2 Linux Ethernet Bonding Driver HOWTO
4 Latest update: 21 June 2005
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
19 The Linux bonding driver provides a method for aggregating
20 multiple network interfaces into a single logical "bonded" interface.
21 The behavior of the bonded interfaces depends upon the mode; generally
22 speaking, modes provide either hot standby or load balancing services.
23 Additionally, link integrity monitoring may be performed.
25 The bonding driver originally came from Donald Becker's
26 beowulf patches for kernel 2.0. It has changed quite a bit since, and
27 the original tools from extreme-linux and beowulf sites will not work
28 with this version of the driver.
30 For new versions of the driver, updated userspace tools, and
31 who to ask for help, please follow the links at the end of this file.
36 1. Bonding Driver Installation
38 2. Bonding Driver Options
40 3. Configuring Bonding Devices
41 3.1 Configuration with sysconfig support
42 3.1.1 Using DHCP with sysconfig
43 3.1.2 Configuring Multiple Bonds with sysconfig
44 3.2 Configuration with initscripts support
45 3.2.1 Using DHCP with initscripts
46 3.2.2 Configuring Multiple Bonds with initscripts
47 3.3 Configuring Bonding Manually
48 3.3.1 Configuring Multiple Bonds Manually
50 5. Querying Bonding Configuration
51 5.1 Bonding Configuration
52 5.2 Network Configuration
54 6. Switch Configuration
56 7. 802.1q VLAN Support
59 8.1 ARP Monitor Operation
60 8.2 Configuring Multiple ARP Targets
61 8.3 MII Monitor Operation
63 9. Potential Trouble Sources
64 9.1 Adventures in Routing
65 9.2 Ethernet Device Renaming
66 9.3 Painfully Slow Or No Failed Link Detection By Miimon
72 12. Configuring Bonding for High Availability
73 12.1 High Availability in a Single Switch Topology
74 12.2 High Availability in a Multiple Switch Topology
75 12.2.1 HA Bonding Mode Selection for Multiple Switch Topology
76 12.2.2 HA Link Monitoring for Multiple Switch Topology
78 13. Configuring Bonding for Maximum Throughput
79 13.1 Maximum Throughput in a Single Switch Topology
80 13.1.1 MT Bonding Mode Selection for Single Switch Topology
81 13.1.2 MT Link Monitoring for Single Switch Topology
82 13.2 Maximum Throughput in a Multiple Switch Topology
83 13.2.1 MT Bonding Mode Selection for Multiple Switch Topology
84 13.2.2 MT Link Monitoring for Multiple Switch Topology
86 14. Switch Behavior Issues
87 14.1 Link Establishment and Failover Delays
88 14.2 Duplicated Incoming Packets
90 15. Hardware Specific Considerations
93 16. Frequently Asked Questions
95 17. Resources and Links
98 1. Bonding Driver Installation
99 ==============================
101 Most popular distro kernels ship with the bonding driver
102 already available as a module and the ifenslave user level control
103 program installed and ready for use. If your distro does not, or you
104 have need to compile bonding from source (e.g., configuring and
105 installing a mainline kernel from kernel.org), you'll need to perform
108 1.1 Configure and build the kernel with bonding
109 -----------------------------------------------
111 The current version of the bonding driver is available in the
112 drivers/net/bonding subdirectory of the most recent kernel source
113 (which is available on http://kernel.org). Most users "rolling their
114 own" will want to use the most recent kernel from kernel.org.
116 Configure kernel with "make menuconfig" (or "make xconfig" or
117 "make config"), then select "Bonding driver support" in the "Network
118 device support" section. It is recommended that you configure the
119 driver as module since it is currently the only way to pass parameters
120 to the driver or configure more than one bonding device.
122 Build and install the new kernel and modules, then continue
123 below to install ifenslave.
125 1.2 Install ifenslave Control Utility
126 -------------------------------------
128 The ifenslave user level control program is included in the
129 kernel source tree, in the file Documentation/networking/ifenslave.c.
130 It is generally recommended that you use the ifenslave that
131 corresponds to the kernel that you are using (either from the same
132 source tree or supplied with the distro), however, ifenslave
133 executables from older kernels should function (but features newer
134 than the ifenslave release are not supported). Running an ifenslave
135 that is newer than the kernel is not supported, and may or may not
138 To install ifenslave, do the following:
140 # gcc -Wall -O -I/usr/src/linux/include ifenslave.c -o ifenslave
141 # cp ifenslave /sbin/ifenslave
143 If your kernel source is not in "/usr/src/linux," then replace
144 "/usr/src/linux/include" in the above with the location of your kernel
145 source include directory.
147 You may wish to back up any existing /sbin/ifenslave, or, for
148 testing or informal use, tag the ifenslave to the kernel version
149 (e.g., name the ifenslave executable /sbin/ifenslave-2.6.10).
153 If you omit the "-I" or specify an incorrect directory, you
154 may end up with an ifenslave that is incompatible with the kernel
155 you're trying to build it for. Some distros (e.g., Red Hat from 7.1
156 onwards) do not have /usr/include/linux symbolically linked to the
157 default kernel source include directory.
160 2. Bonding Driver Options
161 =========================
163 Options for the bonding driver are supplied as parameters to
164 the bonding module at load time. They may be given as command line
165 arguments to the insmod or modprobe command, but are usually specified
166 in either the /etc/modules.conf or /etc/modprobe.conf configuration
167 file, or in a distro-specific configuration file (some of which are
168 detailed in the next section).
170 The available bonding driver parameters are listed below. If a
171 parameter is not specified the default value is used. When initially
172 configuring a bond, it is recommended "tail -f /var/log/messages" be
173 run in a separate window to watch for bonding driver error messages.
175 It is critical that either the miimon or arp_interval and
176 arp_ip_target parameters be specified, otherwise serious network
177 degradation will occur during link failures. Very few devices do not
178 support at least miimon, so there is really no reason not to use it.
180 Options with textual values will accept either the text name
181 or, for backwards compatibility, the option value. E.g.,
182 "mode=802.3ad" and "mode=4" set the same mode.
184 The parameters are as follows:
188 Specifies the ARP link monitoring frequency in milliseconds.
189 If ARP monitoring is used in an etherchannel compatible mode
190 (modes 0 and 2), the switch should be configured in a mode
191 that evenly distributes packets across all links. If the
192 switch is configured to distribute the packets in an XOR
193 fashion, all replies from the ARP targets will be received on
194 the same link which could cause the other team members to
195 fail. ARP monitoring should not be used in conjunction with
196 miimon. A value of 0 disables ARP monitoring. The default
201 Specifies the IP addresses to use as ARP monitoring peers when
202 arp_interval is > 0. These are the targets of the ARP request
203 sent to determine the health of the link to the targets.
204 Specify these values in ddd.ddd.ddd.ddd format. Multiple IP
205 addresses must be separated by a comma. At least one IP
206 address must be given for ARP monitoring to function. The
207 maximum number of targets that can be specified is 16. The
208 default value is no IP addresses.
212 Specifies the time, in milliseconds, to wait before disabling
213 a slave after a link failure has been detected. This option
214 is only valid for the miimon link monitor. The downdelay
215 value should be a multiple of the miimon value; if not, it
216 will be rounded down to the nearest multiple. The default
221 Option specifying the rate in which we'll ask our link partner
222 to transmit LACPDU packets in 802.3ad mode. Possible values
226 Request partner to transmit LACPDUs every 30 seconds
229 Request partner to transmit LACPDUs every 1 second
235 Specifies the number of bonding devices to create for this
236 instance of the bonding driver. E.g., if max_bonds is 3, and
237 the bonding driver is not already loaded, then bond0, bond1
238 and bond2 will be created. The default value is 1.
242 Specifies the MII link monitoring frequency in milliseconds.
243 This determines how often the link state of each slave is
244 inspected for link failures. A value of zero disables MII
245 link monitoring. A value of 100 is a good starting point.
246 The use_carrier option, below, affects how the link state is
247 determined. See the High Availability section for additional
248 information. The default value is 0.
252 Specifies one of the bonding policies. The default is
253 balance-rr (round robin). Possible values are:
257 Round-robin policy: Transmit packets in sequential
258 order from the first available slave through the
259 last. This mode provides load balancing and fault
264 Active-backup policy: Only one slave in the bond is
265 active. A different slave becomes active if, and only
266 if, the active slave fails. The bond's MAC address is
267 externally visible on only one port (network adapter)
268 to avoid confusing the switch.
270 In bonding version 2.6.2 or later, when a failover
271 occurs in active-backup mode, bonding will issue one
272 or more gratuitous ARPs on the newly active slave.
273 One gratutious ARP is issued for the bonding master
274 interface and each VLAN interfaces configured above
275 it, provided that the interface has at least one IP
276 address configured. Gratuitous ARPs issued for VLAN
277 interfaces are tagged with the appropriate VLAN id.
279 This mode provides fault tolerance. The primary
280 option, documented below, affects the behavior of this
285 XOR policy: Transmit based on the selected transmit
286 hash policy. The default policy is a simple [(source
287 MAC address XOR'd with destination MAC address) modulo
288 slave count]. Alternate transmit policies may be
289 selected via the xmit_hash_policy option, described
292 This mode provides load balancing and fault tolerance.
296 Broadcast policy: transmits everything on all slave
297 interfaces. This mode provides fault tolerance.
301 IEEE 802.3ad Dynamic link aggregation. Creates
302 aggregation groups that share the same speed and
303 duplex settings. Utilizes all slaves in the active
304 aggregator according to the 802.3ad specification.
306 Slave selection for outgoing traffic is done according
307 to the transmit hash policy, which may be changed from
308 the default simple XOR policy via the xmit_hash_policy
309 option, documented below. Note that not all transmit
310 policies may be 802.3ad compliant, particularly in
311 regards to the packet mis-ordering requirements of
312 section 43.2.4 of the 802.3ad standard. Differing
313 peer implementations will have varying tolerances for
318 1. Ethtool support in the base drivers for retrieving
319 the speed and duplex of each slave.
321 2. A switch that supports IEEE 802.3ad Dynamic link
324 Most switches will require some type of configuration
325 to enable 802.3ad mode.
329 Adaptive transmit load balancing: channel bonding that
330 does not require any special switch support. The
331 outgoing traffic is distributed according to the
332 current load (computed relative to the speed) on each
333 slave. Incoming traffic is received by the current
334 slave. If the receiving slave fails, another slave
335 takes over the MAC address of the failed receiving
340 Ethtool support in the base drivers for retrieving the
345 Adaptive load balancing: includes balance-tlb plus
346 receive load balancing (rlb) for IPV4 traffic, and
347 does not require any special switch support. The
348 receive load balancing is achieved by ARP negotiation.
349 The bonding driver intercepts the ARP Replies sent by
350 the local system on their way out and overwrites the
351 source hardware address with the unique hardware
352 address of one of the slaves in the bond such that
353 different peers use different hardware addresses for
356 Receive traffic from connections created by the server
357 is also balanced. When the local system sends an ARP
358 Request the bonding driver copies and saves the peer's
359 IP information from the ARP packet. When the ARP
360 Reply arrives from the peer, its hardware address is
361 retrieved and the bonding driver initiates an ARP
362 reply to this peer assigning it to one of the slaves
363 in the bond. A problematic outcome of using ARP
364 negotiation for balancing is that each time that an
365 ARP request is broadcast it uses the hardware address
366 of the bond. Hence, peers learn the hardware address
367 of the bond and the balancing of receive traffic
368 collapses to the current slave. This is handled by
369 sending updates (ARP Replies) to all the peers with
370 their individually assigned hardware address such that
371 the traffic is redistributed. Receive traffic is also
372 redistributed when a new slave is added to the bond
373 and when an inactive slave is re-activated. The
374 receive load is distributed sequentially (round robin)
375 among the group of highest speed slaves in the bond.
377 When a link is reconnected or a new slave joins the
378 bond the receive traffic is redistributed among all
379 active slaves in the bond by initiating ARP Replies
380 with the selected mac address to each of the
381 clients. The updelay parameter (detailed below) must
382 be set to a value equal or greater than the switch's
383 forwarding delay so that the ARP Replies sent to the
384 peers will not be blocked by the switch.
388 1. Ethtool support in the base drivers for retrieving
389 the speed of each slave.
391 2. Base driver support for setting the hardware
392 address of a device while it is open. This is
393 required so that there will always be one slave in the
394 team using the bond hardware address (the
395 curr_active_slave) while having a unique hardware
396 address for each slave in the bond. If the
397 curr_active_slave fails its hardware address is
398 swapped with the new curr_active_slave that was
403 A string (eth0, eth2, etc) specifying which slave is the
404 primary device. The specified device will always be the
405 active slave while it is available. Only when the primary is
406 off-line will alternate devices be used. This is useful when
407 one slave is preferred over another, e.g., when one slave has
408 higher throughput than another.
410 The primary option is only valid for active-backup mode.
414 Specifies the time, in milliseconds, to wait before enabling a
415 slave after a link recovery has been detected. This option is
416 only valid for the miimon link monitor. The updelay value
417 should be a multiple of the miimon value; if not, it will be
418 rounded down to the nearest multiple. The default value is 0.
422 Specifies whether or not miimon should use MII or ETHTOOL
423 ioctls vs. netif_carrier_ok() to determine the link
424 status. The MII or ETHTOOL ioctls are less efficient and
425 utilize a deprecated calling sequence within the kernel. The
426 netif_carrier_ok() relies on the device driver to maintain its
427 state with netif_carrier_on/off; at this writing, most, but
428 not all, device drivers support this facility.
430 If bonding insists that the link is up when it should not be,
431 it may be that your network device driver does not support
432 netif_carrier_on/off. The default state for netif_carrier is
433 "carrier on," so if a driver does not support netif_carrier,
434 it will appear as if the link is always up. In this case,
435 setting use_carrier to 0 will cause bonding to revert to the
436 MII / ETHTOOL ioctl method to determine the link state.
438 A value of 1 enables the use of netif_carrier_ok(), a value of
439 0 will use the deprecated MII / ETHTOOL ioctls. The default
444 Selects the transmit hash policy to use for slave selection in
445 balance-xor and 802.3ad modes. Possible values are:
449 Uses XOR of hardware MAC addresses to generate the
452 (source MAC XOR destination MAC) modulo slave count
454 This algorithm will place all traffic to a particular
455 network peer on the same slave.
457 This algorithm is 802.3ad compliant.
461 This policy uses upper layer protocol information,
462 when available, to generate the hash. This allows for
463 traffic to a particular network peer to span multiple
464 slaves, although a single connection will not span
467 The formula for unfragmented TCP and UDP packets is
469 ((source port XOR dest port) XOR
470 ((source IP XOR dest IP) AND 0xffff)
473 For fragmented TCP or UDP packets and all other IP
474 protocol traffic, the source and destination port
475 information is omitted. For non-IP traffic, the
476 formula is the same as for the layer2 transmit hash
479 This policy is intended to mimic the behavior of
480 certain switches, notably Cisco switches with PFC2 as
481 well as some Foundry and IBM products.
483 This algorithm is not fully 802.3ad compliant. A
484 single TCP or UDP conversation containing both
485 fragmented and unfragmented packets will see packets
486 striped across two interfaces. This may result in out
487 of order delivery. Most traffic types will not meet
488 this criteria, as TCP rarely fragments traffic, and
489 most UDP traffic is not involved in extended
490 conversations. Other implementations of 802.3ad may
491 or may not tolerate this noncompliance.
493 The default value is layer2. This option was added in bonding
494 version 2.6.3. In earlier versions of bonding, this parameter does
495 not exist, and the layer2 policy is the only policy.
498 3. Configuring Bonding Devices
499 ==============================
501 There are, essentially, two methods for configuring bonding:
502 with support from the distro's network initialization scripts, and
503 without. Distros generally use one of two packages for the network
504 initialization scripts: initscripts or sysconfig. Recent versions of
505 these packages have support for bonding, while older versions do not.
507 We will first describe the options for configuring bonding for
508 distros using versions of initscripts and sysconfig with full or
509 partial support for bonding, then provide information on enabling
510 bonding without support from the network initialization scripts (i.e.,
511 older versions of initscripts or sysconfig).
513 If you're unsure whether your distro uses sysconfig or
514 initscripts, or don't know if it's new enough, have no fear.
515 Determining this is fairly straightforward.
517 First, issue the command:
521 It will respond with a line of text starting with either
522 "initscripts" or "sysconfig," followed by some numbers. This is the
523 package that provides your network initialization scripts.
525 Next, to determine if your installation supports bonding,
528 $ grep ifenslave /sbin/ifup
530 If this returns any matches, then your initscripts or
531 sysconfig has support for bonding.
533 3.1 Configuration with sysconfig support
534 ----------------------------------------
536 This section applies to distros using a version of sysconfig
537 with bonding support, for example, SuSE Linux Enterprise Server 9.
539 SuSE SLES 9's networking configuration system does support
540 bonding, however, at this writing, the YaST system configuration
541 frontend does not provide any means to work with bonding devices.
542 Bonding devices can be managed by hand, however, as follows.
544 First, if they have not already been configured, configure the
545 slave devices. On SLES 9, this is most easily done by running the
546 yast2 sysconfig configuration utility. The goal is for to create an
547 ifcfg-id file for each slave device. The simplest way to accomplish
548 this is to configure the devices for DHCP (this is only to get the
549 file ifcfg-id file created; see below for some issues with DHCP). The
550 name of the configuration file for each device will be of the form:
552 ifcfg-id-xx:xx:xx:xx:xx:xx
554 Where the "xx" portion will be replaced with the digits from
555 the device's permanent MAC address.
557 Once the set of ifcfg-id-xx:xx:xx:xx:xx:xx files has been
558 created, it is necessary to edit the configuration files for the slave
559 devices (the MAC addresses correspond to those of the slave devices).
560 Before editing, the file will contain multiple lines, and will look
566 UNIQUE='XNzu.WeZGOGF+4wE'
567 _nm_name='bus-pci-0001:61:01.0'
569 Change the BOOTPROTO and STARTMODE lines to the following:
574 Do not alter the UNIQUE or _nm_name lines. Remove any other
575 lines (USERCTL, etc).
577 Once the ifcfg-id-xx:xx:xx:xx:xx:xx files have been modified,
578 it's time to create the configuration file for the bonding device
579 itself. This file is named ifcfg-bondX, where X is the number of the
580 bonding device to create, starting at 0. The first such file is
581 ifcfg-bond0, the second is ifcfg-bond1, and so on. The sysconfig
582 network configuration system will correctly start multiple instances
585 The contents of the ifcfg-bondX file is as follows:
588 BROADCAST="10.0.2.255"
590 NETMASK="255.255.0.0"
595 BONDING_MODULE_OPTS="mode=active-backup miimon=100"
596 BONDING_SLAVE0="eth0"
597 BONDING_SLAVE1="bus-pci-0000:06:08.1"
599 Replace the sample BROADCAST, IPADDR, NETMASK and NETWORK
600 values with the appropriate values for your network.
602 The STARTMODE specifies when the device is brought online.
603 The possible values are:
605 onboot: The device is started at boot time. If you're not
606 sure, this is probably what you want.
608 manual: The device is started only when ifup is called
609 manually. Bonding devices may be configured this
610 way if you do not wish them to start automatically
611 at boot for some reason.
613 hotplug: The device is started by a hotplug event. This is not
614 a valid choice for a bonding device.
616 off or ignore: The device configuration is ignored.
618 The line BONDING_MASTER='yes' indicates that the device is a
619 bonding master device. The only useful value is "yes."
621 The contents of BONDING_MODULE_OPTS are supplied to the
622 instance of the bonding module for this device. Specify the options
623 for the bonding mode, link monitoring, and so on here. Do not include
624 the max_bonds bonding parameter; this will confuse the configuration
625 system if you have multiple bonding devices.
627 Finally, supply one BONDING_SLAVEn="slave device" for each
628 slave. where "n" is an increasing value, one for each slave. The
629 "slave device" is either an interface name, e.g., "eth0", or a device
630 specifier for the network device. The interface name is easier to
631 find, but the ethN names are subject to change at boot time if, e.g.,
632 a device early in the sequence has failed. The device specifiers
633 (bus-pci-0000:06:08.1 in the example above) specify the physical
634 network device, and will not change unless the device's bus location
635 changes (for example, it is moved from one PCI slot to another). The
636 example above uses one of each type for demonstration purposes; most
637 configurations will choose one or the other for all slave devices.
639 When all configuration files have been modified or created,
640 networking must be restarted for the configuration changes to take
641 effect. This can be accomplished via the following:
643 # /etc/init.d/network restart
645 Note that the network control script (/sbin/ifdown) will
646 remove the bonding module as part of the network shutdown processing,
647 so it is not necessary to remove the module by hand if, e.g., the
648 module parameters have changed.
650 Also, at this writing, YaST/YaST2 will not manage bonding
651 devices (they do not show bonding interfaces on its list of network
652 devices). It is necessary to edit the configuration file by hand to
653 change the bonding configuration.
655 Additional general options and details of the ifcfg file
656 format can be found in an example ifcfg template file:
658 /etc/sysconfig/network/ifcfg.template
660 Note that the template does not document the various BONDING_
661 settings described above, but does describe many of the other options.
663 3.1.1 Using DHCP with sysconfig
664 -------------------------------
666 Under sysconfig, configuring a device with BOOTPROTO='dhcp'
667 will cause it to query DHCP for its IP address information. At this
668 writing, this does not function for bonding devices; the scripts
669 attempt to obtain the device address from DHCP prior to adding any of
670 the slave devices. Without active slaves, the DHCP requests are not
673 3.1.2 Configuring Multiple Bonds with sysconfig
674 -----------------------------------------------
676 The sysconfig network initialization system is capable of
677 handling multiple bonding devices. All that is necessary is for each
678 bonding instance to have an appropriately configured ifcfg-bondX file
679 (as described above). Do not specify the "max_bonds" parameter to any
680 instance of bonding, as this will confuse sysconfig. If you require
681 multiple bonding devices with identical parameters, create multiple
684 Because the sysconfig scripts supply the bonding module
685 options in the ifcfg-bondX file, it is not necessary to add them to
686 the system /etc/modules.conf or /etc/modprobe.conf configuration file.
688 3.2 Configuration with initscripts support
689 ------------------------------------------
691 This section applies to distros using a version of initscripts
692 with bonding support, for example, Red Hat Linux 9 or Red Hat
693 Enterprise Linux version 3 or 4. On these systems, the network
694 initialization scripts have some knowledge of bonding, and can be
695 configured to control bonding devices.
697 These distros will not automatically load the network adapter
698 driver unless the ethX device is configured with an IP address.
699 Because of this constraint, users must manually configure a
700 network-script file for all physical adapters that will be members of
701 a bondX link. Network script files are located in the directory:
703 /etc/sysconfig/network-scripts
705 The file name must be prefixed with "ifcfg-eth" and suffixed
706 with the adapter's physical adapter number. For example, the script
707 for eth0 would be named /etc/sysconfig/network-scripts/ifcfg-eth0.
708 Place the following text in the file:
717 The DEVICE= line will be different for every ethX device and
718 must correspond with the name of the file, i.e., ifcfg-eth1 must have
719 a device line of DEVICE=eth1. The setting of the MASTER= line will
720 also depend on the final bonding interface name chosen for your bond.
721 As with other network devices, these typically start at 0, and go up
722 one for each device, i.e., the first bonding instance is bond0, the
723 second is bond1, and so on.
725 Next, create a bond network script. The file name for this
726 script will be /etc/sysconfig/network-scripts/ifcfg-bondX where X is
727 the number of the bond. For bond0 the file is named "ifcfg-bond0",
728 for bond1 it is named "ifcfg-bond1", and so on. Within that file,
729 place the following text:
733 NETMASK=255.255.255.0
735 BROADCAST=192.168.1.255
740 Be sure to change the networking specific lines (IPADDR,
741 NETMASK, NETWORK and BROADCAST) to match your network configuration.
743 Finally, it is necessary to edit /etc/modules.conf (or
744 /etc/modprobe.conf, depending upon your distro) to load the bonding
745 module with your desired options when the bond0 interface is brought
746 up. The following lines in /etc/modules.conf (or modprobe.conf) will
747 load the bonding module, and select its options:
750 options bond0 mode=balance-alb miimon=100
752 Replace the sample parameters with the appropriate set of
753 options for your configuration.
755 Finally run "/etc/rc.d/init.d/network restart" as root. This
756 will restart the networking subsystem and your bond link should be now
759 3.2.1 Using DHCP with initscripts
760 ---------------------------------
762 Recent versions of initscripts (the version supplied with
763 Fedora Core 3 and Red Hat Enterprise Linux 4 is reported to work) do
764 have support for assigning IP information to bonding devices via DHCP.
766 To configure bonding for DHCP, configure it as described
767 above, except replace the line "BOOTPROTO=none" with "BOOTPROTO=dhcp"
768 and add a line consisting of "TYPE=Bonding". Note that the TYPE value
771 3.2.2 Configuring Multiple Bonds with initscripts
772 -------------------------------------------------
774 At this writing, the initscripts package does not directly
775 support loading the bonding driver multiple times, so the process for
776 doing so is the same as described in the "Configuring Multiple Bonds
777 Manually" section, below.
779 NOTE: It has been observed that some Red Hat supplied kernels
780 are apparently unable to rename modules at load time (the "-o bond1"
781 part). Attempts to pass that option to modprobe will produce an
782 "Operation not permitted" error. This has been reported on some
783 Fedora Core kernels, and has been seen on RHEL 4 as well. On kernels
784 exhibiting this problem, it will be impossible to configure multiple
785 bonds with differing parameters.
787 3.3 Configuring Bonding Manually
788 --------------------------------
790 This section applies to distros whose network initialization
791 scripts (the sysconfig or initscripts package) do not have specific
792 knowledge of bonding. One such distro is SuSE Linux Enterprise Server
795 The general method for these systems is to place the bonding
796 module parameters into /etc/modules.conf or /etc/modprobe.conf (as
797 appropriate for the installed distro), then add modprobe and/or
798 ifenslave commands to the system's global init script. The name of
799 the global init script differs; for sysconfig, it is
800 /etc/init.d/boot.local and for initscripts it is /etc/rc.d/rc.local.
802 For example, if you wanted to make a simple bond of two e100
803 devices (presumed to be eth0 and eth1), and have it persist across
804 reboots, edit the appropriate file (/etc/init.d/boot.local or
805 /etc/rc.d/rc.local), and add the following:
807 modprobe bonding mode=balance-alb miimon=100
809 ifconfig bond0 192.168.1.1 netmask 255.255.255.0 up
813 Replace the example bonding module parameters and bond0
814 network configuration (IP address, netmask, etc) with the appropriate
815 values for your configuration.
817 Unfortunately, this method will not provide support for the
818 ifup and ifdown scripts on the bond devices. To reload the bonding
819 configuration, it is necessary to run the initialization script, e.g.,
821 # /etc/init.d/boot.local
827 It may be desirable in such a case to create a separate script
828 which only initializes the bonding configuration, then call that
829 separate script from within boot.local. This allows for bonding to be
830 enabled without re-running the entire global init script.
832 To shut down the bonding devices, it is necessary to first
833 mark the bonding device itself as being down, then remove the
834 appropriate device driver modules. For our example above, you can do
837 # ifconfig bond0 down
841 Again, for convenience, it may be desirable to create a script
845 3.3.1 Configuring Multiple Bonds Manually
846 -----------------------------------------
848 This section contains information on configuring multiple
849 bonding devices with differing options for those systems whose network
850 initialization scripts lack support for configuring multiple bonds.
852 If you require multiple bonding devices, but all with the same
853 options, you may wish to use the "max_bonds" module parameter,
856 To create multiple bonding devices with differing options, it
857 is necessary to load the bonding driver multiple times. Note that
858 current versions of the sysconfig network initialization scripts
859 handle this automatically; if your distro uses these scripts, no
860 special action is needed. See the section Configuring Bonding
861 Devices, above, if you're not sure about your network initialization
864 To load multiple instances of the module, it is necessary to
865 specify a different name for each instance (the module loading system
866 requires that every loaded module, even multiple instances of the same
867 module, have a unique name). This is accomplished by supplying
868 multiple sets of bonding options in /etc/modprobe.conf, for example:
871 options bond0 -o bond0 mode=balance-rr miimon=100
874 options bond1 -o bond1 mode=balance-alb miimon=50
876 will load the bonding module two times. The first instance is
877 named "bond0" and creates the bond0 device in balance-rr mode with an
878 miimon of 100. The second instance is named "bond1" and creates the
879 bond1 device in balance-alb mode with an miimon of 50.
881 In some circumstances (typically with older distributions),
882 the above does not work, and the second bonding instance never sees
883 its options. In that case, the second options line can be substituted
886 install bond1 /sbin/modprobe --ignore-install bonding -o bond1 \
887 mode=balance-alb miimon=50
889 This may be repeated any number of times, specifying a new and
890 unique name in place of bond1 for each subsequent instance.
893 5. Querying Bonding Configuration
894 =================================
896 5.1 Bonding Configuration
897 -------------------------
899 Each bonding device has a read-only file residing in the
900 /proc/net/bonding directory. The file contents include information
901 about the bonding configuration, options and state of each slave.
903 For example, the contents of /proc/net/bonding/bond0 after the
904 driver is loaded with parameters of mode=0 and miimon=1000 is
905 generally as follows:
907 Ethernet Channel Bonding Driver: 2.6.1 (October 29, 2004)
908 Bonding Mode: load balancing (round-robin)
909 Currently Active Slave: eth0
911 MII Polling Interval (ms): 1000
915 Slave Interface: eth1
917 Link Failure Count: 1
919 Slave Interface: eth0
921 Link Failure Count: 1
923 The precise format and contents will change depending upon the
924 bonding configuration, state, and version of the bonding driver.
926 5.2 Network configuration
927 -------------------------
929 The network configuration can be inspected using the ifconfig
930 command. Bonding devices will have the MASTER flag set; Bonding slave
931 devices will have the SLAVE flag set. The ifconfig output does not
932 contain information on which slaves are associated with which masters.
934 In the example below, the bond0 interface is the master
935 (MASTER) while eth0 and eth1 are slaves (SLAVE). Notice all slaves of
936 bond0 have the same MAC address (HWaddr) as bond0 for all modes except
937 TLB and ALB that require a unique MAC address for each slave.
940 bond0 Link encap:Ethernet HWaddr 00:C0:F0:1F:37:B4
941 inet addr:XXX.XXX.XXX.YYY Bcast:XXX.XXX.XXX.255 Mask:255.255.252.0
942 UP BROADCAST RUNNING MASTER MULTICAST MTU:1500 Metric:1
943 RX packets:7224794 errors:0 dropped:0 overruns:0 frame:0
944 TX packets:3286647 errors:1 dropped:0 overruns:1 carrier:0
945 collisions:0 txqueuelen:0
947 eth0 Link encap:Ethernet HWaddr 00:C0:F0:1F:37:B4
948 inet addr:XXX.XXX.XXX.YYY Bcast:XXX.XXX.XXX.255 Mask:255.255.252.0
949 UP BROADCAST RUNNING SLAVE MULTICAST MTU:1500 Metric:1
950 RX packets:3573025 errors:0 dropped:0 overruns:0 frame:0
951 TX packets:1643167 errors:1 dropped:0 overruns:1 carrier:0
952 collisions:0 txqueuelen:100
953 Interrupt:10 Base address:0x1080
955 eth1 Link encap:Ethernet HWaddr 00:C0:F0:1F:37:B4
956 inet addr:XXX.XXX.XXX.YYY Bcast:XXX.XXX.XXX.255 Mask:255.255.252.0
957 UP BROADCAST RUNNING SLAVE MULTICAST MTU:1500 Metric:1
958 RX packets:3651769 errors:0 dropped:0 overruns:0 frame:0
959 TX packets:1643480 errors:0 dropped:0 overruns:0 carrier:0
960 collisions:0 txqueuelen:100
961 Interrupt:9 Base address:0x1400
963 6. Switch Configuration
964 =======================
966 For this section, "switch" refers to whatever system the
967 bonded devices are directly connected to (i.e., where the other end of
968 the cable plugs into). This may be an actual dedicated switch device,
969 or it may be another regular system (e.g., another computer running
972 The active-backup, balance-tlb and balance-alb modes do not
973 require any specific configuration of the switch.
975 The 802.3ad mode requires that the switch have the appropriate
976 ports configured as an 802.3ad aggregation. The precise method used
977 to configure this varies from switch to switch, but, for example, a
978 Cisco 3550 series switch requires that the appropriate ports first be
979 grouped together in a single etherchannel instance, then that
980 etherchannel is set to mode "lacp" to enable 802.3ad (instead of
981 standard EtherChannel).
983 The balance-rr, balance-xor and broadcast modes generally
984 require that the switch have the appropriate ports grouped together.
985 The nomenclature for such a group differs between switches, it may be
986 called an "etherchannel" (as in the Cisco example, above), a "trunk
987 group" or some other similar variation. For these modes, each switch
988 will also have its own configuration options for the switch's transmit
989 policy to the bond. Typical choices include XOR of either the MAC or
990 IP addresses. The transmit policy of the two peers does not need to
991 match. For these three modes, the bonding mode really selects a
992 transmit policy for an EtherChannel group; all three will interoperate
993 with another EtherChannel group.
996 7. 802.1q VLAN Support
997 ======================
999 It is possible to configure VLAN devices over a bond interface
1000 using the 8021q driver. However, only packets coming from the 8021q
1001 driver and passing through bonding will be tagged by default. Self
1002 generated packets, for example, bonding's learning packets or ARP
1003 packets generated by either ALB mode or the ARP monitor mechanism, are
1004 tagged internally by bonding itself. As a result, bonding must
1005 "learn" the VLAN IDs configured above it, and use those IDs to tag
1006 self generated packets.
1008 For reasons of simplicity, and to support the use of adapters
1009 that can do VLAN hardware acceleration offloading, the bonding
1010 interface declares itself as fully hardware offloading capable, it gets
1011 the add_vid/kill_vid notifications to gather the necessary
1012 information, and it propagates those actions to the slaves. In case
1013 of mixed adapter types, hardware accelerated tagged packets that
1014 should go through an adapter that is not offloading capable are
1015 "un-accelerated" by the bonding driver so the VLAN tag sits in the
1018 VLAN interfaces *must* be added on top of a bonding interface
1019 only after enslaving at least one slave. The bonding interface has a
1020 hardware address of 00:00:00:00:00:00 until the first slave is added.
1021 If the VLAN interface is created prior to the first enslavement, it
1022 would pick up the all-zeroes hardware address. Once the first slave
1023 is attached to the bond, the bond device itself will pick up the
1024 slave's hardware address, which is then available for the VLAN device.
1026 Also, be aware that a similar problem can occur if all slaves
1027 are released from a bond that still has one or more VLAN interfaces on
1028 top of it. When a new slave is added, the bonding interface will
1029 obtain its hardware address from the first slave, which might not
1030 match the hardware address of the VLAN interfaces (which was
1031 ultimately copied from an earlier slave).
1033 There are two methods to insure that the VLAN device operates
1034 with the correct hardware address if all slaves are removed from a
1037 1. Remove all VLAN interfaces then recreate them
1039 2. Set the bonding interface's hardware address so that it
1040 matches the hardware address of the VLAN interfaces.
1042 Note that changing a VLAN interface's HW address would set the
1043 underlying device -- i.e. the bonding interface -- to promiscuous
1044 mode, which might not be what you want.
1050 The bonding driver at present supports two schemes for
1051 monitoring a slave device's link state: the ARP monitor and the MII
1054 At the present time, due to implementation restrictions in the
1055 bonding driver itself, it is not possible to enable both ARP and MII
1056 monitoring simultaneously.
1058 8.1 ARP Monitor Operation
1059 -------------------------
1061 The ARP monitor operates as its name suggests: it sends ARP
1062 queries to one or more designated peer systems on the network, and
1063 uses the response as an indication that the link is operating. This
1064 gives some assurance that traffic is actually flowing to and from one
1065 or more peers on the local network.
1067 The ARP monitor relies on the device driver itself to verify
1068 that traffic is flowing. In particular, the driver must keep up to
1069 date the last receive time, dev->last_rx, and transmit start time,
1070 dev->trans_start. If these are not updated by the driver, then the
1071 ARP monitor will immediately fail any slaves using that driver, and
1072 those slaves will stay down. If networking monitoring (tcpdump, etc)
1073 shows the ARP requests and replies on the network, then it may be that
1074 your device driver is not updating last_rx and trans_start.
1076 8.2 Configuring Multiple ARP Targets
1077 ------------------------------------
1079 While ARP monitoring can be done with just one target, it can
1080 be useful in a High Availability setup to have several targets to
1081 monitor. In the case of just one target, the target itself may go
1082 down or have a problem making it unresponsive to ARP requests. Having
1083 an additional target (or several) increases the reliability of the ARP
1086 Multiple ARP targets must be separated by commas as follows:
1088 # example options for ARP monitoring with three targets
1090 options bond0 arp_interval=60 arp_ip_target=192.168.0.1,192.168.0.3,192.168.0.9
1092 For just a single target the options would resemble:
1094 # example options for ARP monitoring with one target
1096 options bond0 arp_interval=60 arp_ip_target=192.168.0.100
1099 8.3 MII Monitor Operation
1100 -------------------------
1102 The MII monitor monitors only the carrier state of the local
1103 network interface. It accomplishes this in one of three ways: by
1104 depending upon the device driver to maintain its carrier state, by
1105 querying the device's MII registers, or by making an ethtool query to
1108 If the use_carrier module parameter is 1 (the default value),
1109 then the MII monitor will rely on the driver for carrier state
1110 information (via the netif_carrier subsystem). As explained in the
1111 use_carrier parameter information, above, if the MII monitor fails to
1112 detect carrier loss on the device (e.g., when the cable is physically
1113 disconnected), it may be that the driver does not support
1116 If use_carrier is 0, then the MII monitor will first query the
1117 device's (via ioctl) MII registers and check the link state. If that
1118 request fails (not just that it returns carrier down), then the MII
1119 monitor will make an ethtool ETHOOL_GLINK request to attempt to obtain
1120 the same information. If both methods fail (i.e., the driver either
1121 does not support or had some error in processing both the MII register
1122 and ethtool requests), then the MII monitor will assume the link is
1125 9. Potential Sources of Trouble
1126 ===============================
1128 9.1 Adventures in Routing
1129 -------------------------
1131 When bonding is configured, it is important that the slave
1132 devices not have routes that supercede routes of the master (or,
1133 generally, not have routes at all). For example, suppose the bonding
1134 device bond0 has two slaves, eth0 and eth1, and the routing table is
1137 Kernel IP routing table
1138 Destination Gateway Genmask Flags MSS Window irtt Iface
1139 10.0.0.0 0.0.0.0 255.255.0.0 U 40 0 0 eth0
1140 10.0.0.0 0.0.0.0 255.255.0.0 U 40 0 0 eth1
1141 10.0.0.0 0.0.0.0 255.255.0.0 U 40 0 0 bond0
1142 127.0.0.0 0.0.0.0 255.0.0.0 U 40 0 0 lo
1144 This routing configuration will likely still update the
1145 receive/transmit times in the driver (needed by the ARP monitor), but
1146 may bypass the bonding driver (because outgoing traffic to, in this
1147 case, another host on network 10 would use eth0 or eth1 before bond0).
1149 The ARP monitor (and ARP itself) may become confused by this
1150 configuration, because ARP requests (generated by the ARP monitor)
1151 will be sent on one interface (bond0), but the corresponding reply
1152 will arrive on a different interface (eth0). This reply looks to ARP
1153 as an unsolicited ARP reply (because ARP matches replies on an
1154 interface basis), and is discarded. The MII monitor is not affected
1155 by the state of the routing table.
1157 The solution here is simply to insure that slaves do not have
1158 routes of their own, and if for some reason they must, those routes do
1159 not supercede routes of their master. This should generally be the
1160 case, but unusual configurations or errant manual or automatic static
1161 route additions may cause trouble.
1163 9.2 Ethernet Device Renaming
1164 ----------------------------
1166 On systems with network configuration scripts that do not
1167 associate physical devices directly with network interface names (so
1168 that the same physical device always has the same "ethX" name), it may
1169 be necessary to add some special logic to either /etc/modules.conf or
1170 /etc/modprobe.conf (depending upon which is installed on the system).
1172 For example, given a modules.conf containing the following:
1175 options bond0 mode=some-mode miimon=50
1181 If neither eth0 and eth1 are slaves to bond0, then when the
1182 bond0 interface comes up, the devices may end up reordered. This
1183 happens because bonding is loaded first, then its slave device's
1184 drivers are loaded next. Since no other drivers have been loaded,
1185 when the e1000 driver loads, it will receive eth0 and eth1 for its
1186 devices, but the bonding configuration tries to enslave eth2 and eth3
1187 (which may later be assigned to the tg3 devices).
1189 Adding the following:
1191 add above bonding e1000 tg3
1193 causes modprobe to load e1000 then tg3, in that order, when
1194 bonding is loaded. This command is fully documented in the
1195 modules.conf manual page.
1197 On systems utilizing modprobe.conf (or modprobe.conf.local),
1198 an equivalent problem can occur. In this case, the following can be
1199 added to modprobe.conf (or modprobe.conf.local, as appropriate), as
1200 follows (all on one line; it has been split here for clarity):
1202 install bonding /sbin/modprobe tg3; /sbin/modprobe e1000;
1203 /sbin/modprobe --ignore-install bonding
1205 This will, when loading the bonding module, rather than
1206 performing the normal action, instead execute the provided command.
1207 This command loads the device drivers in the order needed, then calls
1208 modprobe with --ignore-install to cause the normal action to then take
1209 place. Full documentation on this can be found in the modprobe.conf
1210 and modprobe manual pages.
1212 9.3. Painfully Slow Or No Failed Link Detection By Miimon
1213 ---------------------------------------------------------
1215 By default, bonding enables the use_carrier option, which
1216 instructs bonding to trust the driver to maintain carrier state.
1218 As discussed in the options section, above, some drivers do
1219 not support the netif_carrier_on/_off link state tracking system.
1220 With use_carrier enabled, bonding will always see these links as up,
1221 regardless of their actual state.
1223 Additionally, other drivers do support netif_carrier, but do
1224 not maintain it in real time, e.g., only polling the link state at
1225 some fixed interval. In this case, miimon will detect failures, but
1226 only after some long period of time has expired. If it appears that
1227 miimon is very slow in detecting link failures, try specifying
1228 use_carrier=0 to see if that improves the failure detection time. If
1229 it does, then it may be that the driver checks the carrier state at a
1230 fixed interval, but does not cache the MII register values (so the
1231 use_carrier=0 method of querying the registers directly works). If
1232 use_carrier=0 does not improve the failover, then the driver may cache
1233 the registers, or the problem may be elsewhere.
1235 Also, remember that miimon only checks for the device's
1236 carrier state. It has no way to determine the state of devices on or
1237 beyond other ports of a switch, or if a switch is refusing to pass
1238 traffic while still maintaining carrier on.
1243 If running SNMP agents, the bonding driver should be loaded
1244 before any network drivers participating in a bond. This requirement
1245 is due to the interface index (ipAdEntIfIndex) being associated to
1246 the first interface found with a given IP address. That is, there is
1247 only one ipAdEntIfIndex for each IP address. For example, if eth0 and
1248 eth1 are slaves of bond0 and the driver for eth0 is loaded before the
1249 bonding driver, the interface for the IP address will be associated
1250 with the eth0 interface. This configuration is shown below, the IP
1251 address 192.168.1.1 has an interface index of 2 which indexes to eth0
1252 in the ifDescr table (ifDescr.2).
1254 interfaces.ifTable.ifEntry.ifDescr.1 = lo
1255 interfaces.ifTable.ifEntry.ifDescr.2 = eth0
1256 interfaces.ifTable.ifEntry.ifDescr.3 = eth1
1257 interfaces.ifTable.ifEntry.ifDescr.4 = eth2
1258 interfaces.ifTable.ifEntry.ifDescr.5 = eth3
1259 interfaces.ifTable.ifEntry.ifDescr.6 = bond0
1260 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.10.10.10 = 5
1261 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.192.168.1.1 = 2
1262 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.74.20.94 = 4
1263 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.127.0.0.1 = 1
1265 This problem is avoided by loading the bonding driver before
1266 any network drivers participating in a bond. Below is an example of
1267 loading the bonding driver first, the IP address 192.168.1.1 is
1268 correctly associated with ifDescr.2.
1270 interfaces.ifTable.ifEntry.ifDescr.1 = lo
1271 interfaces.ifTable.ifEntry.ifDescr.2 = bond0
1272 interfaces.ifTable.ifEntry.ifDescr.3 = eth0
1273 interfaces.ifTable.ifEntry.ifDescr.4 = eth1
1274 interfaces.ifTable.ifEntry.ifDescr.5 = eth2
1275 interfaces.ifTable.ifEntry.ifDescr.6 = eth3
1276 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.10.10.10 = 6
1277 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.192.168.1.1 = 2
1278 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.74.20.94 = 5
1279 ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.127.0.0.1 = 1
1281 While some distributions may not report the interface name in
1282 ifDescr, the association between the IP address and IfIndex remains
1283 and SNMP functions such as Interface_Scan_Next will report that
1286 11. Promiscuous mode
1287 ====================
1289 When running network monitoring tools, e.g., tcpdump, it is
1290 common to enable promiscuous mode on the device, so that all traffic
1291 is seen (instead of seeing only traffic destined for the local host).
1292 The bonding driver handles promiscuous mode changes to the bonding
1293 master device (e.g., bond0), and propagates the setting to the slave
1296 For the balance-rr, balance-xor, broadcast, and 802.3ad modes,
1297 the promiscuous mode setting is propagated to all slaves.
1299 For the active-backup, balance-tlb and balance-alb modes, the
1300 promiscuous mode setting is propagated only to the active slave.
1302 For balance-tlb mode, the active slave is the slave currently
1303 receiving inbound traffic.
1305 For balance-alb mode, the active slave is the slave used as a
1306 "primary." This slave is used for mode-specific control traffic, for
1307 sending to peers that are unassigned or if the load is unbalanced.
1309 For the active-backup, balance-tlb and balance-alb modes, when
1310 the active slave changes (e.g., due to a link failure), the
1311 promiscuous setting will be propagated to the new active slave.
1313 12. Configuring Bonding for High Availability
1314 =============================================
1316 High Availability refers to configurations that provide
1317 maximum network availability by having redundant or backup devices,
1318 links or switches between the host and the rest of the world. The
1319 goal is to provide the maximum availability of network connectivity
1320 (i.e., the network always works), even though other configurations
1321 could provide higher throughput.
1323 12.1 High Availability in a Single Switch Topology
1324 --------------------------------------------------
1326 If two hosts (or a host and a single switch) are directly
1327 connected via multiple physical links, then there is no availability
1328 penalty to optimizing for maximum bandwidth. In this case, there is
1329 only one switch (or peer), so if it fails, there is no alternative
1330 access to fail over to. Additionally, the bonding load balance modes
1331 support link monitoring of their members, so if individual links fail,
1332 the load will be rebalanced across the remaining devices.
1334 See Section 13, "Configuring Bonding for Maximum Throughput"
1335 for information on configuring bonding with one peer device.
1337 12.2 High Availability in a Multiple Switch Topology
1338 ----------------------------------------------------
1340 With multiple switches, the configuration of bonding and the
1341 network changes dramatically. In multiple switch topologies, there is
1342 a trade off between network availability and usable bandwidth.
1344 Below is a sample network, configured to maximize the
1345 availability of the network:
1349 +-----+----+ +-----+----+
1350 | |port2 ISL port2| |
1351 | switch A +--------------------------+ switch B |
1353 +-----+----+ +-----++---+
1356 +-------------+ host1 +---------------+
1359 In this configuration, there is a link between the two
1360 switches (ISL, or inter switch link), and multiple ports connecting to
1361 the outside world ("port3" on each switch). There is no technical
1362 reason that this could not be extended to a third switch.
1364 12.2.1 HA Bonding Mode Selection for Multiple Switch Topology
1365 -------------------------------------------------------------
1367 In a topology such as the example above, the active-backup and
1368 broadcast modes are the only useful bonding modes when optimizing for
1369 availability; the other modes require all links to terminate on the
1370 same peer for them to behave rationally.
1372 active-backup: This is generally the preferred mode, particularly if
1373 the switches have an ISL and play together well. If the
1374 network configuration is such that one switch is specifically
1375 a backup switch (e.g., has lower capacity, higher cost, etc),
1376 then the primary option can be used to insure that the
1377 preferred link is always used when it is available.
1379 broadcast: This mode is really a special purpose mode, and is suitable
1380 only for very specific needs. For example, if the two
1381 switches are not connected (no ISL), and the networks beyond
1382 them are totally independent. In this case, if it is
1383 necessary for some specific one-way traffic to reach both
1384 independent networks, then the broadcast mode may be suitable.
1386 12.2.2 HA Link Monitoring Selection for Multiple Switch Topology
1387 ----------------------------------------------------------------
1389 The choice of link monitoring ultimately depends upon your
1390 switch. If the switch can reliably fail ports in response to other
1391 failures, then either the MII or ARP monitors should work. For
1392 example, in the above example, if the "port3" link fails at the remote
1393 end, the MII monitor has no direct means to detect this. The ARP
1394 monitor could be configured with a target at the remote end of port3,
1395 thus detecting that failure without switch support.
1397 In general, however, in a multiple switch topology, the ARP
1398 monitor can provide a higher level of reliability in detecting end to
1399 end connectivity failures (which may be caused by the failure of any
1400 individual component to pass traffic for any reason). Additionally,
1401 the ARP monitor should be configured with multiple targets (at least
1402 one for each switch in the network). This will insure that,
1403 regardless of which switch is active, the ARP monitor has a suitable
1407 13. Configuring Bonding for Maximum Throughput
1408 ==============================================
1410 13.1 Maximizing Throughput in a Single Switch Topology
1411 ------------------------------------------------------
1413 In a single switch configuration, the best method to maximize
1414 throughput depends upon the application and network environment. The
1415 various load balancing modes each have strengths and weaknesses in
1416 different environments, as detailed below.
1418 For this discussion, we will break down the topologies into
1419 two categories. Depending upon the destination of most traffic, we
1420 categorize them into either "gatewayed" or "local" configurations.
1422 In a gatewayed configuration, the "switch" is acting primarily
1423 as a router, and the majority of traffic passes through this router to
1424 other networks. An example would be the following:
1427 +----------+ +----------+
1428 | |eth0 port1| | to other networks
1429 | Host A +---------------------+ router +------------------->
1430 | +---------------------+ | Hosts B and C are out
1431 | |eth1 port2| | here somewhere
1432 +----------+ +----------+
1434 The router may be a dedicated router device, or another host
1435 acting as a gateway. For our discussion, the important point is that
1436 the majority of traffic from Host A will pass through the router to
1437 some other network before reaching its final destination.
1439 In a gatewayed network configuration, although Host A may
1440 communicate with many other systems, all of its traffic will be sent
1441 and received via one other peer on the local network, the router.
1443 Note that the case of two systems connected directly via
1444 multiple physical links is, for purposes of configuring bonding, the
1445 same as a gatewayed configuration. In that case, it happens that all
1446 traffic is destined for the "gateway" itself, not some other network
1449 In a local configuration, the "switch" is acting primarily as
1450 a switch, and the majority of traffic passes through this switch to
1451 reach other stations on the same network. An example would be the
1454 +----------+ +----------+ +--------+
1455 | |eth0 port1| +-------+ Host B |
1456 | Host A +------------+ switch |port3 +--------+
1457 | +------------+ | +--------+
1458 | |eth1 port2| +------------------+ Host C |
1459 +----------+ +----------+port4 +--------+
1462 Again, the switch may be a dedicated switch device, or another
1463 host acting as a gateway. For our discussion, the important point is
1464 that the majority of traffic from Host A is destined for other hosts
1465 on the same local network (Hosts B and C in the above example).
1467 In summary, in a gatewayed configuration, traffic to and from
1468 the bonded device will be to the same MAC level peer on the network
1469 (the gateway itself, i.e., the router), regardless of its final
1470 destination. In a local configuration, traffic flows directly to and
1471 from the final destinations, thus, each destination (Host B, Host C)
1472 will be addressed directly by their individual MAC addresses.
1474 This distinction between a gatewayed and a local network
1475 configuration is important because many of the load balancing modes
1476 available use the MAC addresses of the local network source and
1477 destination to make load balancing decisions. The behavior of each
1478 mode is described below.
1481 13.1.1 MT Bonding Mode Selection for Single Switch Topology
1482 -----------------------------------------------------------
1484 This configuration is the easiest to set up and to understand,
1485 although you will have to decide which bonding mode best suits your
1486 needs. The trade offs for each mode are detailed below:
1488 balance-rr: This mode is the only mode that will permit a single
1489 TCP/IP connection to stripe traffic across multiple
1490 interfaces. It is therefore the only mode that will allow a
1491 single TCP/IP stream to utilize more than one interface's
1492 worth of throughput. This comes at a cost, however: the
1493 striping often results in peer systems receiving packets out
1494 of order, causing TCP/IP's congestion control system to kick
1495 in, often by retransmitting segments.
1497 It is possible to adjust TCP/IP's congestion limits by
1498 altering the net.ipv4.tcp_reordering sysctl parameter. The
1499 usual default value is 3, and the maximum useful value is 127.
1500 For a four interface balance-rr bond, expect that a single
1501 TCP/IP stream will utilize no more than approximately 2.3
1502 interface's worth of throughput, even after adjusting
1505 Note that this out of order delivery occurs when both the
1506 sending and receiving systems are utilizing a multiple
1507 interface bond. Consider a configuration in which a
1508 balance-rr bond feeds into a single higher capacity network
1509 channel (e.g., multiple 100Mb/sec ethernets feeding a single
1510 gigabit ethernet via an etherchannel capable switch). In this
1511 configuration, traffic sent from the multiple 100Mb devices to
1512 a destination connected to the gigabit device will not see
1513 packets out of order. However, traffic sent from the gigabit
1514 device to the multiple 100Mb devices may or may not see
1515 traffic out of order, depending upon the balance policy of the
1516 switch. Many switches do not support any modes that stripe
1517 traffic (instead choosing a port based upon IP or MAC level
1518 addresses); for those devices, traffic flowing from the
1519 gigabit device to the many 100Mb devices will only utilize one
1522 If you are utilizing protocols other than TCP/IP, UDP for
1523 example, and your application can tolerate out of order
1524 delivery, then this mode can allow for single stream datagram
1525 performance that scales near linearly as interfaces are added
1528 This mode requires the switch to have the appropriate ports
1529 configured for "etherchannel" or "trunking."
1531 active-backup: There is not much advantage in this network topology to
1532 the active-backup mode, as the inactive backup devices are all
1533 connected to the same peer as the primary. In this case, a
1534 load balancing mode (with link monitoring) will provide the
1535 same level of network availability, but with increased
1536 available bandwidth. On the plus side, active-backup mode
1537 does not require any configuration of the switch, so it may
1538 have value if the hardware available does not support any of
1539 the load balance modes.
1541 balance-xor: This mode will limit traffic such that packets destined
1542 for specific peers will always be sent over the same
1543 interface. Since the destination is determined by the MAC
1544 addresses involved, this mode works best in a "local" network
1545 configuration (as described above), with destinations all on
1546 the same local network. This mode is likely to be suboptimal
1547 if all your traffic is passed through a single router (i.e., a
1548 "gatewayed" network configuration, as described above).
1550 As with balance-rr, the switch ports need to be configured for
1551 "etherchannel" or "trunking."
1553 broadcast: Like active-backup, there is not much advantage to this
1554 mode in this type of network topology.
1556 802.3ad: This mode can be a good choice for this type of network
1557 topology. The 802.3ad mode is an IEEE standard, so all peers
1558 that implement 802.3ad should interoperate well. The 802.3ad
1559 protocol includes automatic configuration of the aggregates,
1560 so minimal manual configuration of the switch is needed
1561 (typically only to designate that some set of devices is
1562 available for 802.3ad). The 802.3ad standard also mandates
1563 that frames be delivered in order (within certain limits), so
1564 in general single connections will not see misordering of
1565 packets. The 802.3ad mode does have some drawbacks: the
1566 standard mandates that all devices in the aggregate operate at
1567 the same speed and duplex. Also, as with all bonding load
1568 balance modes other than balance-rr, no single connection will
1569 be able to utilize more than a single interface's worth of
1572 Additionally, the linux bonding 802.3ad implementation
1573 distributes traffic by peer (using an XOR of MAC addresses),
1574 so in a "gatewayed" configuration, all outgoing traffic will
1575 generally use the same device. Incoming traffic may also end
1576 up on a single device, but that is dependent upon the
1577 balancing policy of the peer's 8023.ad implementation. In a
1578 "local" configuration, traffic will be distributed across the
1579 devices in the bond.
1581 Finally, the 802.3ad mode mandates the use of the MII monitor,
1582 therefore, the ARP monitor is not available in this mode.
1584 balance-tlb: The balance-tlb mode balances outgoing traffic by peer.
1585 Since the balancing is done according to MAC address, in a
1586 "gatewayed" configuration (as described above), this mode will
1587 send all traffic across a single device. However, in a
1588 "local" network configuration, this mode balances multiple
1589 local network peers across devices in a vaguely intelligent
1590 manner (not a simple XOR as in balance-xor or 802.3ad mode),
1591 so that mathematically unlucky MAC addresses (i.e., ones that
1592 XOR to the same value) will not all "bunch up" on a single
1595 Unlike 802.3ad, interfaces may be of differing speeds, and no
1596 special switch configuration is required. On the down side,
1597 in this mode all incoming traffic arrives over a single
1598 interface, this mode requires certain ethtool support in the
1599 network device driver of the slave interfaces, and the ARP
1600 monitor is not available.
1602 balance-alb: This mode is everything that balance-tlb is, and more.
1603 It has all of the features (and restrictions) of balance-tlb,
1604 and will also balance incoming traffic from local network
1605 peers (as described in the Bonding Module Options section,
1608 The only additional down side to this mode is that the network
1609 device driver must support changing the hardware address while
1612 13.1.2 MT Link Monitoring for Single Switch Topology
1613 ----------------------------------------------------
1615 The choice of link monitoring may largely depend upon which
1616 mode you choose to use. The more advanced load balancing modes do not
1617 support the use of the ARP monitor, and are thus restricted to using
1618 the MII monitor (which does not provide as high a level of end to end
1619 assurance as the ARP monitor).
1621 13.2 Maximum Throughput in a Multiple Switch Topology
1622 -----------------------------------------------------
1624 Multiple switches may be utilized to optimize for throughput
1625 when they are configured in parallel as part of an isolated network
1626 between two or more systems, for example:
1632 +--------+ | +---------+
1634 +------+---+ +-----+----+ +-----+----+
1635 | Switch A | | Switch B | | Switch C |
1636 +------+---+ +-----+----+ +-----+----+
1638 +--------+ | +---------+
1644 In this configuration, the switches are isolated from one
1645 another. One reason to employ a topology such as this is for an
1646 isolated network with many hosts (a cluster configured for high
1647 performance, for example), using multiple smaller switches can be more
1648 cost effective than a single larger switch, e.g., on a network with 24
1649 hosts, three 24 port switches can be significantly less expensive than
1650 a single 72 port switch.
1652 If access beyond the network is required, an individual host
1653 can be equipped with an additional network device connected to an
1654 external network; this host then additionally acts as a gateway.
1656 13.2.1 MT Bonding Mode Selection for Multiple Switch Topology
1657 -------------------------------------------------------------
1659 In actual practice, the bonding mode typically employed in
1660 configurations of this type is balance-rr. Historically, in this
1661 network configuration, the usual caveats about out of order packet
1662 delivery are mitigated by the use of network adapters that do not do
1663 any kind of packet coalescing (via the use of NAPI, or because the
1664 device itself does not generate interrupts until some number of
1665 packets has arrived). When employed in this fashion, the balance-rr
1666 mode allows individual connections between two hosts to effectively
1667 utilize greater than one interface's bandwidth.
1669 13.2.2 MT Link Monitoring for Multiple Switch Topology
1670 ------------------------------------------------------
1672 Again, in actual practice, the MII monitor is most often used
1673 in this configuration, as performance is given preference over
1674 availability. The ARP monitor will function in this topology, but its
1675 advantages over the MII monitor are mitigated by the volume of probes
1676 needed as the number of systems involved grows (remember that each
1677 host in the network is configured with bonding).
1679 14. Switch Behavior Issues
1680 ==========================
1682 14.1 Link Establishment and Failover Delays
1683 -------------------------------------------
1685 Some switches exhibit undesirable behavior with regard to the
1686 timing of link up and down reporting by the switch.
1688 First, when a link comes up, some switches may indicate that
1689 the link is up (carrier available), but not pass traffic over the
1690 interface for some period of time. This delay is typically due to
1691 some type of autonegotiation or routing protocol, but may also occur
1692 during switch initialization (e.g., during recovery after a switch
1693 failure). If you find this to be a problem, specify an appropriate
1694 value to the updelay bonding module option to delay the use of the
1695 relevant interface(s).
1697 Second, some switches may "bounce" the link state one or more
1698 times while a link is changing state. This occurs most commonly while
1699 the switch is initializing. Again, an appropriate updelay value may
1702 Note that when a bonding interface has no active links, the
1703 driver will immediately reuse the first link that goes up, even if the
1704 updelay parameter has been specified (the updelay is ignored in this
1705 case). If there are slave interfaces waiting for the updelay timeout
1706 to expire, the interface that first went into that state will be
1707 immediately reused. This reduces down time of the network if the
1708 value of updelay has been overestimated, and since this occurs only in
1709 cases with no connectivity, there is no additional penalty for
1710 ignoring the updelay.
1712 In addition to the concerns about switch timings, if your
1713 switches take a long time to go into backup mode, it may be desirable
1714 to not activate a backup interface immediately after a link goes down.
1715 Failover may be delayed via the downdelay bonding module option.
1717 14.2 Duplicated Incoming Packets
1718 --------------------------------
1720 It is not uncommon to observe a short burst of duplicated
1721 traffic when the bonding device is first used, or after it has been
1722 idle for some period of time. This is most easily observed by issuing
1723 a "ping" to some other host on the network, and noticing that the
1724 output from ping flags duplicates (typically one per slave).
1726 For example, on a bond in active-backup mode with five slaves
1727 all connected to one switch, the output may appear as follows:
1730 PING 10.0.4.2 (10.0.4.2) from 10.0.3.10 : 56(84) bytes of data.
1731 64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.7 ms
1732 64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
1733 64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
1734 64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
1735 64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
1736 64 bytes from 10.0.4.2: icmp_seq=2 ttl=64 time=0.216 ms
1737 64 bytes from 10.0.4.2: icmp_seq=3 ttl=64 time=0.267 ms
1738 64 bytes from 10.0.4.2: icmp_seq=4 ttl=64 time=0.222 ms
1740 This is not due to an error in the bonding driver, rather, it
1741 is a side effect of how many switches update their MAC forwarding
1742 tables. Initially, the switch does not associate the MAC address in
1743 the packet with a particular switch port, and so it may send the
1744 traffic to all ports until its MAC forwarding table is updated. Since
1745 the interfaces attached to the bond may occupy multiple ports on a
1746 single switch, when the switch (temporarily) floods the traffic to all
1747 ports, the bond device receives multiple copies of the same packet
1748 (one per slave device).
1750 The duplicated packet behavior is switch dependent, some
1751 switches exhibit this, and some do not. On switches that display this
1752 behavior, it can be induced by clearing the MAC forwarding table (on
1753 most Cisco switches, the privileged command "clear mac address-table
1754 dynamic" will accomplish this).
1756 15. Hardware Specific Considerations
1757 ====================================
1759 This section contains additional information for configuring
1760 bonding on specific hardware platforms, or for interfacing bonding
1761 with particular switches or other devices.
1763 15.1 IBM BladeCenter
1764 --------------------
1766 This applies to the JS20 and similar systems.
1768 On the JS20 blades, the bonding driver supports only
1769 balance-rr, active-backup, balance-tlb and balance-alb modes. This is
1770 largely due to the network topology inside the BladeCenter, detailed
1773 JS20 network adapter information
1774 --------------------------------
1776 All JS20s come with two Broadcom Gigabit Ethernet ports
1777 integrated on the planar (that's "motherboard" in IBM-speak). In the
1778 BladeCenter chassis, the eth0 port of all JS20 blades is hard wired to
1779 I/O Module #1; similarly, all eth1 ports are wired to I/O Module #2.
1780 An add-on Broadcom daughter card can be installed on a JS20 to provide
1781 two more Gigabit Ethernet ports. These ports, eth2 and eth3, are
1782 wired to I/O Modules 3 and 4, respectively.
1784 Each I/O Module may contain either a switch or a passthrough
1785 module (which allows ports to be directly connected to an external
1786 switch). Some bonding modes require a specific BladeCenter internal
1787 network topology in order to function; these are detailed below.
1789 Additional BladeCenter-specific networking information can be
1790 found in two IBM Redbooks (www.ibm.com/redbooks):
1792 "IBM eServer BladeCenter Networking Options"
1793 "IBM eServer BladeCenter Layer 2-7 Network Switching"
1795 BladeCenter networking configuration
1796 ------------------------------------
1798 Because a BladeCenter can be configured in a very large number
1799 of ways, this discussion will be confined to describing basic
1802 Normally, Ethernet Switch Modules (ESMs) are used in I/O
1803 modules 1 and 2. In this configuration, the eth0 and eth1 ports of a
1804 JS20 will be connected to different internal switches (in the
1805 respective I/O modules).
1807 A passthrough module (OPM or CPM, optical or copper,
1808 passthrough module) connects the I/O module directly to an external
1809 switch. By using PMs in I/O module #1 and #2, the eth0 and eth1
1810 interfaces of a JS20 can be redirected to the outside world and
1811 connected to a common external switch.
1813 Depending upon the mix of ESMs and PMs, the network will
1814 appear to bonding as either a single switch topology (all PMs) or as a
1815 multiple switch topology (one or more ESMs, zero or more PMs). It is
1816 also possible to connect ESMs together, resulting in a configuration
1817 much like the example in "High Availability in a Multiple Switch
1820 Requirements for specific modes
1821 -------------------------------
1823 The balance-rr mode requires the use of passthrough modules
1824 for devices in the bond, all connected to an common external switch.
1825 That switch must be configured for "etherchannel" or "trunking" on the
1826 appropriate ports, as is usual for balance-rr.
1828 The balance-alb and balance-tlb modes will function with
1829 either switch modules or passthrough modules (or a mix). The only
1830 specific requirement for these modes is that all network interfaces
1831 must be able to reach all destinations for traffic sent over the
1832 bonding device (i.e., the network must converge at some point outside
1835 The active-backup mode has no additional requirements.
1837 Link monitoring issues
1838 ----------------------
1840 When an Ethernet Switch Module is in place, only the ARP
1841 monitor will reliably detect link loss to an external switch. This is
1842 nothing unusual, but examination of the BladeCenter cabinet would
1843 suggest that the "external" network ports are the ethernet ports for
1844 the system, when it fact there is a switch between these "external"
1845 ports and the devices on the JS20 system itself. The MII monitor is
1846 only able to detect link failures between the ESM and the JS20 system.
1848 When a passthrough module is in place, the MII monitor does
1849 detect failures to the "external" port, which is then directly
1850 connected to the JS20 system.
1855 The Serial Over LAN (SoL) link is established over the primary
1856 ethernet (eth0) only, therefore, any loss of link to eth0 will result
1857 in losing your SoL connection. It will not fail over with other
1858 network traffic, as the SoL system is beyond the control of the
1861 It may be desirable to disable spanning tree on the switch
1862 (either the internal Ethernet Switch Module, or an external switch) to
1863 avoid fail-over delay issues when using bonding.
1866 16. Frequently Asked Questions
1867 ==============================
1871 Yes. The old 2.0.xx channel bonding patch was not SMP safe.
1872 The new driver was designed to be SMP safe from the start.
1874 2. What type of cards will work with it?
1876 Any Ethernet type cards (you can even mix cards - a Intel
1877 EtherExpress PRO/100 and a 3com 3c905b, for example). For most modes,
1878 devices need not be of the same speed.
1880 3. How many bonding devices can I have?
1884 4. How many slaves can a bonding device have?
1886 This is limited only by the number of network interfaces Linux
1887 supports and/or the number of network cards you can place in your
1890 5. What happens when a slave link dies?
1892 If link monitoring is enabled, then the failing device will be
1893 disabled. The active-backup mode will fail over to a backup link, and
1894 other modes will ignore the failed link. The link will continue to be
1895 monitored, and should it recover, it will rejoin the bond (in whatever
1896 manner is appropriate for the mode). See the sections on High
1897 Availability and the documentation for each mode for additional
1900 Link monitoring can be enabled via either the miimon or
1901 arp_interval parameters (described in the module parameters section,
1902 above). In general, miimon monitors the carrier state as sensed by
1903 the underlying network device, and the arp monitor (arp_interval)
1904 monitors connectivity to another host on the local network.
1906 If no link monitoring is configured, the bonding driver will
1907 be unable to detect link failures, and will assume that all links are
1908 always available. This will likely result in lost packets, and a
1909 resulting degradation of performance. The precise performance loss
1910 depends upon the bonding mode and network configuration.
1912 6. Can bonding be used for High Availability?
1914 Yes. See the section on High Availability for details.
1916 7. Which switches/systems does it work with?
1918 The full answer to this depends upon the desired mode.
1920 In the basic balance modes (balance-rr and balance-xor), it
1921 works with any system that supports etherchannel (also called
1922 trunking). Most managed switches currently available have such
1923 support, and many unmanaged switches as well.
1925 The advanced balance modes (balance-tlb and balance-alb) do
1926 not have special switch requirements, but do need device drivers that
1927 support specific features (described in the appropriate section under
1928 module parameters, above).
1930 In 802.3ad mode, it works with with systems that support IEEE
1931 802.3ad Dynamic Link Aggregation. Most managed and many unmanaged
1932 switches currently available support 802.3ad.
1934 The active-backup mode should work with any Layer-II switch.
1936 8. Where does a bonding device get its MAC address from?
1938 If not explicitly configured (with ifconfig or ip link), the
1939 MAC address of the bonding device is taken from its first slave
1940 device. This MAC address is then passed to all following slaves and
1941 remains persistent (even if the first slave is removed) until the
1942 bonding device is brought down or reconfigured.
1944 If you wish to change the MAC address, you can set it with
1945 ifconfig or ip link:
1947 # ifconfig bond0 hw ether 00:11:22:33:44:55
1949 # ip link set bond0 address 66:77:88:99:aa:bb
1951 The MAC address can be also changed by bringing down/up the
1952 device and then changing its slaves (or their order):
1954 # ifconfig bond0 down ; modprobe -r bonding
1955 # ifconfig bond0 .... up
1956 # ifenslave bond0 eth...
1958 This method will automatically take the address from the next
1959 slave that is added.
1961 To restore your slaves' MAC addresses, you need to detach them
1962 from the bond (`ifenslave -d bond0 eth0'). The bonding driver will
1963 then restore the MAC addresses that the slaves had before they were
1966 16. Resources and Links
1967 =======================
1969 The latest version of the bonding driver can be found in the latest
1970 version of the linux kernel, found on http://kernel.org
1972 The latest version of this document can be found in either the latest
1973 kernel source (named Documentation/networking/bonding.txt), or on the
1974 bonding sourceforge site:
1976 http://www.sourceforge.net/projects/bonding
1978 Discussions regarding the bonding driver take place primarily on the
1979 bonding-devel mailing list, hosted at sourceforge.net. If you have
1980 questions or problems, post them to the list. The list address is:
1982 bonding-devel@lists.sourceforge.net
1984 The administrative interface (to subscribe or unsubscribe) can
1987 https://lists.sourceforge.net/lists/listinfo/bonding-devel
1989 Donald Becker's Ethernet Drivers and diag programs may be found at :
1990 - http://www.scyld.com/network/
1992 You will also find a lot of information regarding Ethernet, NWay, MII,
1993 etc. at www.scyld.com.
git.openpandora.org / pandora-kernel.git / blob