
681 setting use_carrier to 0 will cause bonding to revert to the
682 MII / ETHTOOL ioctl method to determine the link state.
684 A value of 1 enables the use of netif_carrier_ok(), a value of
685 0 will use the deprecated MII / ETHTOOL ioctls. The default
690 Selects the transmit hash policy to use for slave selection in
691 balance-xor and 802.3ad modes. Possible values are:
695 Uses XOR of hardware MAC addresses to generate the
698 (source MAC XOR destination MAC) modulo slave count
700 This algorithm will place all traffic to a particular
701 network peer on the same slave.
703 This algorithm is 802.3ad compliant.
707 This policy uses a combination of layer2 and layer3
708 protocol information to generate the hash.
710 Uses XOR of hardware MAC addresses and IP addresses to
711 generate the hash. The formula is
713 (((source IP XOR dest IP) AND 0xffff) XOR
714 ( source MAC XOR destination MAC ))
717 This algorithm will place all traffic to a particular
718 network peer on the same slave. For non-IP traffic,
719 the formula is the same as for the layer2 transmit
722 This policy is intended to provide a more balanced
723 distribution of traffic than layer2 alone, especially
724 in environments where a layer3 gateway device is
725 required to reach most destinations.
727 This algorithm is 802.3ad compliant.
731 This policy uses upper layer protocol information,
732 when available, to generate the hash. This allows for
733 traffic to a particular network peer to span multiple
734 slaves, although a single connection will not span
737 The formula for unfragmented TCP and UDP packets is
739 ((source port XOR dest port) XOR
740 ((source IP XOR dest IP) AND 0xffff)
743 For fragmented TCP or UDP packets and all other IP
744 protocol traffic, the source and destination port
745 information is omitted. For non-IP traffic, the
746 formula is the same as for the layer2 transmit hash
749 This policy is intended to mimic the behavior of
750 certain switches, notably Cisco switches with PFC2 as
751 well as some Foundry and IBM products.
753 This algorithm is not fully 802.3ad compliant. A
754 single TCP or UDP conversation containing both
755 fragmented and unfragmented packets will see packets
756 striped across two interfaces. This may result in out
757 of order delivery. Most traffic types will not meet
758 this criteria, as TCP rarely fragments traffic, and
759 most UDP traffic is not involved in extended
760 conversations. Other implementations of 802.3ad may
761 or may not tolerate this noncompliance.
763 The default value is layer2. This option was added in bonding
764 version 2.6.3. In earlier versions of bonding, this parameter
765 does not exist, and the layer2 policy is the only policy. The
766 layer2+3 value was added for bonding version 3.2.2.
770 Specifies the number of IGMP membership reports to be issued after
771 a failover event. One membership report is issued immediately after
772 the failover, subsequent packets are sent in each 200ms interval.
774 The valid range is 0 - 255; the default value is 1. This option
775 was added for bonding version 3.7.0.
777 3. Configuring Bonding Devices
778 ==============================
780 You can configure bonding using either your distro's network
781 initialization scripts, or manually using either ifenslave or the
782 sysfs interface. Distros generally use one of two packages for the
783 network initialization scripts: initscripts or sysconfig. Recent
784 versions of these packages have support for bonding, while older
787 We will first describe the options for configuring bonding for
788 distros using versions of initscripts and sysconfig with full or
789 partial support for bonding, then provide information on enabling
790 bonding without support from the network initialization scripts (i.e.,
791 older versions of initscripts or sysconfig).
793 If you're unsure whether your distro uses sysconfig or
794 initscripts, or don't know if it's new enough, have no fear.
795 Determining this is fairly straightforward.
797 First, issue the command:
801 It will respond with a line of text starting with either
802 "initscripts" or "sysconfig," followed by some numbers. This is the
803 package that provides your network initialization scripts.
805 Next, to determine if your installation supports bonding,
808 $ grep ifenslave /sbin/ifup
810 If this returns any matches, then your initscripts or
811 sysconfig has support for bonding.
813 3.1 Configuration with Sysconfig Support
814 ----------------------------------------
816 This section applies to distros using a version of sysconfig
817 with bonding support, for example, SuSE Linux Enterprise Server 9.
819 SuSE SLES 9's networking configuration system does support
820 bonding, however, at this writing, the YaST system configuration
821 front end does not provide any means to work with bonding devices.
822 Bonding devices can be managed by hand, however, as follows.
824 First, if they have not already been configured, configure the
825 slave devices. On SLES 9, this is most easily done by running the
826 yast2 sysconfig configuration utility. The goal is for to create an
827 ifcfg-id file for each slave device. The simplest way to accomplish
828 this is to configure the devices for DHCP (this is only to get the
829 file ifcfg-id file created; see below for some issues with DHCP). The
830 name of the configuration file for each device will be of the form:
832 ifcfg-id-xx:xx:xx:xx:xx:xx
834 Where the "xx" portion will be replaced with the digits from
835 the device's permanent MAC address.
837 Once the set of ifcfg-id-xx:xx:xx:xx:xx:xx files has been
838 created, it is necessary to edit the configuration files for the slave
839 devices (the MAC addresses correspond to those of the slave devices).
840 Before editing, the file will contain multiple lines, and will look
846 UNIQUE='XNzu.WeZGOGF+4wE'
847 _nm_name='bus-pci-0001:61:01.0'
849 Change the BOOTPROTO and STARTMODE lines to the following:
854 Do not alter the UNIQUE or _nm_name lines. Remove any other
855 lines (USERCTL, etc).
857 Once the ifcfg-id-xx:xx:xx:xx:xx:xx files have been modified,
858 it's time to create the configuration file for the bonding device
859 itself. This file is named ifcfg-bondX, where X is the number of the
860 bonding device to create, starting at 0. The first such file is
861 ifcfg-bond0, the second is ifcfg-bond1, and so on. The sysconfig
862 network configuration system will correctly start multiple instances
865 The contents of the ifcfg-bondX file is as follows:
868 BROADCAST="10.0.2.255"
870 NETMASK="255.255.0.0"
875 BONDING_MODULE_OPTS="mode=active-backup miimon=100"
876 BONDING_SLAVE0="eth0"
877 BONDING_SLAVE1="bus-pci-0000:06:08.1"
879 Replace the sample BROADCAST, IPADDR, NETMASK and NETWORK
880 values with the appropriate values for your network.
882 The STARTMODE specifies when the device is brought online.
883 The possible values are:
885 onboot: The device is started at boot time. If you're not
886 sure, this is probably what you want.
888 manual: The device is started only when ifup is called
889 manually. Bonding devices may be configured this
890 way if you do not wish them to start automatically
891 at boot for some reason.
893 hotplug: The device is started by a hotplug event. This is not
894 a valid choice for a bonding device.
896 off or ignore: The device configuration is ignored.
898 The line BONDING_MASTER='yes' indicates that the device is a
899 bonding master device. The only useful value is "yes."
901 The contents of BONDING_MODULE_OPTS are supplied to the
902 instance of the bonding module for this device. Specify the options
903 for the bonding mode, link monitoring, and so on here. Do not include
904 the max_bonds bonding parameter; this will confuse the configuration
905 system if you have multiple bonding devices.
907 Finally, supply one BONDING_SLAVEn="slave device" for each
908 slave. where "n" is an increasing value, one for each slave. The
909 "slave device" is either an interface name, e.g., "eth0", or a device
910 specifier for the network device. The interface name is easier to
911 find, but the ethN names are subject to change at boot time if, e.g.,
912 a device early in the sequence has failed. The device specifiers
913 (bus-pci-0000:06:08.1 in the example above) specify the physical
914 network device, and will not change unless the device's bus location
915 changes (for example, it is moved from one PCI slot to another). The
916 example above uses one of each type for demonstration purposes; most
917 configurations will choose one or the other for all slave devices.
919 When all configuration files have been modified or created,
920 networking must be restarted for the configuration changes to take
921 effect. This can be accomplished via the following:
923 # /etc/init.d/network restart
925 Note that the network control script (/sbin/ifdown) will
926 remove the bonding module as part of the network shutdown processing,
927 so it is not necessary to remove the module by hand if, e.g., the
928 module parameters have changed.
930 Also, at this writing, YaST/YaST2 will not manage bonding
931 devices (they do not show bonding interfaces on its list of network
932 devices). It is necessary to edit the configuration file by hand to
933 change the bonding configuration.
935 Additional general options and details of the ifcfg file
936 format can be found in an example ifcfg template file:
938 /etc/sysconfig/network/ifcfg.template
940 Note that the template does not document the various BONDING_
941 settings described above, but does describe many of the other options.
943 3.1.1 Using DHCP with Sysconfig
944 -------------------------------
946 Under sysconfig, configuring a device with BOOTPROTO='dhcp'
947 will cause it to query DHCP for its IP address information. At this
948 writing, this does not function for bonding devices; the scripts
949 attempt to obtain the device address from DHCP prior to adding any of
950 the slave devices. Without active slaves, the DHCP requests are not
953 3.1.2 Configuring Multiple Bonds with Sysconfig
954 -----------------------------------------------
956 The sysconfig network initialization system is capable of
957 handling multiple bonding devices. All that is necessary is for each
958 bonding instance to have an appropriately configured ifcfg-bondX file
959 (as described above). Do not specify the "max_bonds" parameter to any
960 instance of bonding, as this will confuse sysconfig. If you require
961 multiple bonding devices with identical parameters, create multiple
964 Because the sysconfig scripts supply the bonding module
965 options in the ifcfg-bondX file, it is not necessary to add them to
966 the system /etc/modules.conf or /etc/modprobe.conf configuration file.
968 3.2 Configuration with Initscripts Support
969 ------------------------------------------
971 This section applies to distros using a recent version of
972 initscripts with bonding support, for example, Red Hat Enterprise Linux
973 version 3 or later, Fedora, etc. On these systems, the network
974 initialization scripts have knowledge of bonding, and can be configured to
975 control bonding devices. Note that older versions of the initscripts
976 package have lower levels of support for bonding; this will be noted where
979 These distros will not automatically load the network adapter
980 driver unless the ethX device is configured with an IP address.
981 Because of this constraint, users must manually configure a
982 network-script file for all physical adapters that will be members of
983 a bondX link. Network script files are located in the directory:
985 /etc/sysconfig/network-scripts
987 The file name must be prefixed with "ifcfg-eth" and suffixed
988 with the adapter's physical adapter number. For example, the script
989 for eth0 would be named /etc/sysconfig/network-scripts/ifcfg-eth0.
990 Place the following text in the file:
999 The DEVICE= line will be different for every ethX device and
1000 must correspond with the name of the file, i.e., ifcfg-eth1 must have
1001 a device line of DEVICE=eth1. The setting of the MASTER= line will
1002 also depend on the final bonding interface name chosen for your bond.
1003 As with other network devices, these typically start at 0, and go up
1004 one for each device, i.e., the first bonding instance is bond0, the
1005 second is bond1, and so on.
1007 Next, create a bond network script. The file name for this
1008 script will be /etc/sysconfig/network-scripts/ifcfg-bondX where X is
1009 the number of the bond. For bond0 the file is named "ifcfg-bond0",
1010 for bond1 it is named "ifcfg-bond1", and so on. Within that file,
1011 place the following text:
1015 NETMASK=255.255.255.0
1017 BROADCAST=192.168.1.255
1022 Be sure to change the networking specific lines (IPADDR,
1023 NETMASK, NETWORK and BROADCAST) to match your network configuration.
1025 For later versions of initscripts, such as that found with Fedora
1026 7 (or later) and Red Hat Enterprise Linux version 5 (or later), it is possible,
1027 and, indeed, preferable, to specify the bonding options in the ifcfg-bond0
1028 file, e.g. a line of the format:
1030 BONDING_OPTS="mode=active-backup arp_interval=60 arp_ip_target=192.168.1.254"
1032 will configure the bond with the specified options. The options
1033 specified in BONDING_OPTS are identical to the bonding module parameters
1034 except for the arp_ip_target field when using versions of initscripts older
1035 than and 8.57 (Fedora 8) and 8.45.19 (Red Hat Enterprise Linux 5.2). When
1036 using older versions each target should be included as a separate option and
1037 should be preceded by a '+' to indicate it should be added to the list of
1038 queried targets, e.g.,
1040 arp_ip_target=+192.168.1.1 arp_ip_target=+192.168.1.2
1042 is the proper syntax to specify multiple targets. When specifying
1043 options via BONDING_OPTS, it is not necessary to edit /etc/modules.conf or
1046 For even older versions of initscripts that do not support
1047 BONDING_OPTS, it is necessary to edit /etc/modules.conf (or
1048 /etc/modprobe.conf, depending upon your distro) to load the bonding module
1049 with your desired options when the bond0 interface is brought up. The
1050 following lines in /etc/modules.conf (or modprobe.conf) will load the
1051 bonding module, and select its options:
1054 options bond0 mode=balance-alb miimon=100
1056 Replace the sample parameters with the appropriate set of
1057 options for your configuration.
1059 Finally run "/etc/rc.d/init.d/network restart" as root. This
1060 will restart the networking subsystem and your bond link should be now
1063 3.2.1 Using DHCP with Initscripts
1064 ---------------------------------
1066 Recent versions of initscripts (the versions supplied with Fedora
1067 Core 3 and Red Hat Enterprise Linux 4, or later versions, are reported to
1068 work) have support for assigning IP information to bonding devices via
1071 To configure bonding for DHCP, configure it as described
1072 above, except replace the line "BOOTPROTO=none" with "BOOTPROTO=dhcp"
1073 and add a line consisting of "TYPE=Bonding". Note that the TYPE value
1076 3.2.2 Configuring Multiple Bonds with Initscripts
1077 -------------------------------------------------
1079 Initscripts packages that are included with Fedora 7 and Red Hat
1080 Enterprise Linux 5 support multiple bonding interfaces by simply
1081 specifying the appropriate BONDING_OPTS= in ifcfg-bondX where X is the
1082 number of the bond. This support requires sysfs support in the kernel,
1083 and a bonding driver of version 3.0.0 or later. Other configurations may
1084 not support this method for specifying multiple bonding interfaces; for
1085 those instances, see the "Configuring Multiple Bonds Manually" section,
1088 3.3 Configuring Bonding Manually with Ifenslave
1089 -----------------------------------------------
1091 This section applies to distros whose network initialization
1092 scripts (the sysconfig or initscripts package) do not have specific
1093 knowledge of bonding. One such distro is SuSE Linux Enterprise Server
1096 The general method for these systems is to place the bonding
1097 module parameters into /etc/modules.conf or /etc/modprobe.conf (as
1098 appropriate for the installed distro), then add modprobe and/or
1099 ifenslave commands to the system's global init script. The name of
1100 the global init script differs; for sysconfig, it is
1101 /etc/init.d/boot.local and for initscripts it is /etc/rc.d/rc.local.
1103 For example, if you wanted to make a simple bond of two e100
1104 devices (presumed to be eth0 and eth1), and have it persist across
1105 reboots, edit the appropriate file (/etc/init.d/boot.local or
1106 /etc/rc.d/rc.local), and add the following:
1108 modprobe bonding mode=balance-alb miimon=100
1110 ifconfig bond0 192.168.1.1 netmask 255.255.255.0 up
1111 ifenslave bond0 eth0
1112 ifenslave bond0 eth1
1114 Replace the example bonding module parameters and bond0
1115 network configuration (IP address, netmask, etc) with the appropriate
1116 values for your configuration.
1118 Unfortunately, this method will not provide support for the
1119 ifup and ifdown scripts on the bond devices. To reload the bonding
1120 configuration, it is necessary to run the initialization script, e.g.,
1122 # /etc/init.d/boot.local
1126 # /etc/rc.d/rc.local
1128 It may be desirable in such a case to create a separate script
1129 which only initializes the bonding configuration, then call that
1130 separate script from within boot.local. This allows for bonding to be
1131 enabled without re-running the entire global init script.
1133 To shut down the bonding devices, it is necessary to first
1134 mark the bonding device itself as being down, then remove the
1135 appropriate device driver modules. For our example above, you can do
1138 # ifconfig bond0 down
1142 Again, for convenience, it may be desirable to create a script
1143 with these commands.
1146 3.3.1 Configuring Multiple Bonds Manually
1147 -----------------------------------------
1149 This section contains information on configuring multiple
1150 bonding devices with differing options for those systems whose network
1151 initialization scripts lack support for configuring multiple bonds.
1153 If you require multiple bonding devices, but all with the same
1154 options, you may wish to use the "max_bonds" module parameter,
1157 To create multiple bonding devices with differing options, it is
1158 preferrable to use bonding parameters exported by sysfs, documented in the
1161 For versions of bonding without sysfs support, the only means to
1162 provide multiple instances of bonding with differing options is to load
1163 the bonding driver multiple times. Note that current versions of the
1164 sysconfig network initialization scripts handle this automatically; if
1165 your distro uses these scripts, no special action is needed. See the
1166 section Configuring Bonding Devices, above, if you're not sure about your
1167 network initialization scripts.
1169 To load multiple instances of the module, it is necessary to
1170 specify a different name for each instance (the module loading system
1171 requires that every loaded module, even multiple instances of the same
1172 module, have a unique name). This is accomplished by supplying multiple
1173 sets of bonding options in /etc/modprobe.conf, for example:
1176 options bond0 -o bond0 mode=balance-rr miimon=100
1179 options bond1 -o bond1 mode=balance-alb miimon=50
1181 will load the bonding module two times. The first instance is
1182 named "bond0" and creates the bond0 device in balance-rr mode with an
1183 miimon of 100. The second instance is named "bond1" and creates the
1184 bond1 device in balance-alb mode with an miimon of 50.
1186 In some circumstances (typically with older distributions),
1187 the above does not work, and the second bonding instance never sees
1188 its options. In that case, the second options line can be substituted
1191 install bond1 /sbin/modprobe --ignore-install bonding -o bond1 \
1192 mode=balance-alb miimon=50
1194 This may be repeated any number of times, specifying a new and
1195 unique name in place of bond1 for each subsequent instance.
1197 It has been observed that some Red Hat supplied kernels are unable
1198 to rename modules at load time (the "-o bond1" part). Attempts to pass
1199 that option to modprobe will produce an "Operation not permitted" error.
1200 This has been reported on some Fedora Core kernels, and has been seen on
1201 RHEL 4 as well. On kernels exhibiting this problem, it will be impossible
1202 to configure multiple bonds with differing parameters (as they are older
1203 kernels, and also lack sysfs support).
1205 3.4 Configuring Bonding Manually via Sysfs
1206 ------------------------------------------
1208 Starting with version 3.0.0, Channel Bonding may be configured
1209 via the sysfs interface. This interface allows dynamic configuration
1210 of all bonds in the system without unloading the module. It also
1211 allows for adding and removing bonds at runtime. Ifenslave is no
1212 longer required, though it is still supported.
1214 Use of the sysfs interface allows you to use multiple bonds
1215 with different configurations without having to reload the module.
1216 It also allows you to use multiple, differently configured bonds when
1217 bonding is compiled into the kernel.
1219 You must have the sysfs filesystem mounted to configure
1220 bonding this way. The examples in this document assume that you
1221 are using the standard mount point for sysfs, e.g. /sys. If your
1222 sysfs filesystem is mounted elsewhere, you will need to adjust the
1223 example paths accordingly.
1225 Creating and Destroying Bonds
1226 -----------------------------
1227 To add a new bond foo:
1228 # echo +foo > /sys/class/net/bonding_masters
1230 To remove an existing bond bar:
1231 # echo -bar > /sys/class/net/bonding_masters
1233 To show all existing bonds:
1234 # cat /sys/class/net/bonding_masters
1236 NOTE: due to 4K size limitation of sysfs files, this list may be
1237 truncated if you have more than a few hundred bonds. This is unlikely
1238 to occur under normal operating conditions.
1240 Adding and Removing Slaves
1241 --------------------------
1242 Interfaces may be enslaved to a bond using the file
1243 /sys/class/net/<bond>/bonding/slaves. The semantics for this file
1244 are the same as for the bonding_masters file.
1246 To enslave interface eth0 to bond bond0:
1248 # echo +eth0 > /sys/class/net/bond0/bonding/slaves
1250 To free slave eth0 from bond bond0:
1251 # echo -eth0 > /sys/class/net/bond0/bonding/slaves
1253 When an interface is enslaved to a bond, symlinks between the
1254 two are created in the sysfs filesystem. In this case, you would get
1255 /sys/class/net/bond0/slave_eth0 pointing to /sys/class/net/eth0, and
1256 /sys/class/net/eth0/master pointing to /sys/class/net/bond0.
1258 This means that you can tell quickly whether or not an
1259 interface is enslaved by looking for the master symlink. Thus:
1260 # echo -eth0 > /sys/class/net/eth0/master/bonding/slaves
1261 will free eth0 from whatever bond it is enslaved to, regardless of
1262 the name of the bond interface.
1264 Changing a Bond's Configuration
1265 -------------------------------
1266 Each bond may be configured individually by manipulating the
1267 files located in /sys/class/net/<bond name>/bonding
1269 The names of these files correspond directly with the command-
1270 line parameters described elsewhere in this file, and, with the
1271 exception of arp_ip_target, they accept the same values. To see the
1272 current setting, simply cat the appropriate file.
1274 A few examples will be given here; for specific usage
1275 guidelines for each parameter, see the appropriate section in this
1278 To configure bond0 for balance-alb mode:
1279 # ifconfig bond0 down
1280 # echo 6 > /sys/class/net/bond0/bonding/mode
1282 # echo balance-alb > /sys/class/net/bond0/bonding/mode
1283 NOTE: The bond interface must be down before the mode can be
1286 To enable MII monitoring on bond0 with a 1 second interval:
1287 # echo 1000 > /sys/class/net/bond0/bonding/miimon
1288 NOTE: If ARP monitoring is enabled, it will disabled when MII
1289 monitoring is enabled, and vice-versa.
1292 # echo +192.168.0.100 > /sys/class/net/bond0/bonding/arp_ip_target
1293 # echo +192.168.0.101 > /sys/class/net/bond0/bonding/arp_ip_target
1294 NOTE: up to 16 target addresses may be specified.
1296 To remove an ARP target:
1297 # echo -192.168.0.100 > /sys/class/net/bond0/bonding/arp_ip_target
1299 Example Configuration
1300 ---------------------
1301 We begin with the same example that is shown in section 3.3,
1302 executed with sysfs, and without using ifenslave.
1304 To make a simple bond of two e100 devices (presumed to be eth0
1305 and eth1), and have it persist across reboots, edit the appropriate
1306 file (/etc/init.d/boot.local or /etc/rc.d/rc.local), and add the
1311 echo balance-alb > /sys/class/net/bond0/bonding/mode
1312 ifconfig bond0 192.168.1.1 netmask 255.255.255.0 up
1313 echo 100 > /sys/class/net/bond0/bonding/miimon
1314 echo +eth0 > /sys/class/net/bond0/bonding/slaves
1315 echo +eth1 > /sys/class/net/bond0/bonding/slaves
1317 To add a second bond, with two e1000 interfaces in
1318 active-backup mode, using ARP monitoring, add the following lines to
1322 echo +bond1 > /sys/class/net/bonding_masters
1323 echo active-backup > /sys/class/net/bond1/bonding/mode
1324 ifconfig bond1 192.168.2.1 netmask 255.255.255.0 up
1325 echo +192.168.2.100 /sys/class/net/bond1/bonding/arp_ip_target
1326 echo 2000 > /sys/class/net/bond1/bonding/arp_interval
1327 echo +eth2 > /sys/class/net/bond1/bonding/slaves
1328 echo +eth3 > /sys/class/net/bond1/bonding/slaves
1330 3.5 Overriding Configuration for Special Cases
1331 ----------------------------------------------
1332 When using the bonding driver, the physical port which transmits a frame is
1333 typically selected by the bonding driver, and is not relevant to the user or
1334 system administrator. The output port is simply selected using the policies of
1335 the selected bonding mode. On occasion however, it is helpful to direct certain
1336 classes of traffic to certain physical interfaces on output to implement
1337 slightly more complex policies. For example, to reach a web server over a
1338 bonded interface in which eth0 connects to a private network, while eth1
1339 connects via a public network, it may be desirous to bias the bond to send said
1340 traffic over eth0 first, using eth1 only as a fall back, while all other traffic
1341 can safely be sent over either interface. Such configurations may be achieved
1342 using the traffic control utilities inherent in linux.
1344 By default the bonding driver is multiqueue aware and 16 queues are created
1345 when the driver initializes (see Documentation/networking/multiqueue.txt
1346 for details). If more or less queues are desired the module parameter
1347 tx_queues can be used to change this value. There is no sysfs parameter
1348 available as the allocation is done at module init time.
1350 The output of the file /proc/net/bonding/bondX has changed so the output Queue
1351 ID is now printed for each slave:
1353 Bonding Mode: fault-tolerance (active-backup)
1355 Currently Active Slave: eth0
1357 MII Polling Interval (ms): 0
1361 Slave Interface: eth0
1363 Link Failure Count: 0
1364 Permanent HW addr: 00:1a:a0:12:8f:cb
1367 Slave Interface: eth1
1369 Link Failure Count: 0
1370 Permanent HW addr: 00:1a:a0:12:8f:cc
1373 The queue_id for a slave can be set using the command:
1375 # echo "eth1:2" > /sys/class/net/bond0/bonding/queue_id
1377 Any interface that needs a queue_id set should set it with multiple calls
1378 like the one above until proper priorities are set for all interfaces. On
1379 distributions that allow configuration via initscripts, multiple 'queue_id'
1380 arguments can be added to BONDING_OPTS to set all needed slave queues.
1382 These queue id's can be used in conjunction with the tc utility to configure
1383 a multiqueue qdisc and filters to bias certain traffic to transmit on certain
1384 slave devices. For instance, say we wanted, in the above configuration to
1385 force all traffic bound to 192.168.1.100 to use eth1 in the bond as its output
1386 device. The following commands would accomplish this:
1388 # tc qdisc add dev bond0 handle 1 root multiq
1390 # tc filter add dev bond0 protocol ip parent 1: prio 1 u32 match ip dst \
1391 192.168.1.100 action skbedit queue_mapping 2
1393 These commands tell the kernel to attach a multiqueue queue discipline to the
1394 bond0 interface and filter traffic enqueued to it, such that packets with a dst
1395 ip of 192.168.1.100 have their output queue mapping value overwritten to 2.
1396 This value is then passed into the driver, causing the normal output path
1397 selection policy to be overridden, selecting instead qid 2, which maps to eth1.
1399 Note that qid values begin at 1. Qid 0 is reserved to initiate to the driver
1400 that normal output policy selection should take place. One benefit to simply
1401 leaving the qid for a slave to 0 is the multiqueue awareness in the bonding
1402 driver that is now present. This awareness allows tc filters to be placed on
1403 slave devices as well as bond devices and the bonding driver will simply act as
1404 a pass-through for selecting output queues on the slave device rather than
1405 output port selection.
1407 This feature first appeared in bonding driver version 3.7.0 and support for
1408 output slave selection was limited to round-robin and active-backup modes.
1410 4 Querying Bonding Configuration
1411 =================================
1413 4.1 Bonding Configuration
1414 -------------------------
1416 Each bonding device has a read-only file residing in the
1417 /proc/net/bonding directory. The file contents include information
1418 about the bonding configuration, options and state of each slave.
1420 For example, the contents of /proc/net/bonding/bond0 after the
1421 driver is loaded with parameters of mode=0 and miimon=1000 is
1422 generally as follows:
1424 Ethernet Channel Bonding Driver: 2.6.1 (October 29, 2004)
1425 Bonding Mode: load balancing (round-robin)
1426 Currently Active Slave: eth0
1428 MII Polling Interval (ms): 1000
1432 Slave Interface: eth1
1434 Link Failure Count: 1
1436 Slave Interface: eth0
1438 Link Failure Count: 1
1440 The precise format and contents will change depending upon the
1441 bonding configuration, state, and version of the bonding driver.
1443 4.2 Network configuration
1444 -------------------------
1446 The network configuration can be inspected using the ifconfig
1447 command. Bonding devices will have the MASTER flag set; Bonding slave
1448 devices will have the SLAVE flag set. The ifconfig output does not
1449 contain information on which slaves are associated with which masters.
1451 In the example below, the bond0 interface is the master
1452 (MASTER) while eth0 and eth1 are slaves (SLAVE). Notice all slaves of
1453 bond0 have the same MAC address (HWaddr) as bond0 for all modes except
1454 TLB and ALB that require a unique MAC address for each slave.
1457 bond0 Link encap:Ethernet HWaddr 00:C0:F0:1F:37:B4
1458 inet addr:XXX.XXX.XXX.YYY Bcast:XXX.XXX.XXX.255 Mask:255.255.252.0
1459 UP BROADCAST RUNNING MASTER MULTICAST MTU:1500 Metric:1
1460 RX packets:7224794 errors:0 dropped:0 overruns:0 frame:0
1461 TX packets:3286647 errors:1 dropped:0 overruns:1 carrier:0
1462 collisions:0 txqueuelen:0
1464 eth0 Link encap:Ethernet HWaddr 00:C0:F0:1F:37:B4
1465 UP BROADCAST RUNNING SLAVE MULTICAST MTU:1500 Metric:1
1466 RX packets:3573025 errors:0 dropped:0 overruns:0 frame:0
1467 TX packets:1643167 errors:1 dropped:0 overruns:1 carrier:0
1468 collisions:0 txqueuelen:100
1469 Interrupt:10 Base address:0x1080
1471 eth1 Link encap:Ethernet HWaddr 00:C0:F0:1F:37:B4
1472 UP BROADCAST RUNNING SLAVE MULTICAST MTU:1500 Metric:1
1473 RX packets:3651769 errors:0 dropped:0 overruns:0 frame:0
1474 TX packets:1643480 errors:0 dropped:0 overruns:0 carrier:0
1475 collisions:0 txqueuelen:100
1476 Interrupt:9 Base address:0x1400
1478 5. Switch Configuration
1479 =======================
1481 For this section, "switch" refers to whatever system the
1482 bonded devices are directly connected to (i.e., where the other end of
1483 the cable plugs into). This may be an actual dedicated switch device,
1484 or it may be another regular system (e.g., another computer running
1487 The active-backup, balance-tlb and balance-alb modes do not
1488 require any specific configuration of the switch.
1490 The 802.3ad mode requires that the switch have the appropriate
1491 ports configured as an 802.3ad aggregation. The precise method used
1492 to configure this varies from switch to switch, but, for example, a
1493 Cisco 3550 series switch requires that the appropriate ports first be
1494 grouped together in a single etherchannel instance, then that
1495 etherchannel is set to mode "lacp" to enable 802.3ad (instead of
1496 standard EtherChannel).
1498 The balance-rr, balance-xor and broadcast modes generally
1499 require that the switch have the appropriate ports grouped together.
1500 The nomenclature for such a group differs between switches, it may be
1501 called an "etherchannel" (as in the Cisco example, above), a "trunk
1502 group" or some other similar variation. For these modes, each switch
1503 will also have its own configuration options for the switch's transmit
1504 policy to the bond. Typical choices include XOR of either the MAC or
1505 IP addresses. The transmit policy of the two peers does not need to
1506 match. For these three modes, the bonding mode really selects a
1507 transmit policy for an EtherChannel group; all three will interoperate
1508 with another EtherChannel group.
1511 6. 802.1q VLAN Support
1512 ======================
1514 It is possible to configure VLAN devices over a bond interface
1515 using the 8021q driver. However, only packets coming from the 8021q
1516 driver and passing through bonding will be tagged by default. Self
1517 generated packets, for example, bonding's learning packets or ARP
1518 packets generated by either ALB mode or the ARP monitor mechanism, are
1519 tagged internally by bonding itself. As a result, bonding must
1520 "learn" the VLAN IDs configured above it, and use those IDs to tag
1521 self generated packets.
1523 For reasons of simplicity, and to support the use of adapters
1524 that can do VLAN hardware acceleration offloading, the bonding
1525 interface declares itself as fully hardware offloading capable, it gets
1526 the add_vid/kill_vid notifications to gather the necessary
1527 information, and it propagates those actions to the slaves. In case
1528 of mixed adapter types, hardware accelerated tagged packets that
1529 should go through an adapter that is not offloading capable are
1530 "un-accelerated" by the bonding driver so the VLAN tag sits in the
1533 VLAN interfaces *must* be added on top of a bonding interface
1534 only after enslaving at least one slave. The bonding interface has a
1535 hardware address of 00:00:00:00:00:00 until the first slave is added.
1536 If the VLAN interface is created prior to the first enslavement, it
1537 would pick up the all-zeroes hardware address. Once the first slave
1538 is attached to the bond, the bond device itself will pick up the
1539 slave's hardware address, which is then available for the VLAN device.
1541 Also, be aware that a similar problem can occur if all slaves
1542 are released from a bond that still has one or more VLAN interfaces on
1543 top of it. When a new slave is added, the bonding interface will
1544 obtain its hardware address from the first slave, which might not
1545 match the hardware address of the VLAN interfaces (which was
1546 ultimately copied from an earlier slave).
1548 There are two methods to insure that the VLAN device operates
1549 with the correct hardware address if all slaves are removed from a
1552 1. Remove all VLAN interfaces then recreate them
1554 2. Set the bonding interface's hardware address so that it
1555 matches the hardware address of the VLAN interfaces.
1557 Note that changing a VLAN interface's HW address would set the
1558 underlying device -- i.e. the bonding interface -- to promiscuous
1559 mode, which might not be what you want.
1565 The bonding driver at present supports two schemes for
1566 monitoring a slave device's link state: the ARP monitor and the MII
1569 At the present time, due to implementation restrictions in the
1570 bonding driver itself, it is not possible to enable both ARP and MII
1571 monitoring simultaneously.
1573 7.1 ARP Monitor Operation
1574 -------------------------
1576 The ARP monitor operates as its name suggests: it sends ARP
1577 queries to one or more designated peer systems on the network, and
1578 uses the response as an indication that the link is operating. This
1579 gives some assurance that traffic is actually flowing to and from one
1580 or more peers on the local network.
1582 The ARP monitor relies on the device driver itself to verify
1583 that traffic is flowing. In particular, the driver must keep up to
1584 date the last receive time, dev->last_rx, and transmit start time,
1585 dev->trans_start. If these are not updated by the driver, then the
1586 ARP monitor will immediately fail any slaves using that driver, and
1587 those slaves will stay down. If networking monitoring (tcpdump, etc)
1588 shows the ARP requests and replies on the network, then it may be that
1589 your device driver is not updating last_rx and trans_start.
1591 7.2 Configuring Multiple ARP Targets
1592 ------------------------------------
1594 While ARP monitoring can be done with just one target, it can
1595 be useful in a High Availability setup to have several targets to
1596 monitor. In the case of just one target, the target itself may go
1597 down or have a problem making it unresponsive to ARP requests. Having
1598 an additional target (or several) increases the reliability of the ARP
1601 Multiple ARP targets must be separated by commas as follows:
1603 # example options for ARP monitoring with three targets
1605 options bond0 arp_interval=60 arp_ip_target=192.168.0.1,192.168.0.3,192.168.0.9
1607 For just a single target the options would resemble:
1609 # example options for ARP monitoring with one target
1611 options bond0 arp_interval=60 arp_ip_target=192.168.0.100
1614 7.3 MII Monitor Operation
1615 -------------------------
1617 The MII monitor monitors only the carrier state of the local
1618 network interface. It accomplishes this in one of three ways: by
1619 depending upon the device driver to maintain its carrier state, by
1620 querying the device's MII registers, or by making an ethtool query to
1623 If the use_carrier module parameter is 1 (the default value),
1624 then the MII monitor will rely on the driver for carrier state
1625 information (via the netif_carrier subsystem). As explained in the
1626 use_carrier parameter information, above, if the MII monitor fails to
1627 detect carrier loss on the device (e.g., when the cable is physically
1628 disconnected), it may be that the driver does not support
1631 If use_carrier is 0, then the MII monitor will first query the
1632 device's (via ioctl) MII registers and check the link state. If that
1633 request fails (not just that it returns carrier down), then the MII
1634 monitor will make an ethtool ETHOOL_GLINK request to attempt to obtain
1635 the same information. If both methods fail (i.e., the driver either
1636 does not support or had some error in processing both the MII register
1637 and ethtool requests), then the MII monitor will assume the link is
1640 8. Potential Sources of Trouble
1641 ===============================
1643 8.1 Adventures in Routing
1644 -------------------------
1646 When bonding is configured, it is important that the slave
1647 devices not have routes that supersede routes of the master (or,
1648 generally, not have routes at all). For example, suppose the bonding
1649 device bond0 has two slaves, eth0 and eth1, and the routing table is
1652 Kernel IP routing table
1653 Destination Gateway Genmask Flags MSS Window irtt Iface
1654 10.0.0.0 0.0.0.0 255.255.0.0 U 40 0 0 eth0
1655 10.0.0.0 0.0.0.0 255.255.0.0 U 40 0 0 eth1
1656 10.0.0.0 0.0.0.0 255.255.0.0 U 40 0 0 bond0
1657 127.0.0.0 0.0.0.0 255.0.0.0 U 40 0 0 lo
1659 This routing configuration will likely still update the
1660 receive/transmit times in the driver (needed by the ARP monitor), but
1661 may bypass the bonding driver (because outgoing traffic to, in this
1662 case, another host on network 10 would use eth0 or eth1 before bond0).
1664 The ARP monitor (and ARP itself) may become confused by this
1665 configuration, because ARP requests (generated by the ARP monitor)
1666 will be sent on one interface (bond0), but the corresponding reply
1667 will arrive on a different interface (eth0). This reply looks to ARP
1668 as an unsolicited ARP reply (because ARP matches replies on an
1669 interface basis), and is discarded. The MII monitor is not affected
1670 by the state of the routing table.
1672 The solution here is simply to insure that slaves do not have
1673 routes of their own, and if for some reason they must, those routes do
1674 not supersede routes of their master. This should generally be the
1675 case, but unusual configurations or errant manual or automatic static
1676 route additions may cause trouble.
1678 8.2 Ethernet Device Renaming
1679 ----------------------------
1681 On systems with network configuration scripts that do not
1682 associate physical devices directly with network interface names (so
1683 that the same physical device always has the same "ethX" name), it may
1684 be necessary to add some special logic to either /etc/modules.conf or
1685 /etc/modprobe.conf (depending upon which is installed on the system).
1687 For example, given a modules.conf containing the following:
1690 options bond0 mode=some-mode miimon=50
1696 If neither eth0 and eth1 are slaves to bond0, then when the
1697 bond0 interface comes up, the devices may end up reordered. This
1698 happens because bonding is loaded first, then its slave device's
1699 drivers are loaded next. Since no other drivers have been loaded,
1700 when the e1000 driver loads, it will receive eth0 and eth1 for its
1701 devices, but the bonding configuration tries to enslave eth2 and eth3
1702 (which may later be assigned to the tg3 devices).
1704 Adding the following:
1706 add above bonding e1000 tg3
1708 causes modprobe to load e1000 then tg3, in that order, when
1709 bonding is loaded. This command is fully documented in the
1710 modules.conf manual page.
1712 On systems utilizing modprobe.conf (or modprobe.conf.local),
1713 an equivalent problem can occur. In this case, the following can be
1714 added to modprobe.conf (or modprobe.conf.local, as appropriate), as
1715 follows (all on one line; it has been split here for clarity):
1717 install bonding /sbin/modprobe tg3; /sbin/modprobe e1000;
1718 /sbin/modprobe --ignore-install bonding
1720 This will, when loading the bonding module, rather than
1721 performing the normal action, instead execute the provided command.
1722 This command loads the device drivers in the order needed, then calls
1723 modprobe with --ignore-install to cause the normal action to then take
1724 place. Full documentation on this can be found in the modprobe.conf
1725 and modprobe manual pages.
1727 8.3. Painfully Slow Or No Failed Link Detection By Miimon
1728 ---------------------------------------------------------
1730 By default, bonding enables the use_carrier option, which
1731 instructs bonding to trust the driver to maintain carrier state.
1733 As discussed in the options section, above, some drivers do
1734 not support the netif_carrier_on/_off link state tracking system.
1735 With use_carrier enabled, bonding will always see these links as up,
1736 regardless of their actual state.
1738 Additionally, other drivers do support netif_carrier, but do
1739 not maintain it in real time, e.g., only polling the link state at
1740 some fixed interval. In this case, miimon will detect failures, but
1741 only after some long period of time has expired. If it appears that
1742 miimon is very slow in detecting link failures, try specifying
1743 use_carrier=0 to see if that improves the failure detection time. If
1744 it does, then it may be that the driver checks the carrier state at a
1745 fixed interval, but does not cache the MII register values (so the
1746 use_carrier=0 method of querying the registers directly works). If
1747 use_carrier=0 does not improve the failover, then the driver may cache
1748 the registers, or the problem may be elsewhere.
1750 Also, remember that miimon only checks for the device's
1751 carrier state. It has no way to determine the state of devices on or
1752 beyond other ports of a switch, or if a switch is refusing to pass
1753 traffic while still maintaining carrier on.
1758 If running SNMP agents, the bonding driver should be loaded
1759 before any network drivers participating in a bond. This requirement
1760 is due to the interface index (ipAdEntIfIndex) being associated to
1761 the first interface found with a given IP address. That is, there is
1762 only one ipAdEntIfIndex for each IP address. For example, if eth0 and
1763 eth1 are slaves of bond0 and the driver for eth0 is loaded before the
1764 bonding driver, the interface for the IP address will be associated
1765 with the eth0 interface. This configuration is shown below, the IP
1766 address 192.168.1.1 has an interface index of 2 which indexes to eth0
1767 in the ifDescr table (ifDescr.2).
1769 interfaces.ifTable.ifEntry.ifDescr.1 = lo
1770 interfaces.ifTable.ifEntry.ifDescr.2 = eth0
1771 interfaces.ifTable.ifEntry.ifDescr.3 = eth1
1772 interfaces.ifTable.ifEntry.ifDescr.4 = eth2
1773 interfaces.ifTable.ifEntry.ifDescr.5 = eth3
1774 interfaces.ifTable.ifEntry.ifDescr.6 = bond0
1775 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.10.10.10 = 5
1776 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.192.168.1.1 = 2
1777 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.74.20.94 = 4
1778 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.127.0.0.1 = 1
1780 This problem is avoided by loading the bonding driver before
1781 any network drivers participating in a bond. Below is an example of
1782 loading the bonding driver first, the IP address 192.168.1.1 is
1783 correctly associated with ifDescr.2.
1785 interfaces.ifTable.ifEntry.ifDescr.1 = lo
1786 interfaces.ifTable.ifEntry.ifDescr.2 = bond0
1787 interfaces.ifTable.ifEntry.ifDescr.3 = eth0
1788 interfaces.ifTable.ifEntry.ifDescr.4 = eth1
1789 interfaces.ifTable.ifEntry.ifDescr.5 = eth2
1790 interfaces.ifTable.ifEntry.ifDescr.6 = eth3
1791 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.10.10.10 = 6
1792 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.192.168.1.1 = 2
1793 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.74.20.94 = 5
1794 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.127.0.0.1 = 1
1796 While some distributions may not report the interface name in
1797 ifDescr, the association between the IP address and IfIndex remains
1798 and SNMP functions such as Interface_Scan_Next will report that
1801 10. Promiscuous mode
1802 ====================
1804 When running network monitoring tools, e.g., tcpdump, it is
1805 common to enable promiscuous mode on the device, so that all traffic
1806 is seen (instead of seeing only traffic destined for the local host).
1807 The bonding driver handles promiscuous mode changes to the bonding
1808 master device (e.g., bond0), and propagates the setting to the slave
1811 For the balance-rr, balance-xor, broadcast, and 802.3ad modes,
1812 the promiscuous mode setting is propagated to all slaves.
1814 For the active-backup, balance-tlb and balance-alb modes, the
1815 promiscuous mode setting is propagated only to the active slave.
1817 For balance-tlb mode, the active slave is the slave currently
1818 receiving inbound traffic.
1820 For balance-alb mode, the active slave is the slave used as a
1821 "primary." This slave is used for mode-specific control traffic, for
1822 sending to peers that are unassigned or if the load is unbalanced.
1824 For the active-backup, balance-tlb and balance-alb modes, when
1825 the active slave changes (e.g., due to a link failure), the
1826 promiscuous setting will be propagated to the new active slave.
1828 11. Configuring Bonding for High Availability
1829 =============================================
1831 High Availability refers to configurations that provide
1832 maximum network availability by having redundant or backup devices,
1833 links or switches between the host and the rest of the world. The
1834 goal is to provide the maximum availability of network connectivity
1835 (i.e., the network always works), even though other configurations
1836 could provide higher throughput.
1838 11.1 High Availability in a Single Switch Topology
1839 --------------------------------------------------
1841 If two hosts (or a host and a single switch) are directly
1842 connected via multiple physical links, then there is no availability
1843 penalty to optimizing for maximum bandwidth. In this case, there is
1844 only one switch (or peer), so if it fails, there is no alternative
1845 access to fail over to. Additionally, the bonding load balance modes
1846 support link monitoring of their members, so if individual links fail,
1847 the load will be rebalanced across the remaining devices.
1849 See Section 13, "Configuring Bonding for Maximum Throughput"
1850 for information on configuring bonding with one peer device.
1852 11.2 High Availability in a Multiple Switch Topology
1853 ----------------------------------------------------
1855 With multiple switches, the configuration of bonding and the
1856 network changes dramatically. In multiple switch topologies, there is
1857 a trade off between network availability and usable bandwidth.
1859 Below is a sample network, configured to maximize the
1860 availability of the network:
1864 +-----+----+ +-----+----+
1865 | |port2 ISL port2| |
1866 | switch A +--------------------------+ switch B |
1868 +-----+----+ +-----++---+
1871 +-------------+ host1 +---------------+
1874 In this configuration, there is a link between the two
1875 switches (ISL, or inter switch link), and multiple ports connecting to
1876 the outside world ("port3" on each switch). There is no technical
1877 reason that this could not be extended to a third switch.
1879 11.2.1 HA Bonding Mode Selection for Multiple Switch Topology
1880 -------------------------------------------------------------
1882 In a topology such as the example above, the active-backup and
1883 broadcast modes are the only useful bonding modes when optimizing for
1884 availability; the other modes require all links to terminate on the
1885 same peer for them to behave rationally.
1887 active-backup: This is generally the preferred mode, particularly if
1888 the switches have an ISL and play together well. If the
1889 network configuration is such that one switch is specifically
1890 a backup switch (e.g., has lower capacity, higher cost, etc),
1891 then the primary option can be used to insure that the
1892 preferred link is always used when it is available.
1894 broadcast: This mode is really a special purpose mode, and is suitable
1895 only for very specific needs. For example, if the two
1896 switches are not connected (no ISL), and the networks beyond
1897 them are totally independent. In this case, if it is
1898 necessary for some specific one-way traffic to reach both
1899 independent networks, then the broadcast mode may be suitable.
1901 11.2.2 HA Link Monitoring Selection for Multiple Switch Topology
1902 ----------------------------------------------------------------
1904 The choice of link monitoring ultimately depends upon your
1905 switch. If the switch can reliably fail ports in response to other
1906 failures, then either the MII or ARP monitors should work. For
1907 example, in the above example, if the "port3" link fails at the remote
1908 end, the MII monitor has no direct means to detect this. The ARP
1909 monitor could be configured with a target at the remote end of port3,
1910 thus detecting that failure without switch support.
1912 In general, however, in a multiple switch topology, the ARP
1913 monitor can provide a higher level of reliability in detecting end to
1914 end connectivity failures (which may be caused by the failure of any
1915 individual component to pass traffic for any reason). Additionally,
1916 the ARP monitor should be configured with multiple targets (at least
1917 one for each switch in the network). This will insure that,
1918 regardless of which switch is active, the ARP monitor has a suitable
1921 Note, also, that of late many switches now support a functionality
1922 generally referred to as "trunk failover." This is a feature of the
1923 switch that causes the link state of a particular switch port to be set
1924 down (or up) when the state of another switch port goes down (or up).
1925 Its purpose is to propagate link failures from logically "exterior" ports
1926 to the logically "interior" ports that bonding is able to monitor via
1927 miimon. Availability and configuration for trunk failover varies by
1928 switch, but this can be a viable alternative to the ARP monitor when using
1931 12. Configuring Bonding for Maximum Throughput
1932 ==============================================
1934 12.1 Maximizing Throughput in a Single Switch Topology
1935 ------------------------------------------------------
1937 In a single switch configuration, the best method to maximize
1938 throughput depends upon the application and network environment. The
1939 various load balancing modes each have strengths and weaknesses in
1940 different environments, as detailed below.
1942 For this discussion, we will break down the topologies into
1943 two categories. Depending upon the destination of most traffic, we
1944 categorize them into either "gatewayed" or "local" configurations.
1946 In a gatewayed configuration, the "switch" is acting primarily
1947 as a router, and the majority of traffic passes through this router to
1948 other networks. An example would be the following:
1951 +----------+ +----------+
1952 | |eth0 port1| | to other networks
1953 | Host A +---------------------+ router +------------------->
1954 | +---------------------+ | Hosts B and C are out
1955 | |eth1 port2| | here somewhere
1956 +----------+ +----------+
1958 The router may be a dedicated router device, or another host
1959 acting as a gateway. For our discussion, the important point is that
1960 the majority of traffic from Host A will pass through the router to
1961 some other network before reaching its final destination.
1963 In a gatewayed network configuration, although Host A may
1964 communicate with many other systems, all of its traffic will be sent
1965 and received via one other peer on the local network, the router.
1967 Note that the case of two systems connected directly via
1968 multiple physical links is, for purposes of configuring bonding, the
1969 same as a gatewayed configuration. In that case, it happens that all
1970 traffic is destined for the "gateway" itself, not some other network
1973 In a local configuration, the "switch" is acting primarily as
1974 a switch, and the majority of traffic passes through this switch to
1975 reach other stations on the same network. An example would be the
1978 +----------+ +----------+ +--------+
1979 | |eth0 port1| +-------+ Host B |
1980 | Host A +------------+ switch |port3 +--------+
1981 | +------------+ | +--------+
1982 | |eth1 port2| +------------------+ Host C |
1983 +----------+ +----------+port4 +--------+
1986 Again, the switch may be a dedicated switch device, or another
1987 host acting as a gateway. For our discussion, the important point is
1988 that the majority of traffic from Host A is destined for other hosts
1989 on the same local network (Hosts B and C in the above example).
1991 In summary, in a gatewayed configuration, traffic to and from
1992 the bonded device will be to the same MAC level peer on the network
1993 (the gateway itself, i.e., the router), regardless of its final
1994 destination. In a local configuration, traffic flows directly to and
1995 from the final destinations, thus, each destination (Host B, Host C)
1996 will be addressed directly by their individual MAC addresses.
1998 This distinction between a gatewayed and a local network
1999 configuration is important because many of the load balancing modes
2000 available use the MAC addresses of the local network source and
2001 destination to make load balancing decisions. The behavior of each
2002 mode is described below.
2005 12.1.1 MT Bonding Mode Selection for Single Switch Topology
2006 -----------------------------------------------------------
2008 This configuration is the easiest to set up and to understand,
2009 although you will have to decide which bonding mode best suits your
2010 needs. The trade offs for each mode are detailed below:
2012 balance-rr: This mode is the only mode that will permit a single
2013 TCP/IP connection to stripe traffic across multiple
2014 interfaces. It is therefore the only mode that will allow a
2015 single TCP/IP stream to utilize more than one interface's
2016 worth of throughput. This comes at a cost, however: the
2017 striping generally results in peer systems receiving packets out
2018 of order, causing TCP/IP's congestion control system to kick
2019 in, often by retransmitting segments.
2021 It is possible to adjust TCP/IP's congestion limits by
2022 altering the net.ipv4.tcp_reordering sysctl parameter. The
2023 usual default value is 3, and the maximum useful value is 127.
2024 For a four interface balance-rr bond, expect that a single
2025 TCP/IP stream will utilize no more than approximately 2.3
2026 interface's worth of throughput, even after adjusting
2029 Note that the fraction of packets that will be delivered out of
2030 order is highly variable, and is unlikely to be zero. The level
2031 of reordering depends upon a variety of factors, including the
2032 networking interfaces, the switch, and the topology of the
2033 configuration. Speaking in general terms, higher speed network
2034 cards produce more reordering (due to factors such as packet
2035 coalescing), and a "many to many" topology will reorder at a
2036 higher rate than a "many slow to one fast" configuration.
2038 Many switches do not support any modes that stripe traffic
2039 (instead choosing a port based upon IP or MAC level addresses);
2040 for those devices, traffic for a particular connection flowing
2041 through the switch to a balance-rr bond will not utilize greater
2042 than one interface's worth of bandwidth.
2044 If you are utilizing protocols other than TCP/IP, UDP for
2045 example, and your application can tolerate out of order
2046 delivery, then this mode can allow for single stream datagram
2047 performance that scales near linearly as interfaces are added
2050 This mode requires the switch to have the appropriate ports
2051 configured for "etherchannel" or "trunking."
2053 active-backup: There is not much advantage in this network topology to
2054 the active-backup mode, as the inactive backup devices are all
2055 connected to the same peer as the primary. In this case, a
2056 load balancing mode (with link monitoring) will provide the
2057 same level of network availability, but with increased
2058 available bandwidth. On the plus side, active-backup mode
2059 does not require any configuration of the switch, so it may
2060 have value if the hardware available does not support any of
2061 the load balance modes.
2063 balance-xor: This mode will limit traffic such that packets destined
2064 for specific peers will always be sent over the same
2065 interface. Since the destination is determined by the MAC
2066 addresses involved, this mode works best in a "local" network
2067 configuration (as described above), with destinations all on
2068 the same local network. This mode is likely to be suboptimal
2069 if all your traffic is passed through a single router (i.e., a
2070 "gatewayed" network configuration, as described above).
2072 As with balance-rr, the switch ports need to be configured for
2073 "etherchannel" or "trunking."
2075 broadcast: Like active-backup, there is not much advantage to this
2076 mode in this type of network topology.
2078 802.3ad: This mode can be a good choice for this type of network
2079 topology. The 802.3ad mode is an IEEE standard, so all peers
2080 that implement 802.3ad should interoperate well. The 802.3ad
2081 protocol includes automatic configuration of the aggregates,
2082 so minimal manual configuration of the switch is needed
2083 (typically only to designate that some set of devices is
2084 available for 802.3ad). The 802.3ad standard also mandates
2085 that frames be delivered in order (within certain limits), so
2086 in general single connections will not see misordering of
2087 packets. The 802.3ad mode does have some drawbacks: the
2088 standard mandates that all devices in the aggregate operate at
2089 the same speed and duplex. Also, as with all bonding load
2090 balance modes other than balance-rr, no single connection will
2091 be able to utilize more than a single interface's worth of
2094 Additionally, the linux bonding 802.3ad implementation
2095 distributes traffic by peer (using an XOR of MAC addresses),
2096 so in a "gatewayed" configuration, all outgoing traffic will
2097 generally use the same device. Incoming traffic may also end
2098 up on a single device, but that is dependent upon the
2099 balancing policy of the peer's 8023.ad implementation. In a
2100 "local" configuration, traffic will be distributed across the
2101 devices in the bond.
2103 Finally, the 802.3ad mode mandates the use of the MII monitor,
2104 therefore, the ARP monitor is not available in this mode.
2106 balance-tlb: The balance-tlb mode balances outgoing traffic by peer.
2107 Since the balancing is done according to MAC address, in a
2108 "gatewayed" configuration (as described above), this mode will
2109 send all traffic across a single device. However, in a
2110 "local" network configuration, this mode balances multiple
2111 local network peers across devices in a vaguely intelligent
2112 manner (not a simple XOR as in balance-xor or 802.3ad mode),
2113 so that mathematically unlucky MAC addresses (i.e., ones that
2114 XOR to the same value) will not all "bunch up" on a single
2117 Unlike 802.3ad, interfaces may be of differing speeds, and no
2118 special switch configuration is required. On the down side,
2119 in this mode all incoming traffic arrives over a single
2120 interface, this mode requires certain ethtool support in the
2121 network device driver of the slave interfaces, and the ARP
2122 monitor is not available.
2124 balance-alb: This mode is everything that balance-tlb is, and more.
2125 It has all of the features (and restrictions) of balance-tlb,
2126 and will also balance incoming traffic from local network
2127 peers (as described in the Bonding Module Options section,
2130 The only additional down side to this mode is that the network
2131 device driver must support changing the hardware address while
2134 12.1.2 MT Link Monitoring for Single Switch Topology
2135 ----------------------------------------------------
2137 The choice of link monitoring may largely depend upon which
2138 mode you choose to use. The more advanced load balancing modes do not
2139 support the use of the ARP monitor, and are thus restricted to using
2140 the MII monitor (which does not provide as high a level of end to end
2141 assurance as the ARP monitor).
2143 12.2 Maximum Throughput in a Multiple Switch Topology
2144 -----------------------------------------------------
2146 Multiple switches may be utilized to optimize for throughput
2147 when they are configured in parallel as part of an isolated network
2148 between two or more systems, for example:
2154 +--------+ | +---------+
2156 +------+---+ +-----+----+ +-----+----+
2157 | Switch A | | Switch B | | Switch C |
2158 +------+---+ +-----+----+ +-----+----+
2160 +--------+ | +---------+
2166 In this configuration, the switches are isolated from one
2167 another. One reason to employ a topology such as this is for an
2168 isolated network with many hosts (a cluster configured for high
2169 performance, for example), using multiple smaller switches can be more
2170 cost effective than a single larger switch, e.g., on a network with 24
2171 hosts, three 24 port switches can be significantly less expensive than
2172 a single 72 port switch.
2174 If access beyond the network is required, an individual host
2175 can be equipped with an additional network device connected to an
2176 external network; this host then additionally acts as a gateway.
2178 12.2.1 MT Bonding Mode Selection for Multiple Switch Topology
2179 -------------------------------------------------------------
2181 In actual practice, the bonding mode typically employed in
2182 configurations of this type is balance-rr. Historically, in this
2183 network configuration, the usual caveats about out of order packet
2184 delivery are mitigated by the use of network adapters that do not do
2185 any kind of packet coalescing (via the use of NAPI, or because the
2186 device itself does not generate interrupts until some number of
2187 packets has arrived). When employed in this fashion, the balance-rr
2188 mode allows individual connections between two hosts to effectively
2189 utilize greater than one interface's bandwidth.
2191 12.2.2 MT Link Monitoring for Multiple Switch Topology
2192 ------------------------------------------------------
2194 Again, in actual practice, the MII monitor is most often used
2195 in this configuration, as performance is given preference over
2196 availability. The ARP monitor will function in this topology, but its
2197 advantages over the MII monitor are mitigated by the volume of probes
2198 needed as the number of systems involved grows (remember that each
2199 host in the network is configured with bonding).
2201 13. Switch Behavior Issues
2202 ==========================
2204 13.1 Link Establishment and Failover Delays
2205 -------------------------------------------
2207 Some switches exhibit undesirable behavior with regard to the
2208 timing of link up and down reporting by the switch.
2210 First, when a link comes up, some switches may indicate that
2211 the link is up (carrier available), but not pass traffic over the
2212 interface for some period of time. This delay is typically due to
2213 some type of autonegotiation or routing protocol, but may also occur
2214 during switch initialization (e.g., during recovery after a switch
2215 failure). If you find this to be a problem, specify an appropriate
2216 value to the updelay bonding module option to delay the use of the
2217 relevant interface(s).
2219 Second, some switches may "bounce" the link state one or more
2220 times while a link is changing state. This occurs most commonly while
2221 the switch is initializing. Again, an appropriate updelay value may
2224 Note that when a bonding interface has no active links, the
2225 driver will immediately reuse the first link that goes up, even if the
2226 updelay parameter has been specified (the updelay is ignored in this
2227 case). If there are slave interfaces waiting for the updelay timeout
2228 to expire, the interface that first went into that state will be
2229 immediately reused. This reduces down time of the network if the
2230 value of updelay has been overestimated, and since this occurs only in
2231 cases with no connectivity, there is no additional penalty for
2232 ignoring the updelay.
2234 In addition to the concerns about switch timings, if your
2235 switches take a long time to go into backup mode, it may be desirable
2236 to not activate a backup interface immediately after a link goes down.
2237 Failover may be delayed via the downdelay bonding module option.
2239 13.2 Duplicated Incoming Packets
2240 --------------------------------
2242 NOTE: Starting with version 3.0.2, the bonding driver has logic to
2243 suppress duplicate packets, which should largely eliminate this problem.
2244 The following description is kept for reference.
2246 It is not uncommon to observe a short burst of duplicated
2247 traffic when the bonding device is first used, or after it has been
2248 idle for some period of time. This is most easily observed by issuing
2249 a "ping" to some other host on the network, and noticing that the
2250 output from ping flags duplicates (typically one per slave).
2252 For example, on a bond in active-backup mode with five slaves
2253 all connected to one switch, the output may appear as follows:
2256 PING 10.0.4.2 (10.0.4.2) from 10.0.3.10 : 56(84) bytes of data.
2257 64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.7 ms
2258 64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
2259 64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
2260 64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
2261 64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
2262 64 bytes from 10.0.4.2: icmp_seq=2 ttl=64 time=0.216 ms
2263 64 bytes from 10.0.4.2: icmp_seq=3 ttl=64 time=0.267 ms
2264 64 bytes from 10.0.4.2: icmp_seq=4 ttl=64 time=0.222 ms
2266 This is not due to an error in the bonding driver, rather, it
2267 is a side effect of how many switches update their MAC forwarding
2268 tables. Initially, the switch does not associate the MAC address in
2269 the packet with a particular switch port, and so it may send the
2270 traffic to all ports until its MAC forwarding table is updated. Since
2271 the interfaces attached to the bond may occupy multiple ports on a
2272 single switch, when the switch (temporarily) floods the traffic to all
2273 ports, the bond device receives multiple copies of the same packet
2274 (one per slave device).
2276 The duplicated packet behavior is switch dependent, some
2277 switches exhibit this, and some do not. On switches that display this
2278 behavior, it can be induced by clearing the MAC forwarding table (on
2279 most Cisco switches, the privileged command "clear mac address-table
2280 dynamic" will accomplish this).
2282 14. Hardware Specific Considerations
2283 ====================================
2285 This section contains additional information for configuring
2286 bonding on specific hardware platforms, or for interfacing bonding
2287 with particular switches or other devices.
2289 14.1 IBM BladeCenter
2290 --------------------
2292 This applies to the JS20 and similar systems.
2294 On the JS20 blades, the bonding driver supports only
2295 balance-rr, active-backup, balance-tlb and balance-alb modes. This is
2296 largely due to the network topology inside the BladeCenter, detailed
2299 JS20 network adapter information
2300 --------------------------------
2302 All JS20s come with two Broadcom Gigabit Ethernet ports
2303 integrated on the planar (that's "motherboard" in IBM-speak). In the
2304 BladeCenter chassis, the eth0 port of all JS20 blades is hard wired to
2305 I/O Module #1; similarly, all eth1 ports are wired to I/O Module #2.
2306 An add-on Broadcom daughter card can be installed on a JS20 to provide
2307 two more Gigabit Ethernet ports. These ports, eth2 and eth3, are
2308 wired to I/O Modules 3 and 4, respectively.
2310 Each I/O Module may contain either a switch or a passthrough
2311 module (which allows ports to be directly connected to an external
2312 switch). Some bonding modes require a specific BladeCenter internal
2313 network topology in order to function; these are detailed below.
2315 Additional BladeCenter-specific networking information can be
2316 found in two IBM Redbooks (www.ibm.com/redbooks):
2318 "IBM eServer BladeCenter Networking Options"
2319 "IBM eServer BladeCenter Layer 2-7 Network Switching"
2321 BladeCenter networking configuration
2322 ------------------------------------
2324 Because a BladeCenter can be configured in a very large number
2325 of ways, this discussion will be confined to describing basic
2328 Normally, Ethernet Switch Modules (ESMs) are used in I/O
2329 modules 1 and 2. In this configuration, the eth0 and eth1 ports of a
2330 JS20 will be connected to different internal switches (in the
2331 respective I/O modules).
2333 A passthrough module (OPM or CPM, optical or copper,
2334 passthrough module) connects the I/O module directly to an external
2335 switch. By using PMs in I/O module #1 and #2, the eth0 and eth1
2336 interfaces of a JS20 can be redirected to the outside world and
2337 connected to a common external switch.
2339 Depending upon the mix of ESMs and PMs, the network will
2340 appear to bonding as either a single switch topology (all PMs) or as a
2341 multiple switch topology (one or more ESMs, zero or more PMs). It is
2342 also possible to connect ESMs together, resulting in a configuration
2343 much like the example in "High Availability in a Multiple Switch
2346 Requirements for specific modes
2347 -------------------------------
2349 The balance-rr mode requires the use of passthrough modules
2350 for devices in the bond, all connected to an common external switch.
2351 That switch must be configured for "etherchannel" or "trunking" on the
2352 appropriate ports, as is usual for balance-rr.
2354 The balance-alb and balance-tlb modes will function with
2355 either switch modules or passthrough modules (or a mix). The only
2356 specific requirement for these modes is that all network interfaces
2357 must be able to reach all destinations for traffic sent over the
2358 bonding device (i.e., the network must converge at some point outside
2361 The active-backup mode has no additional requirements.
2363 Link monitoring issues
2364 ----------------------
2366 When an Ethernet Switch Module is in place, only the ARP
2367 monitor will reliably detect link loss to an external switch. This is
2368 nothing unusual, but examination of the BladeCenter cabinet would
2369 suggest that the "external" network ports are the ethernet ports for
2370 the system, when it fact there is a switch between these "external"
2371 ports and the devices on the JS20 system itself. The MII monitor is
2372 only able to detect link failures between the ESM and the JS20 system.
2374 When a passthrough module is in place, the MII monitor does
2375 detect failures to the "external" port, which is then directly
2376 connected to the JS20 system.
2381 The Serial Over LAN (SoL) link is established over the primary
2382 ethernet (eth0) only, therefore, any loss of link to eth0 will result
2383 in losing your SoL connection. It will not fail over with other
2384 network traffic, as the SoL system is beyond the control of the
2387 It may be desirable to disable spanning tree on the switch
2388 (either the internal Ethernet Switch Module, or an external switch) to
2389 avoid fail-over delay issues when using bonding.
2392 15. Frequently Asked Questions
2393 ==============================
2397 Yes. The old 2.0.xx channel bonding patch was not SMP safe.
2398 The new driver was designed to be SMP safe from the start.
2400 2. What type of cards will work with it?
2402 Any Ethernet type cards (you can even mix cards - a Intel
2403 EtherExpress PRO/100 and a 3com 3c905b, for example). For most modes,
2404 devices need not be of the same speed.
2406 Starting with version 3.2.1, bonding also supports Infiniband
2407 slaves in active-backup mode.
2409 3. How many bonding devices can I have?
2413 4. How many slaves can a bonding device have?
2415 This is limited only by the number of network interfaces Linux
2416 supports and/or the number of network cards you can place in your
2419 5. What happens when a slave link dies?
2421 If link monitoring is enabled, then the failing device will be
2422 disabled. The active-backup mode will fail over to a backup link, and
2423 other modes will ignore the failed link. The link will continue to be
2424 monitored, and should it recover, it will rejoin the bond (in whatever
2425 manner is appropriate for the mode). See the sections on High
2426 Availability and the documentation for each mode for additional
2429 Link monitoring can be enabled via either the miimon or
2430 arp_interval parameters (described in the module parameters section,
2431 above). In general, miimon monitors the carrier state as sensed by
2432 the underlying network device, and the arp monitor (arp_interval)
2433 monitors connectivity to another host on the local network.
2435 If no link monitoring is configured, the bonding driver will
2436 be unable to detect link failures, and will assume that all links are
2437 always available. This will likely result in lost packets, and a
2438 resulting degradation of performance. The precise performance loss
2439 depends upon the bonding mode and network configuration.
2441 6. Can bonding be used for High Availability?
2443 Yes. See the section on High Availability for details.
2445 7. Which switches/systems does it work with?
2447 The full answer to this depends upon the desired mode.
2449 In the basic balance modes (balance-rr and balance-xor), it
2450 works with any system that supports etherchannel (also called
2451 trunking). Most managed switches currently available have such
2452 support, and many unmanaged switches as well.
2454 The advanced balance modes (balance-tlb and balance-alb) do
2455 not have special switch requirements, but do need device drivers that
2456 support specific features (described in the appropriate section under
2457 module parameters, above).
2459 In 802.3ad mode, it works with systems that support IEEE
2460 802.3ad Dynamic Link Aggregation. Most managed and many unmanaged
2461 switches currently available support 802.3ad.
2463 The active-backup mode should work with any Layer-II switch.
2465 8. Where does a bonding device get its MAC address from?
2467 When using slave devices that have fixed MAC addresses, or when
2468 the fail_over_mac option is enabled, the bonding device's MAC address is
2469 the MAC address of the active slave.
2471 For other configurations, if not explicitly configured (with
2472 ifconfig or ip link), the MAC address of the bonding device is taken from
2473 its first slave device. This MAC address is then passed to all following
2474 slaves and remains persistent (even if the first slave is removed) until
2475 the bonding device is brought down or reconfigured.
2477 If you wish to change the MAC address, you can set it with
2478 ifconfig or ip link:
2480 # ifconfig bond0 hw ether 00:11:22:33:44:55
2482 # ip link set bond0 address 66:77:88:99:aa:bb
2484 The MAC address can be also changed by bringing down/up the
2485 device and then changing its slaves (or their order):
2487 # ifconfig bond0 down ; modprobe -r bonding
2488 # ifconfig bond0 .... up
2489 # ifenslave bond0 eth...
2491 This method will automatically take the address from the next
2492 slave that is added.
2494 To restore your slaves' MAC addresses, you need to detach them
2495 from the bond (`ifenslave -d bond0 eth0'). The bonding driver will
2496 then restore the MAC addresses that the slaves had before they were
2499 16. Resources and Links
2500 =======================
2502 The latest version of the bonding driver can be found in the latest
2503 version of the linux kernel, found on http://kernel.org
2505 The latest version of this document can be found in either the latest
2506 kernel source (named Documentation/networking/bonding.txt), or on the
2507 bonding sourceforge site:
2509 http://www.sourceforge.net/projects/bonding
2511 Discussions regarding the bonding driver take place primarily on the
2512 bonding-devel mailing list, hosted at sourceforge.net. If you have
2513 questions or problems, post them to the list. The list address is:
2515 bonding-devel@lists.sourceforge.net
2517 The administrative interface (to subscribe or unsubscribe) can
2520 https://lists.sourceforge.net/lists/listinfo/bonding-devel
2522 Donald Becker's Ethernet Drivers and diag programs may be found at :
2523 - http://web.archive.org/web/*/http://www.scyld.com/network/
2525 You will also find a lot of information regarding Ethernet, NWay, MII,
2526 etc. at www.scyld.com.
git.openpandora.org / pandora-kernel.git / blob