2 Linux Ethernet Bonding Driver HOWTO
4 Latest update: 12 November 2007
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
53 4. Querying Bonding Configuration
54 4.1 Bonding Configuration
55 4.2 Network Configuration
57 5. Switch Configuration
59 6. 802.1q VLAN Support
62 7.1 ARP Monitor Operation
63 7.2 Configuring Multiple ARP Targets
64 7.3 MII Monitor Operation
66 8. Potential Trouble Sources
67 8.1 Adventures in Routing
68 8.2 Ethernet Device Renaming
69 8.3 Painfully Slow Or No Failed Link Detection By Miimon
75 11. Configuring Bonding for High Availability
76 11.1 High Availability in a Single Switch Topology
77 11.2 High Availability in a Multiple Switch Topology
78 11.2.1 HA Bonding Mode Selection for Multiple Switch Topology
79 11.2.2 HA Link Monitoring for Multiple Switch Topology
81 12. Configuring Bonding for Maximum Throughput
82 12.1 Maximum Throughput in a Single Switch Topology
83 12.1.1 MT Bonding Mode Selection for Single Switch Topology
84 12.1.2 MT Link Monitoring for Single Switch Topology
85 12.2 Maximum Throughput in a Multiple Switch Topology
86 12.2.1 MT Bonding Mode Selection for Multiple Switch Topology
87 12.2.2 MT Link Monitoring for Multiple Switch Topology
89 13. Switch Behavior Issues
90 13.1 Link Establishment and Failover Delays
91 13.2 Duplicated Incoming Packets
93 14. Hardware Specific Considerations
96 15. Frequently Asked Questions
98 16. Resources and Links
101 1. Bonding Driver Installation
102 ==============================
104 Most popular distro kernels ship with the bonding driver
105 already available as a module and the ifenslave user level control
106 program installed and ready for use. If your distro does not, or you
107 have need to compile bonding from source (e.g., configuring and
108 installing a mainline kernel from kernel.org), you'll need to perform
111 1.1 Configure and build the kernel with bonding
112 -----------------------------------------------
114 The current version of the bonding driver is available in the
115 drivers/net/bonding subdirectory of the most recent kernel source
116 (which is available on http://kernel.org). Most users "rolling their
117 own" will want to use the most recent kernel from kernel.org.
119 Configure kernel with "make menuconfig" (or "make xconfig" or
120 "make config"), then select "Bonding driver support" in the "Network
121 device support" section. It is recommended that you configure the
122 driver as module since it is currently the only way to pass parameters
123 to the driver or configure more than one bonding device.
125 Build and install the new kernel and modules, then continue
126 below to install ifenslave.
128 1.2 Install ifenslave Control Utility
129 -------------------------------------
131 The ifenslave user level control program is included in the
132 kernel source tree, in the file Documentation/networking/ifenslave.c.
133 It is generally recommended that you use the ifenslave that
134 corresponds to the kernel that you are using (either from the same
135 source tree or supplied with the distro), however, ifenslave
136 executables from older kernels should function (but features newer
137 than the ifenslave release are not supported). Running an ifenslave
138 that is newer than the kernel is not supported, and may or may not
141 To install ifenslave, do the following:
143 # gcc -Wall -O -I/usr/src/linux/include ifenslave.c -o ifenslave
144 # cp ifenslave /sbin/ifenslave
146 If your kernel source is not in "/usr/src/linux," then replace
147 "/usr/src/linux/include" in the above with the location of your kernel
148 source include directory.
150 You may wish to back up any existing /sbin/ifenslave, or, for
151 testing or informal use, tag the ifenslave to the kernel version
152 (e.g., name the ifenslave executable /sbin/ifenslave-2.6.10).
156 If you omit the "-I" or specify an incorrect directory, you
157 may end up with an ifenslave that is incompatible with the kernel
158 you're trying to build it for. Some distros (e.g., Red Hat from 7.1
159 onwards) do not have /usr/include/linux symbolically linked to the
160 default kernel source include directory.
162 SECOND IMPORTANT NOTE:
163 If you plan to configure bonding using sysfs, you do not need
166 2. Bonding Driver Options
167 =========================
169 Options for the bonding driver are supplied as parameters to the
170 bonding module at load time, or are specified via sysfs.
172 Module options may be given as command line arguments to the
173 insmod or modprobe command, but are usually specified in either the
174 /etc/modules.conf or /etc/modprobe.conf configuration file, or in a
175 distro-specific configuration file (some of which are detailed in the next
178 Details on bonding support for sysfs is provided in the
179 "Configuring Bonding Manually via Sysfs" section, below.
181 The available bonding driver parameters are listed below. If a
182 parameter is not specified the default value is used. When initially
183 configuring a bond, it is recommended "tail -f /var/log/messages" be
184 run in a separate window to watch for bonding driver error messages.
186 It is critical that either the miimon or arp_interval and
187 arp_ip_target parameters be specified, otherwise serious network
188 degradation will occur during link failures. Very few devices do not
189 support at least miimon, so there is really no reason not to use it.
191 Options with textual values will accept either the text name
192 or, for backwards compatibility, the option value. E.g.,
193 "mode=802.3ad" and "mode=4" set the same mode.
195 The parameters are as follows:
199 Specifies the ARP link monitoring frequency in milliseconds.
201 The ARP monitor works by periodically checking the slave
202 devices to determine whether they have sent or received
203 traffic recently (the precise criteria depends upon the
204 bonding mode, and the state of the slave). Regular traffic is
205 generated via ARP probes issued for the addresses specified by
206 the arp_ip_target option.
208 This behavior can be modified by the arp_validate option,
211 If ARP monitoring is used in an etherchannel compatible mode
212 (modes 0 and 2), the switch should be configured in a mode
213 that evenly distributes packets across all links. If the
214 switch is configured to distribute the packets in an XOR
215 fashion, all replies from the ARP targets will be received on
216 the same link which could cause the other team members to
217 fail. ARP monitoring should not be used in conjunction with
218 miimon. A value of 0 disables ARP monitoring. The default
223 Specifies the IP addresses to use as ARP monitoring peers when
224 arp_interval is > 0. These are the targets of the ARP request
225 sent to determine the health of the link to the targets.
226 Specify these values in ddd.ddd.ddd.ddd format. Multiple IP
227 addresses must be separated by a comma. At least one IP
228 address must be given for ARP monitoring to function. The
229 maximum number of targets that can be specified is 16. The
230 default value is no IP addresses.
234 Specifies whether or not ARP probes and replies should be
235 validated in the active-backup mode. This causes the ARP
236 monitor to examine the incoming ARP requests and replies, and
237 only consider a slave to be up if it is receiving the
238 appropriate ARP traffic.
244 No validation is performed. This is the default.
248 Validation is performed only for the active slave.
252 Validation is performed only for backup slaves.
256 Validation is performed for all slaves.
258 For the active slave, the validation checks ARP replies to
259 confirm that they were generated by an arp_ip_target. Since
260 backup slaves do not typically receive these replies, the
261 validation performed for backup slaves is on the ARP request
262 sent out via the active slave. It is possible that some
263 switch or network configurations may result in situations
264 wherein the backup slaves do not receive the ARP requests; in
265 such a situation, validation of backup slaves must be
268 This option is useful in network configurations in which
269 multiple bonding hosts are concurrently issuing ARPs to one or
270 more targets beyond a common switch. Should the link between
271 the switch and target fail (but not the switch itself), the
272 probe traffic generated by the multiple bonding instances will
273 fool the standard ARP monitor into considering the links as
274 still up. Use of the arp_validate option can resolve this, as
275 the ARP monitor will only consider ARP requests and replies
276 associated with its own instance of bonding.
278 This option was added in bonding version 3.1.0.
282 Specifies the time, in milliseconds, to wait before disabling
283 a slave after a link failure has been detected. This option
284 is only valid for the miimon link monitor. The downdelay
285 value should be a multiple of the miimon value; if not, it
286 will be rounded down to the nearest multiple. The default
291 Specifies whether active-backup mode should set all slaves to
292 the same MAC address at enslavement (the traditional
293 behavior), or, when enabled, perform special handling of the
294 bond's MAC address in accordance with the selected policy.
300 This setting disables fail_over_mac, and causes
301 bonding to set all slaves of an active-backup bond to
302 the same MAC address at enslavement time. This is the
307 The "active" fail_over_mac policy indicates that the
308 MAC address of the bond should always be the MAC
309 address of the currently active slave. The MAC
310 address of the slaves is not changed; instead, the MAC
311 address of the bond changes during a failover.
313 This policy is useful for devices that cannot ever
314 alter their MAC address, or for devices that refuse
315 incoming broadcasts with their own source MAC (which
316 interferes with the ARP monitor).
318 The down side of this policy is that every device on
319 the network must be updated via gratuitous ARP,
320 vs. just updating a switch or set of switches (which
321 often takes place for any traffic, not just ARP
322 traffic, if the switch snoops incoming traffic to
323 update its tables) for the traditional method. If the
324 gratuitous ARP is lost, communication may be
327 When this policy is used in conjuction with the mii
328 monitor, devices which assert link up prior to being
329 able to actually transmit and receive are particularly
330 susecptible to loss of the gratuitous ARP, and an
331 appropriate updelay setting may be required.
335 The "follow" fail_over_mac policy causes the MAC
336 address of the bond to be selected normally (normally
337 the MAC address of the first slave added to the bond).
338 However, the second and subsequent slaves are not set
339 to this MAC address while they are in a backup role; a
340 slave is programmed with the bond's MAC address at
341 failover time (and the formerly active slave receives
342 the newly active slave's MAC address).
344 This policy is useful for multiport devices that
345 either become confused or incur a performance penalty
346 when multiple ports are programmed with the same MAC
350 The default policy is none, unless the first slave cannot
351 change its MAC address, in which case the active policy is
354 This option may be modified via sysfs only when no slaves are
357 This option was added in bonding version 3.2.0. The "follow"
358 policy was added in bonding version 3.3.0.
362 Option specifying the rate in which we'll ask our link partner
363 to transmit LACPDU packets in 802.3ad mode. Possible values
367 Request partner to transmit LACPDUs every 30 seconds
370 Request partner to transmit LACPDUs every 1 second
376 Specifies the number of bonding devices to create for this
377 instance of the bonding driver. E.g., if max_bonds is 3, and
378 the bonding driver is not already loaded, then bond0, bond1
379 and bond2 will be created. The default value is 1.
383 Specifies the MII link monitoring frequency in milliseconds.
384 This determines how often the link state of each slave is
385 inspected for link failures. A value of zero disables MII
386 link monitoring. A value of 100 is a good starting point.
387 The use_carrier option, below, affects how the link state is
388 determined. See the High Availability section for additional
389 information. The default value is 0.
393 Specifies one of the bonding policies. The default is
394 balance-rr (round robin). Possible values are:
398 Round-robin policy: Transmit packets in sequential
399 order from the first available slave through the
400 last. This mode provides load balancing and fault
405 Active-backup policy: Only one slave in the bond is
406 active. A different slave becomes active if, and only
407 if, the active slave fails. The bond's MAC address is
408 externally visible on only one port (network adapter)
409 to avoid confusing the switch.
411 In bonding version 2.6.2 or later, when a failover
412 occurs in active-backup mode, bonding will issue one
413 or more gratuitous ARPs on the newly active slave.
414 One gratuitous ARP is issued for the bonding master
415 interface and each VLAN interfaces configured above
416 it, provided that the interface has at least one IP
417 address configured. Gratuitous ARPs issued for VLAN
418 interfaces are tagged with the appropriate VLAN id.
420 This mode provides fault tolerance. The primary
421 option, documented below, affects the behavior of this
426 XOR policy: Transmit based on the selected transmit
427 hash policy. The default policy is a simple [(source
428 MAC address XOR'd with destination MAC address) modulo
429 slave count]. Alternate transmit policies may be
430 selected via the xmit_hash_policy option, described
433 This mode provides load balancing and fault tolerance.
437 Broadcast policy: transmits everything on all slave
438 interfaces. This mode provides fault tolerance.
442 IEEE 802.3ad Dynamic link aggregation. Creates
443 aggregation groups that share the same speed and
444 duplex settings. Utilizes all slaves in the active
445 aggregator according to the 802.3ad specification.
447 Slave selection for outgoing traffic is done according
448 to the transmit hash policy, which may be changed from
449 the default simple XOR policy via the xmit_hash_policy
450 option, documented below. Note that not all transmit
451 policies may be 802.3ad compliant, particularly in
452 regards to the packet mis-ordering requirements of
453 section 43.2.4 of the 802.3ad standard. Differing
454 peer implementations will have varying tolerances for
459 1. Ethtool support in the base drivers for retrieving
460 the speed and duplex of each slave.
462 2. A switch that supports IEEE 802.3ad Dynamic link
465 Most switches will require some type of configuration
466 to enable 802.3ad mode.
470 Adaptive transmit load balancing: channel bonding that
471 does not require any special switch support. The
472 outgoing traffic is distributed according to the
473 current load (computed relative to the speed) on each
474 slave. Incoming traffic is received by the current
475 slave. If the receiving slave fails, another slave
476 takes over the MAC address of the failed receiving
481 Ethtool support in the base drivers for retrieving the
486 Adaptive load balancing: includes balance-tlb plus
487 receive load balancing (rlb) for IPV4 traffic, and
488 does not require any special switch support. The
489 receive load balancing is achieved by ARP negotiation.
490 The bonding driver intercepts the ARP Replies sent by
491 the local system on their way out and overwrites the
492 source hardware address with the unique hardware
493 address of one of the slaves in the bond such that
494 different peers use different hardware addresses for
497 Receive traffic from connections created by the server
498 is also balanced. When the local system sends an ARP
499 Request the bonding driver copies and saves the peer's
500 IP information from the ARP packet. When the ARP
501 Reply arrives from the peer, its hardware address is
502 retrieved and the bonding driver initiates an ARP
503 reply to this peer assigning it to one of the slaves
504 in the bond. A problematic outcome of using ARP
505 negotiation for balancing is that each time that an
506 ARP request is broadcast it uses the hardware address
507 of the bond. Hence, peers learn the hardware address
508 of the bond and the balancing of receive traffic
509 collapses to the current slave. This is handled by
510 sending updates (ARP Replies) to all the peers with
511 their individually assigned hardware address such that
512 the traffic is redistributed. Receive traffic is also
513 redistributed when a new slave is added to the bond
514 and when an inactive slave is re-activated. The
515 receive load is distributed sequentially (round robin)
516 among the group of highest speed slaves in the bond.
518 When a link is reconnected or a new slave joins the
519 bond the receive traffic is redistributed among all
520 active slaves in the bond by initiating ARP Replies
521 with the selected MAC address to each of the
522 clients. The updelay parameter (detailed below) must
523 be set to a value equal or greater than the switch's
524 forwarding delay so that the ARP Replies sent to the
525 peers will not be blocked by the switch.
529 1. Ethtool support in the base drivers for retrieving
530 the speed of each slave.
532 2. Base driver support for setting the hardware
533 address of a device while it is open. This is
534 required so that there will always be one slave in the
535 team using the bond hardware address (the
536 curr_active_slave) while having a unique hardware
537 address for each slave in the bond. If the
538 curr_active_slave fails its hardware address is
539 swapped with the new curr_active_slave that was
544 A string (eth0, eth2, etc) specifying which slave is the
545 primary device. The specified device will always be the
546 active slave while it is available. Only when the primary is
547 off-line will alternate devices be used. This is useful when
548 one slave is preferred over another, e.g., when one slave has
549 higher throughput than another.
551 The primary option is only valid for active-backup mode.
555 Specifies the time, in milliseconds, to wait before enabling a
556 slave after a link recovery has been detected. This option is
557 only valid for the miimon link monitor. The updelay value
558 should be a multiple of the miimon value; if not, it will be
559 rounded down to the nearest multiple. The default value is 0.
563 Specifies whether or not miimon should use MII or ETHTOOL
564 ioctls vs. netif_carrier_ok() to determine the link
565 status. The MII or ETHTOOL ioctls are less efficient and
566 utilize a deprecated calling sequence within the kernel. The
567 netif_carrier_ok() relies on the device driver to maintain its
568 state with netif_carrier_on/off; at this writing, most, but
569 not all, device drivers support this facility.
571 If bonding insists that the link is up when it should not be,
572 it may be that your network device driver does not support
573 netif_carrier_on/off. The default state for netif_carrier is
574 "carrier on," so if a driver does not support netif_carrier,
575 it will appear as if the link is always up. In this case,
576 setting use_carrier to 0 will cause bonding to revert to the
577 MII / ETHTOOL ioctl method to determine the link state.
579 A value of 1 enables the use of netif_carrier_ok(), a value of
580 0 will use the deprecated MII / ETHTOOL ioctls. The default
585 Selects the transmit hash policy to use for slave selection in
586 balance-xor and 802.3ad modes. Possible values are:
590 Uses XOR of hardware MAC addresses to generate the
593 (source MAC XOR destination MAC) modulo slave count
595 This algorithm will place all traffic to a particular
596 network peer on the same slave.
598 This algorithm is 802.3ad compliant.
602 This policy uses a combination of layer2 and layer3
603 protocol information to generate the hash.
605 Uses XOR of hardware MAC addresses and IP addresses to
606 generate the hash. The formula is
608 (((source IP XOR dest IP) AND 0xffff) XOR
609 ( source MAC XOR destination MAC ))
612 This algorithm will place all traffic to a particular
613 network peer on the same slave. For non-IP traffic,
614 the formula is the same as for the layer2 transmit
617 This policy is intended to provide a more balanced
618 distribution of traffic than layer2 alone, especially
619 in environments where a layer3 gateway device is
620 required to reach most destinations.
622 This algorithm is 802.3ad complient.
626 This policy uses upper layer protocol information,
627 when available, to generate the hash. This allows for
628 traffic to a particular network peer to span multiple
629 slaves, although a single connection will not span
632 The formula for unfragmented TCP and UDP packets is
634 ((source port XOR dest port) XOR
635 ((source IP XOR dest IP) AND 0xffff)
638 For fragmented TCP or UDP packets and all other IP
639 protocol traffic, the source and destination port
640 information is omitted. For non-IP traffic, the
641 formula is the same as for the layer2 transmit hash
644 This policy is intended to mimic the behavior of
645 certain switches, notably Cisco switches with PFC2 as
646 well as some Foundry and IBM products.
648 This algorithm is not fully 802.3ad compliant. A
649 single TCP or UDP conversation containing both
650 fragmented and unfragmented packets will see packets
651 striped across two interfaces. This may result in out
652 of order delivery. Most traffic types will not meet
653 this criteria, as TCP rarely fragments traffic, and
654 most UDP traffic is not involved in extended
655 conversations. Other implementations of 802.3ad may
656 or may not tolerate this noncompliance.
658 The default value is layer2. This option was added in bonding
659 version 2.6.3. In earlier versions of bonding, this parameter
660 does not exist, and the layer2 policy is the only policy. The
661 layer2+3 value was added for bonding version 3.2.2.
664 3. Configuring Bonding Devices
665 ==============================
667 You can configure bonding using either your distro's network
668 initialization scripts, or manually using either ifenslave or the
669 sysfs interface. Distros generally use one of two packages for the
670 network initialization scripts: initscripts or sysconfig. Recent
671 versions of these packages have support for bonding, while older
674 We will first describe the options for configuring bonding for
675 distros using versions of initscripts and sysconfig with full or
676 partial support for bonding, then provide information on enabling
677 bonding without support from the network initialization scripts (i.e.,
678 older versions of initscripts or sysconfig).
680 If you're unsure whether your distro uses sysconfig or
681 initscripts, or don't know if it's new enough, have no fear.
682 Determining this is fairly straightforward.
684 First, issue the command:
688 It will respond with a line of text starting with either
689 "initscripts" or "sysconfig," followed by some numbers. This is the
690 package that provides your network initialization scripts.
692 Next, to determine if your installation supports bonding,
695 $ grep ifenslave /sbin/ifup
697 If this returns any matches, then your initscripts or
698 sysconfig has support for bonding.
700 3.1 Configuration with Sysconfig Support
701 ----------------------------------------
703 This section applies to distros using a version of sysconfig
704 with bonding support, for example, SuSE Linux Enterprise Server 9.
706 SuSE SLES 9's networking configuration system does support
707 bonding, however, at this writing, the YaST system configuration
708 front end does not provide any means to work with bonding devices.
709 Bonding devices can be managed by hand, however, as follows.
711 First, if they have not already been configured, configure the
712 slave devices. On SLES 9, this is most easily done by running the
713 yast2 sysconfig configuration utility. The goal is for to create an
714 ifcfg-id file for each slave device. The simplest way to accomplish
715 this is to configure the devices for DHCP (this is only to get the
716 file ifcfg-id file created; see below for some issues with DHCP). The
717 name of the configuration file for each device will be of the form:
719 ifcfg-id-xx:xx:xx:xx:xx:xx
721 Where the "xx" portion will be replaced with the digits from
722 the device's permanent MAC address.
724 Once the set of ifcfg-id-xx:xx:xx:xx:xx:xx files has been
725 created, it is necessary to edit the configuration files for the slave
726 devices (the MAC addresses correspond to those of the slave devices).
727 Before editing, the file will contain multiple lines, and will look
733 UNIQUE='XNzu.WeZGOGF+4wE'
734 _nm_name='bus-pci-0001:61:01.0'
736 Change the BOOTPROTO and STARTMODE lines to the following:
741 Do not alter the UNIQUE or _nm_name lines. Remove any other
742 lines (USERCTL, etc).
744 Once the ifcfg-id-xx:xx:xx:xx:xx:xx files have been modified,
745 it's time to create the configuration file for the bonding device
746 itself. This file is named ifcfg-bondX, where X is the number of the
747 bonding device to create, starting at 0. The first such file is
748 ifcfg-bond0, the second is ifcfg-bond1, and so on. The sysconfig
749 network configuration system will correctly start multiple instances
752 The contents of the ifcfg-bondX file is as follows:
755 BROADCAST="10.0.2.255"
757 NETMASK="255.255.0.0"
762 BONDING_MODULE_OPTS="mode=active-backup miimon=100"
763 BONDING_SLAVE0="eth0"
764 BONDING_SLAVE1="bus-pci-0000:06:08.1"
766 Replace the sample BROADCAST, IPADDR, NETMASK and NETWORK
767 values with the appropriate values for your network.
769 The STARTMODE specifies when the device is brought online.
770 The possible values are:
772 onboot: The device is started at boot time. If you're not
773 sure, this is probably what you want.
775 manual: The device is started only when ifup is called
776 manually. Bonding devices may be configured this
777 way if you do not wish them to start automatically
778 at boot for some reason.
780 hotplug: The device is started by a hotplug event. This is not
781 a valid choice for a bonding device.
783 off or ignore: The device configuration is ignored.
785 The line BONDING_MASTER='yes' indicates that the device is a
786 bonding master device. The only useful value is "yes."
788 The contents of BONDING_MODULE_OPTS are supplied to the
789 instance of the bonding module for this device. Specify the options
790 for the bonding mode, link monitoring, and so on here. Do not include
791 the max_bonds bonding parameter; this will confuse the configuration
792 system if you have multiple bonding devices.
794 Finally, supply one BONDING_SLAVEn="slave device" for each
795 slave. where "n" is an increasing value, one for each slave. The
796 "slave device" is either an interface name, e.g., "eth0", or a device
797 specifier for the network device. The interface name is easier to
798 find, but the ethN names are subject to change at boot time if, e.g.,
799 a device early in the sequence has failed. The device specifiers
800 (bus-pci-0000:06:08.1 in the example above) specify the physical
801 network device, and will not change unless the device's bus location
802 changes (for example, it is moved from one PCI slot to another). The
803 example above uses one of each type for demonstration purposes; most
804 configurations will choose one or the other for all slave devices.
806 When all configuration files have been modified or created,
807 networking must be restarted for the configuration changes to take
808 effect. This can be accomplished via the following:
810 # /etc/init.d/network restart
812 Note that the network control script (/sbin/ifdown) will
813 remove the bonding module as part of the network shutdown processing,
814 so it is not necessary to remove the module by hand if, e.g., the
815 module parameters have changed.
817 Also, at this writing, YaST/YaST2 will not manage bonding
818 devices (they do not show bonding interfaces on its list of network
819 devices). It is necessary to edit the configuration file by hand to
820 change the bonding configuration.
822 Additional general options and details of the ifcfg file
823 format can be found in an example ifcfg template file:
825 /etc/sysconfig/network/ifcfg.template
827 Note that the template does not document the various BONDING_
828 settings described above, but does describe many of the other options.
830 3.1.1 Using DHCP with Sysconfig
831 -------------------------------
833 Under sysconfig, configuring a device with BOOTPROTO='dhcp'
834 will cause it to query DHCP for its IP address information. At this
835 writing, this does not function for bonding devices; the scripts
836 attempt to obtain the device address from DHCP prior to adding any of
837 the slave devices. Without active slaves, the DHCP requests are not
840 3.1.2 Configuring Multiple Bonds with Sysconfig
841 -----------------------------------------------
843 The sysconfig network initialization system is capable of
844 handling multiple bonding devices. All that is necessary is for each
845 bonding instance to have an appropriately configured ifcfg-bondX file
846 (as described above). Do not specify the "max_bonds" parameter to any
847 instance of bonding, as this will confuse sysconfig. If you require
848 multiple bonding devices with identical parameters, create multiple
851 Because the sysconfig scripts supply the bonding module
852 options in the ifcfg-bondX file, it is not necessary to add them to
853 the system /etc/modules.conf or /etc/modprobe.conf configuration file.
855 3.2 Configuration with Initscripts Support
856 ------------------------------------------
858 This section applies to distros using a recent version of
859 initscripts with bonding support, for example, Red Hat Enterprise Linux
860 version 3 or later, Fedora, etc. On these systems, the network
861 initialization scripts have knowledge of bonding, and can be configured to
862 control bonding devices. Note that older versions of the initscripts
863 package have lower levels of support for bonding; this will be noted where
866 These distros will not automatically load the network adapter
867 driver unless the ethX device is configured with an IP address.
868 Because of this constraint, users must manually configure a
869 network-script file for all physical adapters that will be members of
870 a bondX link. Network script files are located in the directory:
872 /etc/sysconfig/network-scripts
874 The file name must be prefixed with "ifcfg-eth" and suffixed
875 with the adapter's physical adapter number. For example, the script
876 for eth0 would be named /etc/sysconfig/network-scripts/ifcfg-eth0.
877 Place the following text in the file:
886 The DEVICE= line will be different for every ethX device and
887 must correspond with the name of the file, i.e., ifcfg-eth1 must have
888 a device line of DEVICE=eth1. The setting of the MASTER= line will
889 also depend on the final bonding interface name chosen for your bond.
890 As with other network devices, these typically start at 0, and go up
891 one for each device, i.e., the first bonding instance is bond0, the
892 second is bond1, and so on.
894 Next, create a bond network script. The file name for this
895 script will be /etc/sysconfig/network-scripts/ifcfg-bondX where X is
896 the number of the bond. For bond0 the file is named "ifcfg-bond0",
897 for bond1 it is named "ifcfg-bond1", and so on. Within that file,
898 place the following text:
902 NETMASK=255.255.255.0
904 BROADCAST=192.168.1.255
909 Be sure to change the networking specific lines (IPADDR,
910 NETMASK, NETWORK and BROADCAST) to match your network configuration.
912 For later versions of initscripts, such as that found with Fedora
913 7 and Red Hat Enterprise Linux version 5 (or later), it is possible, and,
914 indeed, preferable, to specify the bonding options in the ifcfg-bond0
915 file, e.g. a line of the format:
917 BONDING_OPTS="mode=active-backup arp_interval=60 arp_ip_target=+192.168.1.254"
919 will configure the bond with the specified options. The options
920 specified in BONDING_OPTS are identical to the bonding module parameters
921 except for the arp_ip_target field. Each target should be included as a
922 separate option and should be preceded by a '+' to indicate it should be
923 added to the list of queried targets, e.g.,
925 arp_ip_target=+192.168.1.1 arp_ip_target=+192.168.1.2
927 is the proper syntax to specify multiple targets. When specifying
928 options via BONDING_OPTS, it is not necessary to edit /etc/modules.conf or
931 For older versions of initscripts that do not support
932 BONDING_OPTS, it is necessary to edit /etc/modules.conf (or
933 /etc/modprobe.conf, depending upon your distro) to load the bonding module
934 with your desired options when the bond0 interface is brought up. The
935 following lines in /etc/modules.conf (or modprobe.conf) will load the
936 bonding module, and select its options:
939 options bond0 mode=balance-alb miimon=100
941 Replace the sample parameters with the appropriate set of
942 options for your configuration.
944 Finally run "/etc/rc.d/init.d/network restart" as root. This
945 will restart the networking subsystem and your bond link should be now
948 3.2.1 Using DHCP with Initscripts
949 ---------------------------------
951 Recent versions of initscripts (the versions supplied with Fedora
952 Core 3 and Red Hat Enterprise Linux 4, or later versions, are reported to
953 work) have support for assigning IP information to bonding devices via
956 To configure bonding for DHCP, configure it as described
957 above, except replace the line "BOOTPROTO=none" with "BOOTPROTO=dhcp"
958 and add a line consisting of "TYPE=Bonding". Note that the TYPE value
961 3.2.2 Configuring Multiple Bonds with Initscripts
962 -------------------------------------------------
964 Initscripts packages that are included with Fedora 7 and Red Hat
965 Enterprise Linux 5 support multiple bonding interfaces by simply
966 specifying the appropriate BONDING_OPTS= in ifcfg-bondX where X is the
967 number of the bond. This support requires sysfs support in the kernel,
968 and a bonding driver of version 3.0.0 or later. Other configurations may
969 not support this method for specifying multiple bonding interfaces; for
970 those instances, see the "Configuring Multiple Bonds Manually" section,
973 3.3 Configuring Bonding Manually with Ifenslave
974 -----------------------------------------------
976 This section applies to distros whose network initialization
977 scripts (the sysconfig or initscripts package) do not have specific
978 knowledge of bonding. One such distro is SuSE Linux Enterprise Server
981 The general method for these systems is to place the bonding
982 module parameters into /etc/modules.conf or /etc/modprobe.conf (as
983 appropriate for the installed distro), then add modprobe and/or
984 ifenslave commands to the system's global init script. The name of
985 the global init script differs; for sysconfig, it is
986 /etc/init.d/boot.local and for initscripts it is /etc/rc.d/rc.local.
988 For example, if you wanted to make a simple bond of two e100
989 devices (presumed to be eth0 and eth1), and have it persist across
990 reboots, edit the appropriate file (/etc/init.d/boot.local or
991 /etc/rc.d/rc.local), and add the following:
993 modprobe bonding mode=balance-alb miimon=100
995 ifconfig bond0 192.168.1.1 netmask 255.255.255.0 up
999 Replace the example bonding module parameters and bond0
1000 network configuration (IP address, netmask, etc) with the appropriate
1001 values for your configuration.
1003 Unfortunately, this method will not provide support for the
1004 ifup and ifdown scripts on the bond devices. To reload the bonding
1005 configuration, it is necessary to run the initialization script, e.g.,
1007 # /etc/init.d/boot.local
1011 # /etc/rc.d/rc.local
1013 It may be desirable in such a case to create a separate script
1014 which only initializes the bonding configuration, then call that
1015 separate script from within boot.local. This allows for bonding to be
1016 enabled without re-running the entire global init script.
1018 To shut down the bonding devices, it is necessary to first
1019 mark the bonding device itself as being down, then remove the
1020 appropriate device driver modules. For our example above, you can do
1023 # ifconfig bond0 down
1027 Again, for convenience, it may be desirable to create a script
1028 with these commands.
1031 3.3.1 Configuring Multiple Bonds Manually
1032 -----------------------------------------
1034 This section contains information on configuring multiple
1035 bonding devices with differing options for those systems whose network
1036 initialization scripts lack support for configuring multiple bonds.
1038 If you require multiple bonding devices, but all with the same
1039 options, you may wish to use the "max_bonds" module parameter,
1042 To create multiple bonding devices with differing options, it is
1043 preferrable to use bonding parameters exported by sysfs, documented in the
1046 For versions of bonding without sysfs support, the only means to
1047 provide multiple instances of bonding with differing options is to load
1048 the bonding driver multiple times. Note that current versions of the
1049 sysconfig network initialization scripts handle this automatically; if
1050 your distro uses these scripts, no special action is needed. See the
1051 section Configuring Bonding Devices, above, if you're not sure about your
1052 network initialization scripts.
1054 To load multiple instances of the module, it is necessary to
1055 specify a different name for each instance (the module loading system
1056 requires that every loaded module, even multiple instances of the same
1057 module, have a unique name). This is accomplished by supplying multiple
1058 sets of bonding options in /etc/modprobe.conf, for example:
1061 options bond0 -o bond0 mode=balance-rr miimon=100
1064 options bond1 -o bond1 mode=balance-alb miimon=50
1066 will load the bonding module two times. The first instance is
1067 named "bond0" and creates the bond0 device in balance-rr mode with an
1068 miimon of 100. The second instance is named "bond1" and creates the
1069 bond1 device in balance-alb mode with an miimon of 50.
1071 In some circumstances (typically with older distributions),
1072 the above does not work, and the second bonding instance never sees
1073 its options. In that case, the second options line can be substituted
1076 install bond1 /sbin/modprobe --ignore-install bonding -o bond1 \
1077 mode=balance-alb miimon=50
1079 This may be repeated any number of times, specifying a new and
1080 unique name in place of bond1 for each subsequent instance.
1082 It has been observed that some Red Hat supplied kernels are unable
1083 to rename modules at load time (the "-o bond1" part). Attempts to pass
1084 that option to modprobe will produce an "Operation not permitted" error.
1085 This has been reported on some Fedora Core kernels, and has been seen on
1086 RHEL 4 as well. On kernels exhibiting this problem, it will be impossible
1087 to configure multiple bonds with differing parameters (as they are older
1088 kernels, and also lack sysfs support).
1090 3.4 Configuring Bonding Manually via Sysfs
1091 ------------------------------------------
1093 Starting with version 3.0.0, Channel Bonding may be configured
1094 via the sysfs interface. This interface allows dynamic configuration
1095 of all bonds in the system without unloading the module. It also
1096 allows for adding and removing bonds at runtime. Ifenslave is no
1097 longer required, though it is still supported.
1099 Use of the sysfs interface allows you to use multiple bonds
1100 with different configurations without having to reload the module.
1101 It also allows you to use multiple, differently configured bonds when
1102 bonding is compiled into the kernel.
1104 You must have the sysfs filesystem mounted to configure
1105 bonding this way. The examples in this document assume that you
1106 are using the standard mount point for sysfs, e.g. /sys. If your
1107 sysfs filesystem is mounted elsewhere, you will need to adjust the
1108 example paths accordingly.
1110 Creating and Destroying Bonds
1111 -----------------------------
1112 To add a new bond foo:
1113 # echo +foo > /sys/class/net/bonding_masters
1115 To remove an existing bond bar:
1116 # echo -bar > /sys/class/net/bonding_masters
1118 To show all existing bonds:
1119 # cat /sys/class/net/bonding_masters
1121 NOTE: due to 4K size limitation of sysfs files, this list may be
1122 truncated if you have more than a few hundred bonds. This is unlikely
1123 to occur under normal operating conditions.
1125 Adding and Removing Slaves
1126 --------------------------
1127 Interfaces may be enslaved to a bond using the file
1128 /sys/class/net/<bond>/bonding/slaves. The semantics for this file
1129 are the same as for the bonding_masters file.
1131 To enslave interface eth0 to bond bond0:
1133 # echo +eth0 > /sys/class/net/bond0/bonding/slaves
1135 To free slave eth0 from bond bond0:
1136 # echo -eth0 > /sys/class/net/bond0/bonding/slaves
1138 When an interface is enslaved to a bond, symlinks between the
1139 two are created in the sysfs filesystem. In this case, you would get
1140 /sys/class/net/bond0/slave_eth0 pointing to /sys/class/net/eth0, and
1141 /sys/class/net/eth0/master pointing to /sys/class/net/bond0.
1143 This means that you can tell quickly whether or not an
1144 interface is enslaved by looking for the master symlink. Thus:
1145 # echo -eth0 > /sys/class/net/eth0/master/bonding/slaves
1146 will free eth0 from whatever bond it is enslaved to, regardless of
1147 the name of the bond interface.
1149 Changing a Bond's Configuration
1150 -------------------------------
1151 Each bond may be configured individually by manipulating the
1152 files located in /sys/class/net/<bond name>/bonding
1154 The names of these files correspond directly with the command-
1155 line parameters described elsewhere in this file, and, with the
1156 exception of arp_ip_target, they accept the same values. To see the
1157 current setting, simply cat the appropriate file.
1159 A few examples will be given here; for specific usage
1160 guidelines for each parameter, see the appropriate section in this
1163 To configure bond0 for balance-alb mode:
1164 # ifconfig bond0 down
1165 # echo 6 > /sys/class/net/bond0/bonding/mode
1167 # echo balance-alb > /sys/class/net/bond0/bonding/mode
1168 NOTE: The bond interface must be down before the mode can be
1171 To enable MII monitoring on bond0 with a 1 second interval:
1172 # echo 1000 > /sys/class/net/bond0/bonding/miimon
1173 NOTE: If ARP monitoring is enabled, it will disabled when MII
1174 monitoring is enabled, and vice-versa.
1177 # echo +192.168.0.100 > /sys/class/net/bond0/bonding/arp_ip_target
1178 # echo +192.168.0.101 > /sys/class/net/bond0/bonding/arp_ip_target
1179 NOTE: up to 10 target addresses may be specified.
1181 To remove an ARP target:
1182 # echo -192.168.0.100 > /sys/class/net/bond0/bonding/arp_ip_target
1184 Example Configuration
1185 ---------------------
1186 We begin with the same example that is shown in section 3.3,
1187 executed with sysfs, and without using ifenslave.
1189 To make a simple bond of two e100 devices (presumed to be eth0
1190 and eth1), and have it persist across reboots, edit the appropriate
1191 file (/etc/init.d/boot.local or /etc/rc.d/rc.local), and add the
1196 echo balance-alb > /sys/class/net/bond0/bonding/mode
1197 ifconfig bond0 192.168.1.1 netmask 255.255.255.0 up
1198 echo 100 > /sys/class/net/bond0/bonding/miimon
1199 echo +eth0 > /sys/class/net/bond0/bonding/slaves
1200 echo +eth1 > /sys/class/net/bond0/bonding/slaves
1202 To add a second bond, with two e1000 interfaces in
1203 active-backup mode, using ARP monitoring, add the following lines to
1207 echo +bond1 > /sys/class/net/bonding_masters
1208 echo active-backup > /sys/class/net/bond1/bonding/mode
1209 ifconfig bond1 192.168.2.1 netmask 255.255.255.0 up
1210 echo +192.168.2.100 /sys/class/net/bond1/bonding/arp_ip_target
1211 echo 2000 > /sys/class/net/bond1/bonding/arp_interval
1212 echo +eth2 > /sys/class/net/bond1/bonding/slaves
1213 echo +eth3 > /sys/class/net/bond1/bonding/slaves
1216 4. Querying Bonding Configuration
1217 =================================
1219 4.1 Bonding Configuration
1220 -------------------------
1222 Each bonding device has a read-only file residing in the
1223 /proc/net/bonding directory. The file contents include information
1224 about the bonding configuration, options and state of each slave.
1226 For example, the contents of /proc/net/bonding/bond0 after the
1227 driver is loaded with parameters of mode=0 and miimon=1000 is
1228 generally as follows:
1230 Ethernet Channel Bonding Driver: 2.6.1 (October 29, 2004)
1231 Bonding Mode: load balancing (round-robin)
1232 Currently Active Slave: eth0
1234 MII Polling Interval (ms): 1000
1238 Slave Interface: eth1
1240 Link Failure Count: 1
1242 Slave Interface: eth0
1244 Link Failure Count: 1
1246 The precise format and contents will change depending upon the
1247 bonding configuration, state, and version of the bonding driver.
1249 4.2 Network configuration
1250 -------------------------
1252 The network configuration can be inspected using the ifconfig
1253 command. Bonding devices will have the MASTER flag set; Bonding slave
1254 devices will have the SLAVE flag set. The ifconfig output does not
1255 contain information on which slaves are associated with which masters.
1257 In the example below, the bond0 interface is the master
1258 (MASTER) while eth0 and eth1 are slaves (SLAVE). Notice all slaves of
1259 bond0 have the same MAC address (HWaddr) as bond0 for all modes except
1260 TLB and ALB that require a unique MAC address for each slave.
1263 bond0 Link encap:Ethernet HWaddr 00:C0:F0:1F:37:B4
1264 inet addr:XXX.XXX.XXX.YYY Bcast:XXX.XXX.XXX.255 Mask:255.255.252.0
1265 UP BROADCAST RUNNING MASTER MULTICAST MTU:1500 Metric:1
1266 RX packets:7224794 errors:0 dropped:0 overruns:0 frame:0
1267 TX packets:3286647 errors:1 dropped:0 overruns:1 carrier:0
1268 collisions:0 txqueuelen:0
1270 eth0 Link encap:Ethernet HWaddr 00:C0:F0:1F:37:B4
1271 UP BROADCAST RUNNING SLAVE MULTICAST MTU:1500 Metric:1
1272 RX packets:3573025 errors:0 dropped:0 overruns:0 frame:0
1273 TX packets:1643167 errors:1 dropped:0 overruns:1 carrier:0
1274 collisions:0 txqueuelen:100
1275 Interrupt:10 Base address:0x1080
1277 eth1 Link encap:Ethernet HWaddr 00:C0:F0:1F:37:B4
1278 UP BROADCAST RUNNING SLAVE MULTICAST MTU:1500 Metric:1
1279 RX packets:3651769 errors:0 dropped:0 overruns:0 frame:0
1280 TX packets:1643480 errors:0 dropped:0 overruns:0 carrier:0
1281 collisions:0 txqueuelen:100
1282 Interrupt:9 Base address:0x1400
1284 5. Switch Configuration
1285 =======================
1287 For this section, "switch" refers to whatever system the
1288 bonded devices are directly connected to (i.e., where the other end of
1289 the cable plugs into). This may be an actual dedicated switch device,
1290 or it may be another regular system (e.g., another computer running
1293 The active-backup, balance-tlb and balance-alb modes do not
1294 require any specific configuration of the switch.
1296 The 802.3ad mode requires that the switch have the appropriate
1297 ports configured as an 802.3ad aggregation. The precise method used
1298 to configure this varies from switch to switch, but, for example, a
1299 Cisco 3550 series switch requires that the appropriate ports first be
1300 grouped together in a single etherchannel instance, then that
1301 etherchannel is set to mode "lacp" to enable 802.3ad (instead of
1302 standard EtherChannel).
1304 The balance-rr, balance-xor and broadcast modes generally
1305 require that the switch have the appropriate ports grouped together.
1306 The nomenclature for such a group differs between switches, it may be
1307 called an "etherchannel" (as in the Cisco example, above), a "trunk
1308 group" or some other similar variation. For these modes, each switch
1309 will also have its own configuration options for the switch's transmit
1310 policy to the bond. Typical choices include XOR of either the MAC or
1311 IP addresses. The transmit policy of the two peers does not need to
1312 match. For these three modes, the bonding mode really selects a
1313 transmit policy for an EtherChannel group; all three will interoperate
1314 with another EtherChannel group.
1317 6. 802.1q VLAN Support
1318 ======================
1320 It is possible to configure VLAN devices over a bond interface
1321 using the 8021q driver. However, only packets coming from the 8021q
1322 driver and passing through bonding will be tagged by default. Self
1323 generated packets, for example, bonding's learning packets or ARP
1324 packets generated by either ALB mode or the ARP monitor mechanism, are
1325 tagged internally by bonding itself. As a result, bonding must
1326 "learn" the VLAN IDs configured above it, and use those IDs to tag
1327 self generated packets.
1329 For reasons of simplicity, and to support the use of adapters
1330 that can do VLAN hardware acceleration offloading, the bonding
1331 interface declares itself as fully hardware offloading capable, it gets
1332 the add_vid/kill_vid notifications to gather the necessary
1333 information, and it propagates those actions to the slaves. In case
1334 of mixed adapter types, hardware accelerated tagged packets that
1335 should go through an adapter that is not offloading capable are
1336 "un-accelerated" by the bonding driver so the VLAN tag sits in the
1339 VLAN interfaces *must* be added on top of a bonding interface
1340 only after enslaving at least one slave. The bonding interface has a
1341 hardware address of 00:00:00:00:00:00 until the first slave is added.
1342 If the VLAN interface is created prior to the first enslavement, it
1343 would pick up the all-zeroes hardware address. Once the first slave
1344 is attached to the bond, the bond device itself will pick up the
1345 slave's hardware address, which is then available for the VLAN device.
1347 Also, be aware that a similar problem can occur if all slaves
1348 are released from a bond that still has one or more VLAN interfaces on
1349 top of it. When a new slave is added, the bonding interface will
1350 obtain its hardware address from the first slave, which might not
1351 match the hardware address of the VLAN interfaces (which was
1352 ultimately copied from an earlier slave).
1354 There are two methods to insure that the VLAN device operates
1355 with the correct hardware address if all slaves are removed from a
1358 1. Remove all VLAN interfaces then recreate them
1360 2. Set the bonding interface's hardware address so that it
1361 matches the hardware address of the VLAN interfaces.
1363 Note that changing a VLAN interface's HW address would set the
1364 underlying device -- i.e. the bonding interface -- to promiscuous
1365 mode, which might not be what you want.
1371 The bonding driver at present supports two schemes for
1372 monitoring a slave device's link state: the ARP monitor and the MII
1375 At the present time, due to implementation restrictions in the
1376 bonding driver itself, it is not possible to enable both ARP and MII
1377 monitoring simultaneously.
1379 7.1 ARP Monitor Operation
1380 -------------------------
1382 The ARP monitor operates as its name suggests: it sends ARP
1383 queries to one or more designated peer systems on the network, and
1384 uses the response as an indication that the link is operating. This
1385 gives some assurance that traffic is actually flowing to and from one
1386 or more peers on the local network.
1388 The ARP monitor relies on the device driver itself to verify
1389 that traffic is flowing. In particular, the driver must keep up to
1390 date the last receive time, dev->last_rx, and transmit start time,
1391 dev->trans_start. If these are not updated by the driver, then the
1392 ARP monitor will immediately fail any slaves using that driver, and
1393 those slaves will stay down. If networking monitoring (tcpdump, etc)
1394 shows the ARP requests and replies on the network, then it may be that
1395 your device driver is not updating last_rx and trans_start.
1397 7.2 Configuring Multiple ARP Targets
1398 ------------------------------------
1400 While ARP monitoring can be done with just one target, it can
1401 be useful in a High Availability setup to have several targets to
1402 monitor. In the case of just one target, the target itself may go
1403 down or have a problem making it unresponsive to ARP requests. Having
1404 an additional target (or several) increases the reliability of the ARP
1407 Multiple ARP targets must be separated by commas as follows:
1409 # example options for ARP monitoring with three targets
1411 options bond0 arp_interval=60 arp_ip_target=192.168.0.1,192.168.0.3,192.168.0.9
1413 For just a single target the options would resemble:
1415 # example options for ARP monitoring with one target
1417 options bond0 arp_interval=60 arp_ip_target=192.168.0.100
1420 7.3 MII Monitor Operation
1421 -------------------------
1423 The MII monitor monitors only the carrier state of the local
1424 network interface. It accomplishes this in one of three ways: by
1425 depending upon the device driver to maintain its carrier state, by
1426 querying the device's MII registers, or by making an ethtool query to
1429 If the use_carrier module parameter is 1 (the default value),
1430 then the MII monitor will rely on the driver for carrier state
1431 information (via the netif_carrier subsystem). As explained in the
1432 use_carrier parameter information, above, if the MII monitor fails to
1433 detect carrier loss on the device (e.g., when the cable is physically
1434 disconnected), it may be that the driver does not support
1437 If use_carrier is 0, then the MII monitor will first query the
1438 device's (via ioctl) MII registers and check the link state. If that
1439 request fails (not just that it returns carrier down), then the MII
1440 monitor will make an ethtool ETHOOL_GLINK request to attempt to obtain
1441 the same information. If both methods fail (i.e., the driver either
1442 does not support or had some error in processing both the MII register
1443 and ethtool requests), then the MII monitor will assume the link is
1446 8. Potential Sources of Trouble
1447 ===============================
1449 8.1 Adventures in Routing
1450 -------------------------
1452 When bonding is configured, it is important that the slave
1453 devices not have routes that supersede routes of the master (or,
1454 generally, not have routes at all). For example, suppose the bonding
1455 device bond0 has two slaves, eth0 and eth1, and the routing table is
1458 Kernel IP routing table
1459 Destination Gateway Genmask Flags MSS Window irtt Iface
1460 10.0.0.0 0.0.0.0 255.255.0.0 U 40 0 0 eth0
1461 10.0.0.0 0.0.0.0 255.255.0.0 U 40 0 0 eth1
1462 10.0.0.0 0.0.0.0 255.255.0.0 U 40 0 0 bond0
1463 127.0.0.0 0.0.0.0 255.0.0.0 U 40 0 0 lo
1465 This routing configuration will likely still update the
1466 receive/transmit times in the driver (needed by the ARP monitor), but
1467 may bypass the bonding driver (because outgoing traffic to, in this
1468 case, another host on network 10 would use eth0 or eth1 before bond0).
1470 The ARP monitor (and ARP itself) may become confused by this
1471 configuration, because ARP requests (generated by the ARP monitor)
1472 will be sent on one interface (bond0), but the corresponding reply
1473 will arrive on a different interface (eth0). This reply looks to ARP
1474 as an unsolicited ARP reply (because ARP matches replies on an
1475 interface basis), and is discarded. The MII monitor is not affected
1476 by the state of the routing table.
1478 The solution here is simply to insure that slaves do not have
1479 routes of their own, and if for some reason they must, those routes do
1480 not supersede routes of their master. This should generally be the
1481 case, but unusual configurations or errant manual or automatic static
1482 route additions may cause trouble.
1484 8.2 Ethernet Device Renaming
1485 ----------------------------
1487 On systems with network configuration scripts that do not
1488 associate physical devices directly with network interface names (so
1489 that the same physical device always has the same "ethX" name), it may
1490 be necessary to add some special logic to either /etc/modules.conf or
1491 /etc/modprobe.conf (depending upon which is installed on the system).
1493 For example, given a modules.conf containing the following:
1496 options bond0 mode=some-mode miimon=50
1502 If neither eth0 and eth1 are slaves to bond0, then when the
1503 bond0 interface comes up, the devices may end up reordered. This
1504 happens because bonding is loaded first, then its slave device's
1505 drivers are loaded next. Since no other drivers have been loaded,
1506 when the e1000 driver loads, it will receive eth0 and eth1 for its
1507 devices, but the bonding configuration tries to enslave eth2 and eth3
1508 (which may later be assigned to the tg3 devices).
1510 Adding the following:
1512 add above bonding e1000 tg3
1514 causes modprobe to load e1000 then tg3, in that order, when
1515 bonding is loaded. This command is fully documented in the
1516 modules.conf manual page.
1518 On systems utilizing modprobe.conf (or modprobe.conf.local),
1519 an equivalent problem can occur. In this case, the following can be
1520 added to modprobe.conf (or modprobe.conf.local, as appropriate), as
1521 follows (all on one line; it has been split here for clarity):
1523 install bonding /sbin/modprobe tg3; /sbin/modprobe e1000;
1524 /sbin/modprobe --ignore-install bonding
1526 This will, when loading the bonding module, rather than
1527 performing the normal action, instead execute the provided command.
1528 This command loads the device drivers in the order needed, then calls
1529 modprobe with --ignore-install to cause the normal action to then take
1530 place. Full documentation on this can be found in the modprobe.conf
1531 and modprobe manual pages.
1533 8.3. Painfully Slow Or No Failed Link Detection By Miimon
1534 ---------------------------------------------------------
1536 By default, bonding enables the use_carrier option, which
1537 instructs bonding to trust the driver to maintain carrier state.
1539 As discussed in the options section, above, some drivers do
1540 not support the netif_carrier_on/_off link state tracking system.
1541 With use_carrier enabled, bonding will always see these links as up,
1542 regardless of their actual state.
1544 Additionally, other drivers do support netif_carrier, but do
1545 not maintain it in real time, e.g., only polling the link state at
1546 some fixed interval. In this case, miimon will detect failures, but
1547 only after some long period of time has expired. If it appears that
1548 miimon is very slow in detecting link failures, try specifying
1549 use_carrier=0 to see if that improves the failure detection time. If
1550 it does, then it may be that the driver checks the carrier state at a
1551 fixed interval, but does not cache the MII register values (so the
1552 use_carrier=0 method of querying the registers directly works). If
1553 use_carrier=0 does not improve the failover, then the driver may cache
1554 the registers, or the problem may be elsewhere.
1556 Also, remember that miimon only checks for the device's
1557 carrier state. It has no way to determine the state of devices on or
1558 beyond other ports of a switch, or if a switch is refusing to pass
1559 traffic while still maintaining carrier on.
1564 If running SNMP agents, the bonding driver should be loaded
1565 before any network drivers participating in a bond. This requirement
1566 is due to the interface index (ipAdEntIfIndex) being associated to
1567 the first interface found with a given IP address. That is, there is
1568 only one ipAdEntIfIndex for each IP address. For example, if eth0 and
1569 eth1 are slaves of bond0 and the driver for eth0 is loaded before the
1570 bonding driver, the interface for the IP address will be associated
1571 with the eth0 interface. This configuration is shown below, the IP
1572 address 192.168.1.1 has an interface index of 2 which indexes to eth0
1573 in the ifDescr table (ifDescr.2).
1575 interfaces.ifTable.ifEntry.ifDescr.1 = lo
1576 interfaces.ifTable.ifEntry.ifDescr.2 = eth0
1577 interfaces.ifTable.ifEntry.ifDescr.3 = eth1
1578 interfaces.ifTable.ifEntry.ifDescr.4 = eth2
1579 interfaces.ifTable.ifEntry.ifDescr.5 = eth3
1580 interfaces.ifTable.ifEntry.ifDescr.6 = bond0
1581 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.10.10.10 = 5
1582 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.192.168.1.1 = 2
1583 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.74.20.94 = 4
1584 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.127.0.0.1 = 1
1586 This problem is avoided by loading the bonding driver before
1587 any network drivers participating in a bond. Below is an example of
1588 loading the bonding driver first, the IP address 192.168.1.1 is
1589 correctly associated with ifDescr.2.
1591 interfaces.ifTable.ifEntry.ifDescr.1 = lo
1592 interfaces.ifTable.ifEntry.ifDescr.2 = bond0
1593 interfaces.ifTable.ifEntry.ifDescr.3 = eth0
1594 interfaces.ifTable.ifEntry.ifDescr.4 = eth1
1595 interfaces.ifTable.ifEntry.ifDescr.5 = eth2
1596 interfaces.ifTable.ifEntry.ifDescr.6 = eth3
1597 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.10.10.10 = 6
1598 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.192.168.1.1 = 2
1599 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.74.20.94 = 5
1600 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.127.0.0.1 = 1
1602 While some distributions may not report the interface name in
1603 ifDescr, the association between the IP address and IfIndex remains
1604 and SNMP functions such as Interface_Scan_Next will report that
1607 10. Promiscuous mode
1608 ====================
1610 When running network monitoring tools, e.g., tcpdump, it is
1611 common to enable promiscuous mode on the device, so that all traffic
1612 is seen (instead of seeing only traffic destined for the local host).
1613 The bonding driver handles promiscuous mode changes to the bonding
1614 master device (e.g., bond0), and propagates the setting to the slave
1617 For the balance-rr, balance-xor, broadcast, and 802.3ad modes,
1618 the promiscuous mode setting is propagated to all slaves.
1620 For the active-backup, balance-tlb and balance-alb modes, the
1621 promiscuous mode setting is propagated only to the active slave.
1623 For balance-tlb mode, the active slave is the slave currently
1624 receiving inbound traffic.
1626 For balance-alb mode, the active slave is the slave used as a
1627 "primary." This slave is used for mode-specific control traffic, for
1628 sending to peers that are unassigned or if the load is unbalanced.
1630 For the active-backup, balance-tlb and balance-alb modes, when
1631 the active slave changes (e.g., due to a link failure), the
1632 promiscuous setting will be propagated to the new active slave.
1634 11. Configuring Bonding for High Availability
1635 =============================================
1637 High Availability refers to configurations that provide
1638 maximum network availability by having redundant or backup devices,
1639 links or switches between the host and the rest of the world. The
1640 goal is to provide the maximum availability of network connectivity
1641 (i.e., the network always works), even though other configurations
1642 could provide higher throughput.
1644 11.1 High Availability in a Single Switch Topology
1645 --------------------------------------------------
1647 If two hosts (or a host and a single switch) are directly
1648 connected via multiple physical links, then there is no availability
1649 penalty to optimizing for maximum bandwidth. In this case, there is
1650 only one switch (or peer), so if it fails, there is no alternative
1651 access to fail over to. Additionally, the bonding load balance modes
1652 support link monitoring of their members, so if individual links fail,
1653 the load will be rebalanced across the remaining devices.
1655 See Section 13, "Configuring Bonding for Maximum Throughput"
1656 for information on configuring bonding with one peer device.
1658 11.2 High Availability in a Multiple Switch Topology
1659 ----------------------------------------------------
1661 With multiple switches, the configuration of bonding and the
1662 network changes dramatically. In multiple switch topologies, there is
1663 a trade off between network availability and usable bandwidth.
1665 Below is a sample network, configured to maximize the
1666 availability of the network:
1670 +-----+----+ +-----+----+
1671 | |port2 ISL port2| |
1672 | switch A +--------------------------+ switch B |
1674 +-----+----+ +-----++---+
1677 +-------------+ host1 +---------------+
1680 In this configuration, there is a link between the two
1681 switches (ISL, or inter switch link), and multiple ports connecting to
1682 the outside world ("port3" on each switch). There is no technical
1683 reason that this could not be extended to a third switch.
1685 11.2.1 HA Bonding Mode Selection for Multiple Switch Topology
1686 -------------------------------------------------------------
1688 In a topology such as the example above, the active-backup and
1689 broadcast modes are the only useful bonding modes when optimizing for
1690 availability; the other modes require all links to terminate on the
1691 same peer for them to behave rationally.
1693 active-backup: This is generally the preferred mode, particularly if
1694 the switches have an ISL and play together well. If the
1695 network configuration is such that one switch is specifically
1696 a backup switch (e.g., has lower capacity, higher cost, etc),
1697 then the primary option can be used to insure that the
1698 preferred link is always used when it is available.
1700 broadcast: This mode is really a special purpose mode, and is suitable
1701 only for very specific needs. For example, if the two
1702 switches are not connected (no ISL), and the networks beyond
1703 them are totally independent. In this case, if it is
1704 necessary for some specific one-way traffic to reach both
1705 independent networks, then the broadcast mode may be suitable.
1707 11.2.2 HA Link Monitoring Selection for Multiple Switch Topology
1708 ----------------------------------------------------------------
1710 The choice of link monitoring ultimately depends upon your
1711 switch. If the switch can reliably fail ports in response to other
1712 failures, then either the MII or ARP monitors should work. For
1713 example, in the above example, if the "port3" link fails at the remote
1714 end, the MII monitor has no direct means to detect this. The ARP
1715 monitor could be configured with a target at the remote end of port3,
1716 thus detecting that failure without switch support.
1718 In general, however, in a multiple switch topology, the ARP
1719 monitor can provide a higher level of reliability in detecting end to
1720 end connectivity failures (which may be caused by the failure of any
1721 individual component to pass traffic for any reason). Additionally,
1722 the ARP monitor should be configured with multiple targets (at least
1723 one for each switch in the network). This will insure that,
1724 regardless of which switch is active, the ARP monitor has a suitable
1727 Note, also, that of late many switches now support a functionality
1728 generally referred to as "trunk failover." This is a feature of the
1729 switch that causes the link state of a particular switch port to be set
1730 down (or up) when the state of another switch port goes down (or up).
1731 It's purpose is to propogate link failures from logically "exterior" ports
1732 to the logically "interior" ports that bonding is able to monitor via
1733 miimon. Availability and configuration for trunk failover varies by
1734 switch, but this can be a viable alternative to the ARP monitor when using
1737 12. Configuring Bonding for Maximum Throughput
1738 ==============================================
1740 12.1 Maximizing Throughput in a Single Switch Topology
1741 ------------------------------------------------------
1743 In a single switch configuration, the best method to maximize
1744 throughput depends upon the application and network environment. The
1745 various load balancing modes each have strengths and weaknesses in
1746 different environments, as detailed below.
1748 For this discussion, we will break down the topologies into
1749 two categories. Depending upon the destination of most traffic, we
1750 categorize them into either "gatewayed" or "local" configurations.
1752 In a gatewayed configuration, the "switch" is acting primarily
1753 as a router, and the majority of traffic passes through this router to
1754 other networks. An example would be the following:
1757 +----------+ +----------+
1758 | |eth0 port1| | to other networks
1759 | Host A +---------------------+ router +------------------->
1760 | +---------------------+ | Hosts B and C are out
1761 | |eth1 port2| | here somewhere
1762 +----------+ +----------+
1764 The router may be a dedicated router device, or another host
1765 acting as a gateway. For our discussion, the important point is that
1766 the majority of traffic from Host A will pass through the router to
1767 some other network before reaching its final destination.
1769 In a gatewayed network configuration, although Host A may
1770 communicate with many other systems, all of its traffic will be sent
1771 and received via one other peer on the local network, the router.
1773 Note that the case of two systems connected directly via
1774 multiple physical links is, for purposes of configuring bonding, the
1775 same as a gatewayed configuration. In that case, it happens that all
1776 traffic is destined for the "gateway" itself, not some other network
1779 In a local configuration, the "switch" is acting primarily as
1780 a switch, and the majority of traffic passes through this switch to
1781 reach other stations on the same network. An example would be the
1784 +----------+ +----------+ +--------+
1785 | |eth0 port1| +-------+ Host B |
1786 | Host A +------------+ switch |port3 +--------+
1787 | +------------+ | +--------+
1788 | |eth1 port2| +------------------+ Host C |
1789 +----------+ +----------+port4 +--------+
1792 Again, the switch may be a dedicated switch device, or another
1793 host acting as a gateway. For our discussion, the important point is
1794 that the majority of traffic from Host A is destined for other hosts
1795 on the same local network (Hosts B and C in the above example).
1797 In summary, in a gatewayed configuration, traffic to and from
1798 the bonded device will be to the same MAC level peer on the network
1799 (the gateway itself, i.e., the router), regardless of its final
1800 destination. In a local configuration, traffic flows directly to and
1801 from the final destinations, thus, each destination (Host B, Host C)
1802 will be addressed directly by their individual MAC addresses.
1804 This distinction between a gatewayed and a local network
1805 configuration is important because many of the load balancing modes
1806 available use the MAC addresses of the local network source and
1807 destination to make load balancing decisions. The behavior of each
1808 mode is described below.
1811 12.1.1 MT Bonding Mode Selection for Single Switch Topology
1812 -----------------------------------------------------------
1814 This configuration is the easiest to set up and to understand,
1815 although you will have to decide which bonding mode best suits your
1816 needs. The trade offs for each mode are detailed below:
1818 balance-rr: This mode is the only mode that will permit a single
1819 TCP/IP connection to stripe traffic across multiple
1820 interfaces. It is therefore the only mode that will allow a
1821 single TCP/IP stream to utilize more than one interface's
1822 worth of throughput. This comes at a cost, however: the
1823 striping generally results in peer systems receiving packets out
1824 of order, causing TCP/IP's congestion control system to kick
1825 in, often by retransmitting segments.
1827 It is possible to adjust TCP/IP's congestion limits by
1828 altering the net.ipv4.tcp_reordering sysctl parameter. The
1829 usual default value is 3, and the maximum useful value is 127.
1830 For a four interface balance-rr bond, expect that a single
1831 TCP/IP stream will utilize no more than approximately 2.3
1832 interface's worth of throughput, even after adjusting
1835 Note that the fraction of packets that will be delivered out of
1836 order is highly variable, and is unlikely to be zero. The level
1837 of reordering depends upon a variety of factors, including the
1838 networking interfaces, the switch, and the topology of the
1839 configuration. Speaking in general terms, higher speed network
1840 cards produce more reordering (due to factors such as packet
1841 coalescing), and a "many to many" topology will reorder at a
1842 higher rate than a "many slow to one fast" configuration.
1844 Many switches do not support any modes that stripe traffic
1845 (instead choosing a port based upon IP or MAC level addresses);
1846 for those devices, traffic for a particular connection flowing
1847 through the switch to a balance-rr bond will not utilize greater
1848 than one interface's worth of bandwidth.
1850 If you are utilizing protocols other than TCP/IP, UDP for
1851 example, and your application can tolerate out of order
1852 delivery, then this mode can allow for single stream datagram
1853 performance that scales near linearly as interfaces are added
1856 This mode requires the switch to have the appropriate ports
1857 configured for "etherchannel" or "trunking."
1859 active-backup: There is not much advantage in this network topology to
1860 the active-backup mode, as the inactive backup devices are all
1861 connected to the same peer as the primary. In this case, a
1862 load balancing mode (with link monitoring) will provide the
1863 same level of network availability, but with increased
1864 available bandwidth. On the plus side, active-backup mode
1865 does not require any configuration of the switch, so it may
1866 have value if the hardware available does not support any of
1867 the load balance modes.
1869 balance-xor: This mode will limit traffic such that packets destined
1870 for specific peers will always be sent over the same
1871 interface. Since the destination is determined by the MAC
1872 addresses involved, this mode works best in a "local" network
1873 configuration (as described above), with destinations all on
1874 the same local network. This mode is likely to be suboptimal
1875 if all your traffic is passed through a single router (i.e., a
1876 "gatewayed" network configuration, as described above).
1878 As with balance-rr, the switch ports need to be configured for
1879 "etherchannel" or "trunking."
1881 broadcast: Like active-backup, there is not much advantage to this
1882 mode in this type of network topology.
1884 802.3ad: This mode can be a good choice for this type of network
1885 topology. The 802.3ad mode is an IEEE standard, so all peers
1886 that implement 802.3ad should interoperate well. The 802.3ad
1887 protocol includes automatic configuration of the aggregates,
1888 so minimal manual configuration of the switch is needed
1889 (typically only to designate that some set of devices is
1890 available for 802.3ad). The 802.3ad standard also mandates
1891 that frames be delivered in order (within certain limits), so
1892 in general single connections will not see misordering of
1893 packets. The 802.3ad mode does have some drawbacks: the
1894 standard mandates that all devices in the aggregate operate at
1895 the same speed and duplex. Also, as with all bonding load
1896 balance modes other than balance-rr, no single connection will
1897 be able to utilize more than a single interface's worth of
1900 Additionally, the linux bonding 802.3ad implementation
1901 distributes traffic by peer (using an XOR of MAC addresses),
1902 so in a "gatewayed" configuration, all outgoing traffic will
1903 generally use the same device. Incoming traffic may also end
1904 up on a single device, but that is dependent upon the
1905 balancing policy of the peer's 8023.ad implementation. In a
1906 "local" configuration, traffic will be distributed across the
1907 devices in the bond.
1909 Finally, the 802.3ad mode mandates the use of the MII monitor,
1910 therefore, the ARP monitor is not available in this mode.
1912 balance-tlb: The balance-tlb mode balances outgoing traffic by peer.
1913 Since the balancing is done according to MAC address, in a
1914 "gatewayed" configuration (as described above), this mode will
1915 send all traffic across a single device. However, in a
1916 "local" network configuration, this mode balances multiple
1917 local network peers across devices in a vaguely intelligent
1918 manner (not a simple XOR as in balance-xor or 802.3ad mode),
1919 so that mathematically unlucky MAC addresses (i.e., ones that
1920 XOR to the same value) will not all "bunch up" on a single
1923 Unlike 802.3ad, interfaces may be of differing speeds, and no
1924 special switch configuration is required. On the down side,
1925 in this mode all incoming traffic arrives over a single
1926 interface, this mode requires certain ethtool support in the
1927 network device driver of the slave interfaces, and the ARP
1928 monitor is not available.
1930 balance-alb: This mode is everything that balance-tlb is, and more.
1931 It has all of the features (and restrictions) of balance-tlb,
1932 and will also balance incoming traffic from local network
1933 peers (as described in the Bonding Module Options section,
1936 The only additional down side to this mode is that the network
1937 device driver must support changing the hardware address while
1940 12.1.2 MT Link Monitoring for Single Switch Topology
1941 ----------------------------------------------------
1943 The choice of link monitoring may largely depend upon which
1944 mode you choose to use. The more advanced load balancing modes do not
1945 support the use of the ARP monitor, and are thus restricted to using
1946 the MII monitor (which does not provide as high a level of end to end
1947 assurance as the ARP monitor).
1949 12.2 Maximum Throughput in a Multiple Switch Topology
1950 -----------------------------------------------------
1952 Multiple switches may be utilized to optimize for throughput
1953 when they are configured in parallel as part of an isolated network
1954 between two or more systems, for example:
1960 +--------+ | +---------+
1962 +------+---+ +-----+----+ +-----+----+
1963 | Switch A | | Switch B | | Switch C |
1964 +------+---+ +-----+----+ +-----+----+
1966 +--------+ | +---------+
1972 In this configuration, the switches are isolated from one
1973 another. One reason to employ a topology such as this is for an
1974 isolated network with many hosts (a cluster configured for high
1975 performance, for example), using multiple smaller switches can be more
1976 cost effective than a single larger switch, e.g., on a network with 24
1977 hosts, three 24 port switches can be significantly less expensive than
1978 a single 72 port switch.
1980 If access beyond the network is required, an individual host
1981 can be equipped with an additional network device connected to an
1982 external network; this host then additionally acts as a gateway.
1984 12.2.1 MT Bonding Mode Selection for Multiple Switch Topology
1985 -------------------------------------------------------------
1987 In actual practice, the bonding mode typically employed in
1988 configurations of this type is balance-rr. Historically, in this
1989 network configuration, the usual caveats about out of order packet
1990 delivery are mitigated by the use of network adapters that do not do
1991 any kind of packet coalescing (via the use of NAPI, or because the
1992 device itself does not generate interrupts until some number of
1993 packets has arrived). When employed in this fashion, the balance-rr
1994 mode allows individual connections between two hosts to effectively
1995 utilize greater than one interface's bandwidth.
1997 12.2.2 MT Link Monitoring for Multiple Switch Topology
1998 ------------------------------------------------------
2000 Again, in actual practice, the MII monitor is most often used
2001 in this configuration, as performance is given preference over
2002 availability. The ARP monitor will function in this topology, but its
2003 advantages over the MII monitor are mitigated by the volume of probes
2004 needed as the number of systems involved grows (remember that each
2005 host in the network is configured with bonding).
2007 13. Switch Behavior Issues
2008 ==========================
2010 13.1 Link Establishment and Failover Delays
2011 -------------------------------------------
2013 Some switches exhibit undesirable behavior with regard to the
2014 timing of link up and down reporting by the switch.
2016 First, when a link comes up, some switches may indicate that
2017 the link is up (carrier available), but not pass traffic over the
2018 interface for some period of time. This delay is typically due to
2019 some type of autonegotiation or routing protocol, but may also occur
2020 during switch initialization (e.g., during recovery after a switch
2021 failure). If you find this to be a problem, specify an appropriate
2022 value to the updelay bonding module option to delay the use of the
2023 relevant interface(s).
2025 Second, some switches may "bounce" the link state one or more
2026 times while a link is changing state. This occurs most commonly while
2027 the switch is initializing. Again, an appropriate updelay value may
2030 Note that when a bonding interface has no active links, the
2031 driver will immediately reuse the first link that goes up, even if the
2032 updelay parameter has been specified (the updelay is ignored in this
2033 case). If there are slave interfaces waiting for the updelay timeout
2034 to expire, the interface that first went into that state will be
2035 immediately reused. This reduces down time of the network if the
2036 value of updelay has been overestimated, and since this occurs only in
2037 cases with no connectivity, there is no additional penalty for
2038 ignoring the updelay.
2040 In addition to the concerns about switch timings, if your
2041 switches take a long time to go into backup mode, it may be desirable
2042 to not activate a backup interface immediately after a link goes down.
2043 Failover may be delayed via the downdelay bonding module option.
2045 13.2 Duplicated Incoming Packets
2046 --------------------------------
2048 NOTE: Starting with version 3.0.2, the bonding driver has logic to
2049 suppress duplicate packets, which should largely eliminate this problem.
2050 The following description is kept for reference.
2052 It is not uncommon to observe a short burst of duplicated
2053 traffic when the bonding device is first used, or after it has been
2054 idle for some period of time. This is most easily observed by issuing
2055 a "ping" to some other host on the network, and noticing that the
2056 output from ping flags duplicates (typically one per slave).
2058 For example, on a bond in active-backup mode with five slaves
2059 all connected to one switch, the output may appear as follows:
2062 PING 10.0.4.2 (10.0.4.2) from 10.0.3.10 : 56(84) bytes of data.
2063 64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.7 ms
2064 64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
2065 64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
2066 64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
2067 64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
2068 64 bytes from 10.0.4.2: icmp_seq=2 ttl=64 time=0.216 ms
2069 64 bytes from 10.0.4.2: icmp_seq=3 ttl=64 time=0.267 ms
2070 64 bytes from 10.0.4.2: icmp_seq=4 ttl=64 time=0.222 ms
2072 This is not due to an error in the bonding driver, rather, it
2073 is a side effect of how many switches update their MAC forwarding
2074 tables. Initially, the switch does not associate the MAC address in
2075 the packet with a particular switch port, and so it may send the
2076 traffic to all ports until its MAC forwarding table is updated. Since
2077 the interfaces attached to the bond may occupy multiple ports on a
2078 single switch, when the switch (temporarily) floods the traffic to all
2079 ports, the bond device receives multiple copies of the same packet
2080 (one per slave device).
2082 The duplicated packet behavior is switch dependent, some
2083 switches exhibit this, and some do not. On switches that display this
2084 behavior, it can be induced by clearing the MAC forwarding table (on
2085 most Cisco switches, the privileged command "clear mac address-table
2086 dynamic" will accomplish this).
2088 14. Hardware Specific Considerations
2089 ====================================
2091 This section contains additional information for configuring
2092 bonding on specific hardware platforms, or for interfacing bonding
2093 with particular switches or other devices.
2095 14.1 IBM BladeCenter
2096 --------------------
2098 This applies to the JS20 and similar systems.
2100 On the JS20 blades, the bonding driver supports only
2101 balance-rr, active-backup, balance-tlb and balance-alb modes. This is
2102 largely due to the network topology inside the BladeCenter, detailed
2105 JS20 network adapter information
2106 --------------------------------
2108 All JS20s come with two Broadcom Gigabit Ethernet ports
2109 integrated on the planar (that's "motherboard" in IBM-speak). In the
2110 BladeCenter chassis, the eth0 port of all JS20 blades is hard wired to
2111 I/O Module #1; similarly, all eth1 ports are wired to I/O Module #2.
2112 An add-on Broadcom daughter card can be installed on a JS20 to provide
2113 two more Gigabit Ethernet ports. These ports, eth2 and eth3, are
2114 wired to I/O Modules 3 and 4, respectively.
2116 Each I/O Module may contain either a switch or a passthrough
2117 module (which allows ports to be directly connected to an external
2118 switch). Some bonding modes require a specific BladeCenter internal
2119 network topology in order to function; these are detailed below.
2121 Additional BladeCenter-specific networking information can be
2122 found in two IBM Redbooks (www.ibm.com/redbooks):
2124 "IBM eServer BladeCenter Networking Options"
2125 "IBM eServer BladeCenter Layer 2-7 Network Switching"
2127 BladeCenter networking configuration
2128 ------------------------------------
2130 Because a BladeCenter can be configured in a very large number
2131 of ways, this discussion will be confined to describing basic
2134 Normally, Ethernet Switch Modules (ESMs) are used in I/O
2135 modules 1 and 2. In this configuration, the eth0 and eth1 ports of a
2136 JS20 will be connected to different internal switches (in the
2137 respective I/O modules).
2139 A passthrough module (OPM or CPM, optical or copper,
2140 passthrough module) connects the I/O module directly to an external
2141 switch. By using PMs in I/O module #1 and #2, the eth0 and eth1
2142 interfaces of a JS20 can be redirected to the outside world and
2143 connected to a common external switch.
2145 Depending upon the mix of ESMs and PMs, the network will
2146 appear to bonding as either a single switch topology (all PMs) or as a
2147 multiple switch topology (one or more ESMs, zero or more PMs). It is
2148 also possible to connect ESMs together, resulting in a configuration
2149 much like the example in "High Availability in a Multiple Switch
2152 Requirements for specific modes
2153 -------------------------------
2155 The balance-rr mode requires the use of passthrough modules
2156 for devices in the bond, all connected to an common external switch.
2157 That switch must be configured for "etherchannel" or "trunking" on the
2158 appropriate ports, as is usual for balance-rr.
2160 The balance-alb and balance-tlb modes will function with
2161 either switch modules or passthrough modules (or a mix). The only
2162 specific requirement for these modes is that all network interfaces
2163 must be able to reach all destinations for traffic sent over the
2164 bonding device (i.e., the network must converge at some point outside
2167 The active-backup mode has no additional requirements.
2169 Link monitoring issues
2170 ----------------------
2172 When an Ethernet Switch Module is in place, only the ARP
2173 monitor will reliably detect link loss to an external switch. This is
2174 nothing unusual, but examination of the BladeCenter cabinet would
2175 suggest that the "external" network ports are the ethernet ports for
2176 the system, when it fact there is a switch between these "external"
2177 ports and the devices on the JS20 system itself. The MII monitor is
2178 only able to detect link failures between the ESM and the JS20 system.
2180 When a passthrough module is in place, the MII monitor does
2181 detect failures to the "external" port, which is then directly
2182 connected to the JS20 system.
2187 The Serial Over LAN (SoL) link is established over the primary
2188 ethernet (eth0) only, therefore, any loss of link to eth0 will result
2189 in losing your SoL connection. It will not fail over with other
2190 network traffic, as the SoL system is beyond the control of the
2193 It may be desirable to disable spanning tree on the switch
2194 (either the internal Ethernet Switch Module, or an external switch) to
2195 avoid fail-over delay issues when using bonding.
2198 15. Frequently Asked Questions
2199 ==============================
2203 Yes. The old 2.0.xx channel bonding patch was not SMP safe.
2204 The new driver was designed to be SMP safe from the start.
2206 2. What type of cards will work with it?
2208 Any Ethernet type cards (you can even mix cards - a Intel
2209 EtherExpress PRO/100 and a 3com 3c905b, for example). For most modes,
2210 devices need not be of the same speed.
2212 Starting with version 3.2.1, bonding also supports Infiniband
2213 slaves in active-backup mode.
2215 3. How many bonding devices can I have?
2219 4. How many slaves can a bonding device have?
2221 This is limited only by the number of network interfaces Linux
2222 supports and/or the number of network cards you can place in your
2225 5. What happens when a slave link dies?
2227 If link monitoring is enabled, then the failing device will be
2228 disabled. The active-backup mode will fail over to a backup link, and
2229 other modes will ignore the failed link. The link will continue to be
2230 monitored, and should it recover, it will rejoin the bond (in whatever
2231 manner is appropriate for the mode). See the sections on High
2232 Availability and the documentation for each mode for additional
2235 Link monitoring can be enabled via either the miimon or
2236 arp_interval parameters (described in the module parameters section,
2237 above). In general, miimon monitors the carrier state as sensed by
2238 the underlying network device, and the arp monitor (arp_interval)
2239 monitors connectivity to another host on the local network.
2241 If no link monitoring is configured, the bonding driver will
2242 be unable to detect link failures, and will assume that all links are
2243 always available. This will likely result in lost packets, and a
2244 resulting degradation of performance. The precise performance loss
2245 depends upon the bonding mode and network configuration.
2247 6. Can bonding be used for High Availability?
2249 Yes. See the section on High Availability for details.
2251 7. Which switches/systems does it work with?
2253 The full answer to this depends upon the desired mode.
2255 In the basic balance modes (balance-rr and balance-xor), it
2256 works with any system that supports etherchannel (also called
2257 trunking). Most managed switches currently available have such
2258 support, and many unmanaged switches as well.
2260 The advanced balance modes (balance-tlb and balance-alb) do
2261 not have special switch requirements, but do need device drivers that
2262 support specific features (described in the appropriate section under
2263 module parameters, above).
2265 In 802.3ad mode, it works with systems that support IEEE
2266 802.3ad Dynamic Link Aggregation. Most managed and many unmanaged
2267 switches currently available support 802.3ad.
2269 The active-backup mode should work with any Layer-II switch.
2271 8. Where does a bonding device get its MAC address from?
2273 When using slave devices that have fixed MAC addresses, or when
2274 the fail_over_mac option is enabled, the bonding device's MAC address is
2275 the MAC address of the active slave.
2277 For other configurations, if not explicitly configured (with
2278 ifconfig or ip link), the MAC address of the bonding device is taken from
2279 its first slave device. This MAC address is then passed to all following
2280 slaves and remains persistent (even if the first slave is removed) until
2281 the bonding device is brought down or reconfigured.
2283 If you wish to change the MAC address, you can set it with
2284 ifconfig or ip link:
2286 # ifconfig bond0 hw ether 00:11:22:33:44:55
2288 # ip link set bond0 address 66:77:88:99:aa:bb
2290 The MAC address can be also changed by bringing down/up the
2291 device and then changing its slaves (or their order):
2293 # ifconfig bond0 down ; modprobe -r bonding
2294 # ifconfig bond0 .... up
2295 # ifenslave bond0 eth...
2297 This method will automatically take the address from the next
2298 slave that is added.
2300 To restore your slaves' MAC addresses, you need to detach them
2301 from the bond (`ifenslave -d bond0 eth0'). The bonding driver will
2302 then restore the MAC addresses that the slaves had before they were
2305 16. Resources and Links
2306 =======================
2308 The latest version of the bonding driver can be found in the latest
2309 version of the linux kernel, found on http://kernel.org
2311 The latest version of this document can be found in either the latest
2312 kernel source (named Documentation/networking/bonding.txt), or on the
2313 bonding sourceforge site:
2315 http://www.sourceforge.net/projects/bonding
2317 Discussions regarding the bonding driver take place primarily on the
2318 bonding-devel mailing list, hosted at sourceforge.net. If you have
2319 questions or problems, post them to the list. The list address is:
2321 bonding-devel@lists.sourceforge.net
2323 The administrative interface (to subscribe or unsubscribe) can
2326 https://lists.sourceforge.net/lists/listinfo/bonding-devel
2328 Donald Becker's Ethernet Drivers and diag programs may be found at :
2329 - http://www.scyld.com/network/
2331 You will also find a lot of information regarding Ethernet, NWay, MII,
2332 etc. at www.scyld.com.
git.openpandora.org / pandora-kernel.git / blob