2                Linux Ethernet Bonding Driver HOWTO
   4                Latest update: 27 April 2011
   6Initial release : Thomas Davis <tadavis at>
   7Corrections, HA extensions : 2000/10/03-15 :
   8  - Willy Tarreau <willy at>
   9  - Constantine Gavrilov <const-g at>
  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>
  14Reorganized and updated Feb 2005 by Jay Vosburgh
  15Added Sysfs information: 2006/04/24
  16  - Mitch Williams <mitch.a.williams at>
  21        The Linux bonding driver provides a method for aggregating
  22multiple network interfaces into a single logical "bonded" interface.
  23The behavior of the bonded interfaces depends upon the mode; generally
  24speaking, modes provide either hot standby or load balancing services.
  25Additionally, link integrity monitoring may be performed.
  27        The bonding driver originally came from Donald Becker's
  28beowulf patches for kernel 2.0. It has changed quite a bit since, and
  29the original tools from extreme-linux and beowulf sites will not work
  30with this version of the driver.
  32        For new versions of the driver, updated userspace tools, and
  33who to ask for help, please follow the links at the end of this file.
  35Table of Contents
  381. Bonding Driver Installation
  402. Bonding Driver Options
  423. Configuring Bonding Devices
  433.1     Configuration with Sysconfig Support
  443.1.1           Using DHCP with Sysconfig
  453.1.2           Configuring Multiple Bonds with Sysconfig
  463.2     Configuration with Initscripts Support
  473.2.1           Using DHCP with Initscripts
  483.2.2           Configuring Multiple Bonds with Initscripts
  493.3     Configuring Bonding Manually with Ifenslave
  503.3.1           Configuring Multiple Bonds Manually
  513.4     Configuring Bonding Manually via Sysfs
  523.5     Configuration with Interfaces Support
  533.6     Overriding Configuration for Special Cases
  554. Querying Bonding Configuration
  564.1     Bonding Configuration
  574.2     Network Configuration
  595. Switch Configuration
  616. 802.1q VLAN Support
  637. Link Monitoring
  647.1     ARP Monitor Operation
  657.2     Configuring Multiple ARP Targets
  667.3     MII Monitor Operation
  688. Potential Trouble Sources
  698.1     Adventures in Routing
  708.2     Ethernet Device Renaming
  718.3     Painfully Slow Or No Failed Link Detection By Miimon
  739. SNMP agents
  7510. Promiscuous mode
  7711. Configuring Bonding for High Availability
  7811.1    High Availability in a Single Switch Topology
  7911.2    High Availability in a Multiple Switch Topology
  8011.2.1          HA Bonding Mode Selection for Multiple Switch Topology
  8111.2.2          HA Link Monitoring for Multiple Switch Topology
  8312. Configuring Bonding for Maximum Throughput
  8412.1    Maximum Throughput in a Single Switch Topology
  8512.1.1          MT Bonding Mode Selection for Single Switch Topology
  8612.1.2          MT Link Monitoring for Single Switch Topology
  8712.2    Maximum Throughput in a Multiple Switch Topology
  8812.2.1          MT Bonding Mode Selection for Multiple Switch Topology
  8912.2.2          MT Link Monitoring for Multiple Switch Topology
  9113. Switch Behavior Issues
  9213.1    Link Establishment and Failover Delays
  9313.2    Duplicated Incoming Packets
  9514. Hardware Specific Considerations
  9614.1    IBM BladeCenter
  9815. Frequently Asked Questions
 10016. Resources and Links
 1031. Bonding Driver Installation
 106        Most popular distro kernels ship with the bonding driver
 107already available as a module and the ifenslave user level control
 108program installed and ready for use. If your distro does not, or you
 109have need to compile bonding from source (e.g., configuring and
 110installing a mainline kernel from, you'll need to perform
 111the following steps:
 1131.1 Configure and build the kernel with bonding
 116        The current version of the bonding driver is available in the
 117drivers/net/bonding subdirectory of the most recent kernel source
 118(which is available on  Most users "rolling their
 119own" will want to use the most recent kernel from
 121        Configure kernel with "make menuconfig" (or "make xconfig" or
 122"make config"), then select "Bonding driver support" in the "Network
 123device support" section.  It is recommended that you configure the
 124driver as module since it is currently the only way to pass parameters
 125to the driver or configure more than one bonding device.
 127        Build and install the new kernel and modules, then continue
 128below to install ifenslave.
 1301.2 Install ifenslave Control Utility
 133        The ifenslave user level control program is included in the
 134kernel source tree, in the file Documentation/networking/ifenslave.c.
 135It is generally recommended that you use the ifenslave that
 136corresponds to the kernel that you are using (either from the same
 137source tree or supplied with the distro), however, ifenslave
 138executables from older kernels should function (but features newer
 139than the ifenslave release are not supported).  Running an ifenslave
 140that is newer than the kernel is not supported, and may or may not
 143        To install ifenslave, do the following:
 145# gcc -Wall -O -I/usr/src/linux/include ifenslave.c -o ifenslave
 146# cp ifenslave /sbin/ifenslave
 148        If your kernel source is not in "/usr/src/linux," then replace
 149"/usr/src/linux/include" in the above with the location of your kernel
 150source include directory.
 152        You may wish to back up any existing /sbin/ifenslave, or, for
 153testing or informal use, tag the ifenslave to the kernel version
 154(e.g., name the ifenslave executable /sbin/ifenslave-2.6.10).
 158        If you omit the "-I" or specify an incorrect directory, you
 159may end up with an ifenslave that is incompatible with the kernel
 160you're trying to build it for.  Some distros (e.g., Red Hat from 7.1
 161onwards) do not have /usr/include/linux symbolically linked to the
 162default kernel source include directory.
 165        If you plan to configure bonding using sysfs or using the
 166/etc/network/interfaces file, you do not need to use ifenslave.
 1682. Bonding Driver Options
 171        Options for the bonding driver are supplied as parameters to the
 172bonding module at load time, or are specified via sysfs.
 174        Module options may be given as command line arguments to the
 175insmod or modprobe command, but are usually specified in either the
 176/etc/modrobe.d/*.conf configuration files, or in a distro-specific
 177configuration file (some of which are detailed in the next section).
 179        Details on bonding support for sysfs is provided in the
 180"Configuring Bonding Manually via Sysfs" section, below.
 182        The available bonding driver parameters are listed below. If a
 183parameter is not specified the default value is used.  When initially
 184configuring a bond, it is recommended "tail -f /var/log/messages" be
 185run in a separate window to watch for bonding driver error messages.
 187        It is critical that either the miimon or arp_interval and
 188arp_ip_target parameters be specified, otherwise serious network
 189degradation will occur during link failures.  Very few devices do not
 190support at least miimon, so there is really no reason not to use it.
 192        Options with textual values will accept either the text name
 193or, for backwards compatibility, the option value.  E.g.,
 194"mode=802.3ad" and "mode=4" set the same mode.
 196        The parameters are as follows:
 200        Specifies the new active slave for modes that support it
 201        (active-backup, balance-alb and balance-tlb).  Possible values
 202        are the name of any currently enslaved interface, or an empty
 203        string.  If a name is given, the slave and its link must be up in order
 204        to be selected as the new active slave.  If an empty string is
 205        specified, the current active slave is cleared, and a new active
 206        slave is selected automatically.
 208        Note that this is only available through the sysfs interface. No module
 209        parameter by this name exists.
 211        The normal value of this option is the name of the currently
 212        active slave, or the empty string if there is no active slave or
 213        the current mode does not use an active slave.
 217        Specifies the 802.3ad aggregation selection logic to use.  The
 218        possible values and their effects are:
 220        stable or 0
 222                The active aggregator is chosen by largest aggregate
 223                bandwidth.
 225                Reselection of the active aggregator occurs only when all
 226                slaves of the active aggregator are down or the active
 227                aggregator has no slaves.
 229                This is the default value.
 231        bandwidth or 1
 233                The active aggregator is chosen by largest aggregate
 234                bandwidth.  Reselection occurs if:
 236                - A slave is added to or removed from the bond
 238                - Any slave's link state changes
 240                - Any slave's 802.3ad association state changes
 242                - The bond's administrative state changes to up
 244        count or 2
 246                The active aggregator is chosen by the largest number of
 247                ports (slaves).  Reselection occurs as described under the
 248                "bandwidth" setting, above.
 250        The bandwidth and count selection policies permit failover of
 251        802.3ad aggregations when partial failure of the active aggregator
 252        occurs.  This keeps the aggregator with the highest availability
 253        (either in bandwidth or in number of ports) active at all times.
 255        This option was added in bonding version 3.4.0.
 259        Specifies that duplicate frames (received on inactive ports) should be
 260        dropped (0) or delivered (1).
 262        Normally, bonding will drop duplicate frames (received on inactive
 263        ports), which is desirable for most users. But there are some times
 264        it is nice to allow duplicate frames to be delivered.
 266        The default value is 0 (drop duplicate frames received on inactive
 267        ports).
 271        Specifies the ARP link monitoring frequency in milliseconds.
 273        The ARP monitor works by periodically checking the slave
 274        devices to determine whether they have sent or received
 275        traffic recently (the precise criteria depends upon the
 276        bonding mode, and the state of the slave).  Regular traffic is
 277        generated via ARP probes issued for the addresses specified by
 278        the arp_ip_target option.
 280        This behavior can be modified by the arp_validate option,
 281        below.
 283        If ARP monitoring is used in an etherchannel compatible mode
 284        (modes 0 and 2), the switch should be configured in a mode
 285        that evenly distributes packets across all links. If the
 286        switch is configured to distribute the packets in an XOR
 287        fashion, all replies from the ARP targets will be received on
 288        the same link which could cause the other team members to
 289        fail.  ARP monitoring should not be used in conjunction with
 290        miimon.  A value of 0 disables ARP monitoring.  The default
 291        value is 0.
 295        Specifies the IP addresses to use as ARP monitoring peers when
 296        arp_interval is > 0.  These are the targets of the ARP request
 297        sent to determine the health of the link to the targets.
 298        Specify these values in ddd.ddd.ddd.ddd format.  Multiple IP
 299        addresses must be separated by a comma.  At least one IP
 300        address must be given for ARP monitoring to function.  The
 301        maximum number of targets that can be specified is 16.  The
 302        default value is no IP addresses.
 306        Specifies whether or not ARP probes and replies should be
 307        validated in the active-backup mode.  This causes the ARP
 308        monitor to examine the incoming ARP requests and replies, and
 309        only consider a slave to be up if it is receiving the
 310        appropriate ARP traffic.
 312        Possible values are:
 314        none or 0
 316                No validation is performed.  This is the default.
 318        active or 1
 320                Validation is performed only for the active slave.
 322        backup or 2
 324                Validation is performed only for backup slaves.
 326        all or 3
 328                Validation is performed for all slaves.
 330        For the active slave, the validation checks ARP replies to
 331        confirm that they were generated by an arp_ip_target.  Since
 332        backup slaves do not typically receive these replies, the
 333        validation performed for backup slaves is on the ARP request
 334        sent out via the active slave.  It is possible that some
 335        switch or network configurations may result in situations
 336        wherein the backup slaves do not receive the ARP requests; in
 337        such a situation, validation of backup slaves must be
 338        disabled.
 340        This option is useful in network configurations in which
 341        multiple bonding hosts are concurrently issuing ARPs to one or
 342        more targets beyond a common switch.  Should the link between
 343        the switch and target fail (but not the switch itself), the
 344        probe traffic generated by the multiple bonding instances will
 345        fool the standard ARP monitor into considering the links as
 346        still up.  Use of the arp_validate option can resolve this, as
 347        the ARP monitor will only consider ARP requests and replies
 348        associated with its own instance of bonding.
 350        This option was added in bonding version 3.1.0.
 354        Specifies the time, in milliseconds, to wait before disabling
 355        a slave after a link failure has been detected.  This option
 356        is only valid for the miimon link monitor.  The downdelay
 357        value should be a multiple of the miimon value; if not, it
 358        will be rounded down to the nearest multiple.  The default
 359        value is 0.
 363        Specifies whether active-backup mode should set all slaves to
 364        the same MAC address at enslavement (the traditional
 365        behavior), or, when enabled, perform special handling of the
 366        bond's MAC address in accordance with the selected policy.
 368        Possible values are:
 370        none or 0
 372                This setting disables fail_over_mac, and causes
 373                bonding to set all slaves of an active-backup bond to
 374                the same MAC address at enslavement time.  This is the
 375                default.
 377        active or 1
 379                The "active" fail_over_mac policy indicates that the
 380                MAC address of the bond should always be the MAC
 381                address of the currently active slave.  The MAC
 382                address of the slaves is not changed; instead, the MAC
 383                address of the bond changes during a failover.
 385                This policy is useful for devices that cannot ever
 386                alter their MAC address, or for devices that refuse
 387                incoming broadcasts with their own source MAC (which
 388                interferes with the ARP monitor).
 390                The down side of this policy is that every device on
 391                the network must be updated via gratuitous ARP,
 392                vs. just updating a switch or set of switches (which
 393                often takes place for any traffic, not just ARP
 394                traffic, if the switch snoops incoming traffic to
 395                update its tables) for the traditional method.  If the
 396                gratuitous ARP is lost, communication may be
 397                disrupted.
 399                When this policy is used in conjunction with the mii
 400                monitor, devices which assert link up prior to being
 401                able to actually transmit and receive are particularly
 402                susceptible to loss of the gratuitous ARP, and an
 403                appropriate updelay setting may be required.
 405        follow or 2
 407                The "follow" fail_over_mac policy causes the MAC
 408                address of the bond to be selected normally (normally
 409                the MAC address of the first slave added to the bond).
 410                However, the second and subsequent slaves are not set
 411                to this MAC address while they are in a backup role; a
 412                slave is programmed with the bond's MAC address at
 413                failover time (and the formerly active slave receives
 414                the newly active slave's MAC address).
 416                This policy is useful for multiport devices that
 417                either become confused or incur a performance penalty
 418                when multiple ports are programmed with the same MAC
 419                address.
 422        The default policy is none, unless the first slave cannot
 423        change its MAC address, in which case the active policy is
 424        selected by default.
 426        This option may be modified via sysfs only when no slaves are
 427        present in the bond.
 429        This option was added in bonding version 3.2.0.  The "follow"
 430        policy was added in bonding version 3.3.0.
 434        Option specifying the rate in which we'll ask our link partner
 435        to transmit LACPDU packets in 802.3ad mode.  Possible values
 436        are:
 438        slow or 0
 439                Request partner to transmit LACPDUs every 30 seconds
 441        fast or 1
 442                Request partner to transmit LACPDUs every 1 second
 444        The default is slow.
 448        Specifies the number of bonding devices to create for this
 449        instance of the bonding driver.  E.g., if max_bonds is 3, and
 450        the bonding driver is not already loaded, then bond0, bond1
 451        and bond2 will be created.  The default value is 1.  Specifying
 452        a value of 0 will load bonding, but will not create any devices.
 456        Specifies the MII link monitoring frequency in milliseconds.
 457        This determines how often the link state of each slave is
 458        inspected for link failures.  A value of zero disables MII
 459        link monitoring.  A value of 100 is a good starting point.
 460        The use_carrier option, below, affects how the link state is
 461        determined.  See the High Availability section for additional
 462        information.  The default value is 0.
 466        Specifies the minimum number of links that must be active before
 467        asserting carrier. It is similar to the Cisco EtherChannel min-links
 468        feature. This allows setting the minimum number of member ports that
 469        must be up (link-up state) before marking the bond device as up
 470        (carrier on). This is useful for situations where higher level services
 471        such as clustering want to ensure a minimum number of low bandwidth
 472        links are active before switchover. This option only affect 802.3ad
 473        mode.
 475        The default value is 0. This will cause carrier to be asserted (for
 476        802.3ad mode) whenever there is an active aggregator, regardless of the
 477        number of available links in that aggregator. Note that, because an
 478        aggregator cannot be active without at least one available link,
 479        setting this option to 0 or to 1 has the exact same effect.
 483        Specifies one of the bonding policies. The default is
 484        balance-rr (round robin).  Possible values are:
 486        balance-rr or 0
 488                Round-robin policy: Transmit packets in sequential
 489                order from the first available slave through the
 490                last.  This mode provides load balancing and fault
 491                tolerance.
 493        active-backup or 1
 495                Active-backup policy: Only one slave in the bond is
 496                active.  A different slave becomes active if, and only
 497                if, the active slave fails.  The bond's MAC address is
 498                externally visible on only one port (network adapter)
 499                to avoid confusing the switch.
 501                In bonding version 2.6.2 or later, when a failover
 502                occurs in active-backup mode, bonding will issue one
 503                or more gratuitous ARPs on the newly active slave.
 504                One gratuitous ARP is issued for the bonding master
 505                interface and each VLAN interfaces configured above
 506                it, provided that the interface has at least one IP
 507                address configured.  Gratuitous ARPs issued for VLAN
 508                interfaces are tagged with the appropriate VLAN id.
 510                This mode provides fault tolerance.  The primary
 511                option, documented below, affects the behavior of this
 512                mode.
 514        balance-xor or 2
 516                XOR policy: Transmit based on the selected transmit
 517                hash policy.  The default policy is a simple [(source
 518                MAC address XOR'd with destination MAC address) modulo
 519                slave count].  Alternate transmit policies may be
 520                selected via the xmit_hash_policy option, described
 521                below.
 523                This mode provides load balancing and fault tolerance.
 525        broadcast or 3
 527                Broadcast policy: transmits everything on all slave
 528                interfaces.  This mode provides fault tolerance.
 530        802.3ad or 4
 532                IEEE 802.3ad Dynamic link aggregation.  Creates
 533                aggregation groups that share the same speed and
 534                duplex settings.  Utilizes all slaves in the active
 535                aggregator according to the 802.3ad specification.
 537                Slave selection for outgoing traffic is done according
 538                to the transmit hash policy, which may be changed from
 539                the default simple XOR policy via the xmit_hash_policy
 540                option, documented below.  Note that not all transmit
 541                policies may be 802.3ad compliant, particularly in
 542                regards to the packet mis-ordering requirements of
 543                section 43.2.4 of the 802.3ad standard.  Differing
 544                peer implementations will have varying tolerances for
 545                noncompliance.
 547                Prerequisites:
 549                1. Ethtool support in the base drivers for retrieving
 550                the speed and duplex of each slave.
 552                2. A switch that supports IEEE 802.3ad Dynamic link
 553                aggregation.
 555                Most switches will require some type of configuration
 556                to enable 802.3ad mode.
 558        balance-tlb or 5
 560                Adaptive transmit load balancing: channel bonding that
 561                does not require any special switch support.  The
 562                outgoing traffic is distributed according to the
 563                current load (computed relative to the speed) on each
 564                slave.  Incoming traffic is received by the current
 565                slave.  If the receiving slave fails, another slave
 566                takes over the MAC address of the failed receiving
 567                slave.
 569                Prerequisite:
 571                Ethtool support in the base drivers for retrieving the
 572                speed of each slave.
 574        balance-alb or 6
 576                Adaptive load balancing: includes balance-tlb plus
 577                receive load balancing (rlb) for IPV4 traffic, and
 578                does not require any special switch support.  The
 579                receive load balancing is achieved by ARP negotiation.
 580                The bonding driver intercepts the ARP Replies sent by
 581                the local system on their way out and overwrites the
 582                source hardware address with the unique hardware
 583                address of one of the slaves in the bond such that
 584                different peers use different hardware addresses for
 585                the server.
 587                Receive traffic from connections created by the server
 588                is also balanced.  When the local system sends an ARP
 589                Request the bonding driver copies and saves the peer's
 590                IP information from the ARP packet.  When the ARP
 591                Reply arrives from the peer, its hardware address is
 592                retrieved and the bonding driver initiates an ARP
 593                reply to this peer assigning it to one of the slaves
 594                in the bond.  A problematic outcome of using ARP
 595                negotiation for balancing is that each time that an
 596                ARP request is broadcast it uses the hardware address
 597                of the bond.  Hence, peers learn the hardware address
 598                of the bond and the balancing of receive traffic
 599                collapses to the current slave.  This is handled by
 600                sending updates (ARP Replies) to all the peers with
 601                their individually assigned hardware address such that
 602                the traffic is redistributed.  Receive traffic is also
 603                redistributed when a new slave is added to the bond
 604                and when an inactive slave is re-activated.  The
 605                receive load is distributed sequentially (round robin)
 606                among the group of highest speed slaves in the bond.
 608                When a link is reconnected or a new slave joins the
 609                bond the receive traffic is redistributed among all
 610                active slaves in the bond by initiating ARP Replies
 611                with the selected MAC address to each of the
 612                clients. The updelay parameter (detailed below) must
 613                be set to a value equal or greater than the switch's
 614                forwarding delay so that the ARP Replies sent to the
 615                peers will not be blocked by the switch.
 617                Prerequisites:
 619                1. Ethtool support in the base drivers for retrieving
 620                the speed of each slave.
 622                2. Base driver support for setting the hardware
 623                address of a device while it is open.  This is
 624                required so that there will always be one slave in the
 625                team using the bond hardware address (the
 626                curr_active_slave) while having a unique hardware
 627                address for each slave in the bond.  If the
 628                curr_active_slave fails its hardware address is
 629                swapped with the new curr_active_slave that was
 630                chosen.
 635        Specify the number of peer notifications (gratuitous ARPs and
 636        unsolicited IPv6 Neighbor Advertisements) to be issued after a
 637        failover event.  As soon as the link is up on the new slave
 638        (possibly immediately) a peer notification is sent on the
 639        bonding device and each VLAN sub-device.  This is repeated at
 640        each link monitor interval (arp_interval or miimon, whichever
 641        is active) if the number is greater than 1.
 643        The valid range is 0 - 255; the default value is 1.  These options
 644        affect only the active-backup mode.  These options were added for
 645        bonding versions 3.3.0 and 3.4.0 respectively.
 647        From Linux 3.0 and bonding version 3.7.1, these notifications
 648        are generated by the ipv4 and ipv6 code and the numbers of
 649        repetitions cannot be set independently.
 653        A string (eth0, eth2, etc) specifying which slave is the
 654        primary device.  The specified device will always be the
 655        active slave while it is available.  Only when the primary is
 656        off-line will alternate devices be used.  This is useful when
 657        one slave is preferred over another, e.g., when one slave has
 658        higher throughput than another.
 660        The primary option is only valid for active-backup mode.
 664        Specifies the reselection policy for the primary slave.  This
 665        affects how the primary slave is chosen to become the active slave
 666        when failure of the active slave or recovery of the primary slave
 667        occurs.  This option is designed to prevent flip-flopping between
 668        the primary slave and other slaves.  Possible values are:
 670        always or 0 (default)
 672                The primary slave becomes the active slave whenever it
 673                comes back up.
 675        better or 1
 677                The primary slave becomes the active slave when it comes
 678                back up, if the speed and duplex of the primary slave is
 679                better than the speed and duplex of the current active
 680                slave.
 682        failure or 2
 684                The primary slave becomes the active slave only if the
 685                current active slave fails and the primary slave is up.
 687        The primary_reselect setting is ignored in two cases:
 689                If no slaves are active, the first slave to recover is
 690                made the active slave.
 692                When initially enslaved, the primary slave is always made
 693                the active slave.
 695        Changing the primary_reselect policy via sysfs will cause an
 696        immediate selection of the best active slave according to the new
 697        policy.  This may or may not result in a change of the active
 698        slave, depending upon the circumstances.
 700        This option was added for bonding version 3.6.0.
 704        Specifies the time, in milliseconds, to wait before enabling a
 705        slave after a link recovery has been detected.  This option is
 706        only valid for the miimon link monitor.  The updelay value
 707        should be a multiple of the miimon value; if not, it will be
 708        rounded down to the nearest multiple.  The default value is 0.
 712        Specifies whether or not miimon should use MII or ETHTOOL
 713        ioctls vs. netif_carrier_ok() to determine the link
 714        status. The MII or ETHTOOL ioctls are less efficient and
 715        utilize a deprecated calling sequence within the kernel.  The
 716        netif_carrier_ok() relies on the device driver to maintain its
 717        state with netif_carrier_on/off; at this writing, most, but
 718        not all, device drivers support this facility.
 720        If bonding insists that the link is up when it should not be,
 721        it may be that your network device driver does not support
 722        netif_carrier_on/off.  The default state for netif_carrier is
 723        "carrier on," so if a driver does not support netif_carrier,
 724        it will appear as if the link is always up.  In this case,
 725        setting use_carrier to 0 will cause bonding to revert to the
 726        MII / ETHTOOL ioctl method to determine the link state.
 728        A value of 1 enables the use of netif_carrier_ok(), a value of
 729        0 will use the deprecated MII / ETHTOOL ioctls.  The default
 730        value is 1.
 734        Selects the transmit hash policy to use for slave selection in
 735        balance-xor and 802.3ad modes.  Possible values are:
 737        layer2
 739                Uses XOR of hardware MAC addresses to generate the
 740                hash.  The formula is
 742                (source MAC XOR destination MAC) modulo slave count
 744                This algorithm will place all traffic to a particular
 745                network peer on the same slave.
 747                This algorithm is 802.3ad compliant.
 749        layer2+3
 751                This policy uses a combination of layer2 and layer3
 752                protocol information to generate the hash.
 754                Uses XOR of hardware MAC addresses and IP addresses to
 755                generate the hash.  The formula is
 757                (((source IP XOR dest IP) AND 0xffff) XOR
 758                        ( source MAC XOR destination MAC ))
 759                                modulo slave count
 761                This algorithm will place all traffic to a particular
 762                network peer on the same slave.  For non-IP traffic,
 763                the formula is the same as for the layer2 transmit
 764                hash policy.
 766                This policy is intended to provide a more balanced
 767                distribution of traffic than layer2 alone, especially
 768                in environments where a layer3 gateway device is
 769                required to reach most destinations.
 771                This algorithm is 802.3ad compliant.
 773        layer3+4
 775                This policy uses upper layer protocol information,
 776                when available, to generate the hash.  This allows for
 777                traffic to a particular network peer to span multiple
 778                slaves, although a single connection will not span
 779                multiple slaves.
 781                The formula for unfragmented TCP and UDP packets is
 783                ((source port XOR dest port) XOR
 784                         ((source IP XOR dest IP) AND 0xffff)
 785                                modulo slave count
 787                For fragmented TCP or UDP packets and all other IP
 788                protocol traffic, the source and destination port
 789                information is omitted.  For non-IP traffic, the
 790                formula is the same as for the layer2 transmit hash
 791                policy.
 793                This policy is intended to mimic the behavior of
 794                certain switches, notably Cisco switches with PFC2 as
 795                well as some Foundry and IBM products.
 797                This algorithm is not fully 802.3ad compliant.  A
 798                single TCP or UDP conversation containing both
 799                fragmented and unfragmented packets will see packets
 800                striped across two interfaces.  This may result in out
 801                of order delivery.  Most traffic types will not meet
 802                this criteria, as TCP rarely fragments traffic, and
 803                most UDP traffic is not involved in extended
 804                conversations.  Other implementations of 802.3ad may
 805                or may not tolerate this noncompliance.
 807        The default value is layer2.  This option was added in bonding
 808        version 2.6.3.  In earlier versions of bonding, this parameter
 809        does not exist, and the layer2 policy is the only policy.  The
 810        layer2+3 value was added for bonding version 3.2.2.
 814        Specifies the number of IGMP membership reports to be issued after
 815        a failover event. One membership report is issued immediately after
 816        the failover, subsequent packets are sent in each 200ms interval.
 818        The valid range is 0 - 255; the default value is 1. A value of 0
 819        prevents the IGMP membership report from being issued in response
 820        to the failover event.
 822        This option is useful for bonding modes balance-rr (0), active-backup
 823        (1), balance-tlb (5) and balance-alb (6), in which a failover can
 824        switch the IGMP traffic from one slave to another.  Therefore a fresh
 825        IGMP report must be issued to cause the switch to forward the incoming
 826        IGMP traffic over the newly selected slave.
 828        This option was added for bonding version 3.7.0.
 8303. Configuring Bonding Devices
 833        You can configure bonding using either your distro's network
 834initialization scripts, or manually using either ifenslave or the
 835sysfs interface.  Distros generally use one of three packages for the
 836network initialization scripts: initscripts, sysconfig or interfaces.
 837Recent versions of these packages have support for bonding, while older
 838versions do not.
 840        We will first describe the options for configuring bonding for
 841distros using versions of initscripts, sysconfig and interfaces with full
 842or partial support for bonding, then provide information on enabling
 843bonding without support from the network initialization scripts (i.e.,
 844older versions of initscripts or sysconfig).
 846        If you're unsure whether your distro uses sysconfig,
 847initscripts or interfaces, or don't know if it's new enough, have no fear.
 848Determining this is fairly straightforward.
 850        First, look for a file called interfaces in /etc/network directory.
 851If this file is present in your system, then your system use interfaces. See
 852Configuration with Interfaces Support.
 854        Else, issue the command:
 856$ rpm -qf /sbin/ifup
 858        It will respond with a line of text starting with either
 859"initscripts" or "sysconfig," followed by some numbers.  This is the
 860package that provides your network initialization scripts.
 862        Next, to determine if your installation supports bonding,
 863issue the command:
 865$ grep ifenslave /sbin/ifup
 867        If this returns any matches, then your initscripts or
 868sysconfig has support for bonding.
 8703.1 Configuration with Sysconfig Support
 873        This section applies to distros using a version of sysconfig
 874with bonding support, for example, SuSE Linux Enterprise Server 9.
 876        SuSE SLES 9's networking configuration system does support
 877bonding, however, at this writing, the YaST system configuration
 878front end does not provide any means to work with bonding devices.
 879Bonding devices can be managed by hand, however, as follows.
 881        First, if they have not already been configured, configure the
 882slave devices.  On SLES 9, this is most easily done by running the
 883yast2 sysconfig configuration utility.  The goal is for to create an
 884ifcfg-id file for each slave device.  The simplest way to accomplish
 885this is to configure the devices for DHCP (this is only to get the
 886file ifcfg-id file created; see below for some issues with DHCP).  The
 887name of the configuration file for each device will be of the form:
 891        Where the "xx" portion will be replaced with the digits from
 892the device's permanent MAC address.
 894        Once the set of ifcfg-id-xx:xx:xx:xx:xx:xx files has been
 895created, it is necessary to edit the configuration files for the slave
 896devices (the MAC addresses correspond to those of the slave devices).
 897Before editing, the file will contain multiple lines, and will look
 898something like this:
 906        Change the BOOTPROTO and STARTMODE lines to the following:
 911        Do not alter the UNIQUE or _nm_name lines.  Remove any other
 912lines (USERCTL, etc).
 914        Once the ifcfg-id-xx:xx:xx:xx:xx:xx files have been modified,
 915it's time to create the configuration file for the bonding device
 916itself.  This file is named ifcfg-bondX, where X is the number of the
 917bonding device to create, starting at 0.  The first such file is
 918ifcfg-bond0, the second is ifcfg-bond1, and so on.  The sysconfig
 919network configuration system will correctly start multiple instances
 920of bonding.
 922        The contents of the ifcfg-bondX file is as follows:
 932BONDING_MODULE_OPTS="mode=active-backup miimon=100"
 936        Replace the sample BROADCAST, IPADDR, NETMASK and NETWORK
 937values with the appropriate values for your network.
 939        The STARTMODE specifies when the device is brought online.
 940The possible values are:
 942        onboot:  The device is started at boot time.  If you're not
 943                 sure, this is probably what you want.
 945        manual:  The device is started only when ifup is called
 946                 manually.  Bonding devices may be configured this
 947                 way if you do not wish them to start automatically
 948                 at boot for some reason.
 950        hotplug: The device is started by a hotplug event.  This is not
 951                 a valid choice for a bonding device.
 953        off or ignore: The device configuration is ignored.
 955        The line BONDING_MASTER='yes' indicates that the device is a
 956bonding master device.  The only useful value is "yes."
 958        The contents of BONDING_MODULE_OPTS are supplied to the
 959instance of the bonding module for this device.  Specify the options
 960for the bonding mode, link monitoring, and so on here.  Do not include
 961the max_bonds bonding parameter; this will confuse the configuration
 962system if you have multiple bonding devices.
 964        Finally, supply one BONDING_SLAVEn="slave device" for each
 965slave.  where "n" is an increasing value, one for each slave.  The
 966"slave device" is either an interface name, e.g., "eth0", or a device
 967specifier for the network device.  The interface name is easier to
 968find, but the ethN names are subject to change at boot time if, e.g.,
 969a device early in the sequence has failed.  The device specifiers
 970(bus-pci-0000:06:08.1 in the example above) specify the physical
 971network device, and will not change unless the device's bus location
 972changes (for example, it is moved from one PCI slot to another).  The
 973example above uses one of each type for demonstration purposes; most
 974configurations will choose one or the other for all slave devices.
 976        When all configuration files have been modified or created,
 977networking must be restarted for the configuration changes to take
 978effect.  This can be accomplished via the following:
 980# /etc/init.d/network restart
 982        Note that the network control script (/sbin/ifdown) will
 983remove the bonding module as part of the network shutdown processing,
 984so it is not necessary to remove the module by hand if, e.g., the
 985module parameters have changed.
 987        Also, at this writing, YaST/YaST2 will not manage bonding
 988devices (they do not show bonding interfaces on its list of network
 989devices).  It is necessary to edit the configuration file by hand to
 990change the bonding configuration.
 992        Additional general options and details of the ifcfg file
 993format can be found in an example ifcfg template file:
 997        Note that the template does not document the various BONDING_
 998settings described above, but does describe many of the other options.
10003.1.1 Using DHCP with Sysconfig
1003        Under sysconfig, configuring a device with BOOTPROTO='dhcp'
1004will cause it to query DHCP for its IP address information.  At this
1005writing, this does not function for bonding devices; the scripts
1006attempt to obtain the device address from DHCP prior to adding any of
1007the slave devices.  Without active slaves, the DHCP requests are not
1008sent to the network.
10103.1.2 Configuring Multiple Bonds with Sysconfig
1013        The sysconfig network initialization system is capable of
1014handling multiple bonding devices.  All that is necessary is for each
1015bonding instance to have an appropriately configured ifcfg-bondX file
1016(as described above).  Do not specify the "max_bonds" parameter to any
1017instance of bonding, as this will confuse sysconfig.  If you require
1018multiple bonding devices with identical parameters, create multiple
1019ifcfg-bondX files.
1021        Because the sysconfig scripts supply the bonding module
1022options in the ifcfg-bondX file, it is not necessary to add them to
1023the system /etc/modules.d/*.conf configuration files.
10253.2 Configuration with Initscripts Support
1028        This section applies to distros using a recent version of
1029initscripts with bonding support, for example, Red Hat Enterprise Linux
1030version 3 or later, Fedora, etc.  On these systems, the network
1031initialization scripts have knowledge of bonding, and can be configured to
1032control bonding devices.  Note that older versions of the initscripts
1033package have lower levels of support for bonding; this will be noted where
1036        These distros will not automatically load the network adapter
1037driver unless the ethX device is configured with an IP address.
1038Because of this constraint, users must manually configure a
1039network-script file for all physical adapters that will be members of
1040a bondX link.  Network script files are located in the directory:
1044        The file name must be prefixed with "ifcfg-eth" and suffixed
1045with the adapter's physical adapter number.  For example, the script
1046for eth0 would be named /etc/sysconfig/network-scripts/ifcfg-eth0.
1047Place the following text in the file:
1056        The DEVICE= line will be different for every ethX device and
1057must correspond with the name of the file, i.e., ifcfg-eth1 must have
1058a device line of DEVICE=eth1.  The setting of the MASTER= line will
1059also depend on the final bonding interface name chosen for your bond.
1060As with other network devices, these typically start at 0, and go up
1061one for each device, i.e., the first bonding instance is bond0, the
1062second is bond1, and so on.
1064        Next, create a bond network script.  The file name for this
1065script will be /etc/sysconfig/network-scripts/ifcfg-bondX where X is
1066the number of the bond.  For bond0 the file is named "ifcfg-bond0",
1067for bond1 it is named "ifcfg-bond1", and so on.  Within that file,
1068place the following text:
1079        Be sure to change the networking specific lines (IPADDR,
1080NETMASK, NETWORK and BROADCAST) to match your network configuration.
1082        For later versions of initscripts, such as that found with Fedora
10837 (or later) and Red Hat Enterprise Linux version 5 (or later), it is possible,
1084and, indeed, preferable, to specify the bonding options in the ifcfg-bond0
1085file, e.g. a line of the format:
1087BONDING_OPTS="mode=active-backup arp_interval=60 arp_ip_target="
1089        will configure the bond with the specified options.  The options
1090specified in BONDING_OPTS are identical to the bonding module parameters
1091except for the arp_ip_target field when using versions of initscripts older
1092than and 8.57 (Fedora 8) and 8.45.19 (Red Hat Enterprise Linux 5.2).  When
1093using older versions each target should be included as a separate option and
1094should be preceded by a '+' to indicate it should be added to the list of
1095queried targets, e.g.,
1097        arp_ip_target=+ arp_ip_target=+
1099        is the proper syntax to specify multiple targets.  When specifying
1100options via BONDING_OPTS, it is not necessary to edit /etc/modprobe.d/*.conf.
1102        For even older versions of initscripts that do not support
1103BONDING_OPTS, it is necessary to edit /etc/modprobe.d/*.conf, depending upon
1104your distro) to load the bonding module with your desired options when the
1105bond0 interface is brought up.  The following lines in /etc/modprobe.d/*.conf
1106will load the bonding module, and select its options:
1108alias bond0 bonding
1109options bond0 mode=balance-alb miimon=100
1111        Replace the sample parameters with the appropriate set of
1112options for your configuration.
1114        Finally run "/etc/rc.d/init.d/network restart" as root.  This
1115will restart the networking subsystem and your bond link should be now
1116up and running.
11183.2.1 Using DHCP with Initscripts
1121        Recent versions of initscripts (the versions supplied with Fedora
1122Core 3 and Red Hat Enterprise Linux 4, or later versions, are reported to
1123work) have support for assigning IP information to bonding devices via
1126        To configure bonding for DHCP, configure it as described
1127above, except replace the line "BOOTPROTO=none" with "BOOTPROTO=dhcp"
1128and add a line consisting of "TYPE=Bonding".  Note that the TYPE value
1129is case sensitive.
11313.2.2 Configuring Multiple Bonds with Initscripts
1134        Initscripts packages that are included with Fedora 7 and Red Hat
1135Enterprise Linux 5 support multiple bonding interfaces by simply
1136specifying the appropriate BONDING_OPTS= in ifcfg-bondX where X is the
1137number of the bond.  This support requires sysfs support in the kernel,
1138and a bonding driver of version 3.0.0 or later.  Other configurations may
1139not support this method for specifying multiple bonding interfaces; for
1140those instances, see the "Configuring Multiple Bonds Manually" section,
11433.3 Configuring Bonding Manually with Ifenslave
1146        This section applies to distros whose network initialization
1147scripts (the sysconfig or initscripts package) do not have specific
1148knowledge of bonding.  One such distro is SuSE Linux Enterprise Server
1149version 8.
1151        The general method for these systems is to place the bonding
1152module parameters into a config file in /etc/modprobe.d/ (as
1153appropriate for the installed distro), then add modprobe and/or
1154ifenslave commands to the system's global init script.  The name of
1155the global init script differs; for sysconfig, it is
1156/etc/init.d/boot.local and for initscripts it is /etc/rc.d/rc.local.
1158        For example, if you wanted to make a simple bond of two e100
1159devices (presumed to be eth0 and eth1), and have it persist across
1160reboots, edit the appropriate file (/etc/init.d/boot.local or
1161/etc/rc.d/rc.local), and add the following:
1163modprobe bonding mode=balance-alb miimon=100
1164modprobe e100
1165ifconfig bond0 netmask up
1166ifenslave bond0 eth0
1167ifenslave bond0 eth1
1169        Replace the example bonding module parameters and bond0
1170network configuration (IP address, netmask, etc) with the appropriate
1171values for your configuration.
1173        Unfortunately, this method will not provide support for the
1174ifup and ifdown scripts on the bond devices.  To reload the bonding
1175configuration, it is necessary to run the initialization script, e.g.,
1177# /etc/init.d/boot.local
1179        or
1181# /etc/rc.d/rc.local
1183        It may be desirable in such a case to create a separate script
1184which only initializes the bonding configuration, then call that
1185separate script from within boot.local.  This allows for bonding to be
1186enabled without re-running the entire global init script.
1188        To shut down the bonding devices, it is necessary to first
1189mark the bonding device itself as being down, then remove the
1190appropriate device driver modules.  For our example above, you can do
1191the following:
1193# ifconfig bond0 down
1194# rmmod bonding
1195# rmmod e100
1197        Again, for convenience, it may be desirable to create a script
1198with these commands.
12013.3.1 Configuring Multiple Bonds Manually
1204        This section contains information on configuring multiple
1205bonding devices with differing options for those systems whose network
1206initialization scripts lack support for configuring multiple bonds.
1208        If you require multiple bonding devices, but all with the same
1209options, you may wish to use the "max_bonds" module parameter,
1210documented above.
1212        To create multiple bonding devices with differing options, it is
1213preferable to use bonding parameters exported by sysfs, documented in the
1214section below.
1216        For versions of bonding without sysfs support, the only means to
1217provide multiple instances of bonding with differing options is to load
1218the bonding driver multiple times.  Note that current versions of the
1219sysconfig network initialization scripts handle this automatically; if
1220your distro uses these scripts, no special action is needed.  See the
1221section Configuring Bonding Devices, above, if you're not sure about your
1222network initialization scripts.
1224        To load multiple instances of the module, it is necessary to
1225specify a different name for each instance (the module loading system
1226requires that every loaded module, even multiple instances of the same
1227module, have a unique name).  This is accomplished by supplying multiple
1228sets of bonding options in /etc/modprobe.d/*.conf, for example:
1230alias bond0 bonding
1231options bond0 -o bond0 mode=balance-rr miimon=100
1233alias bond1 bonding
1234options bond1 -o bond1 mode=balance-alb miimon=50
1236        will load the bonding module two times.  The first instance is
1237named "bond0" and creates the bond0 device in balance-rr mode with an
1238miimon of 100.  The second instance is named "bond1" and creates the
1239bond1 device in balance-alb mode with an miimon of 50.
1241        In some circumstances (typically with older distributions),
1242the above does not work, and the second bonding instance never sees
1243its options.  In that case, the second options line can be substituted
1244as follows:
1246install bond1 /sbin/modprobe --ignore-install bonding -o bond1 \
1247        mode=balance-alb miimon=50
1249        This may be repeated any number of times, specifying a new and
1250unique name in place of bond1 for each subsequent instance.
1252        It has been observed that some Red Hat supplied kernels are unable
1253to rename modules at load time (the "-o bond1" part).  Attempts to pass
1254that option to modprobe will produce an "Operation not permitted" error.
1255This has been reported on some Fedora Core kernels, and has been seen on
1256RHEL 4 as well.  On kernels exhibiting this problem, it will be impossible
1257to configure multiple bonds with differing parameters (as they are older
1258kernels, and also lack sysfs support).
12603.4 Configuring Bonding Manually via Sysfs
1263        Starting with version 3.0.0, Channel Bonding may be configured
1264via the sysfs interface.  This interface allows dynamic configuration
1265of all bonds in the system without unloading the module.  It also
1266allows for adding and removing bonds at runtime.  Ifenslave is no
1267longer required, though it is still supported.
1269        Use of the sysfs interface allows you to use multiple bonds
1270with different configurations without having to reload the module.
1271It also allows you to use multiple, differently configured bonds when
1272bonding is compiled into the kernel.
1274        You must have the sysfs filesystem mounted to configure
1275bonding this way.  The examples in this document assume that you
1276are using the standard mount point for sysfs, e.g. /sys.  If your
1277sysfs filesystem is mounted elsewhere, you will need to adjust the
1278example paths accordingly.
1280Creating and Destroying Bonds
1282To add a new bond foo:
1283# echo +foo > /sys/class/net/bonding_masters
1285To remove an existing bond bar:
1286# echo -bar > /sys/class/net/bonding_masters
1288To show all existing bonds:
1289# cat /sys/class/net/bonding_masters
1291NOTE: due to 4K size limitation of sysfs files, this list may be
1292truncated if you have more than a few hundred bonds.  This is unlikely
1293to occur under normal operating conditions.
1295Adding and Removing Slaves
1297        Interfaces may be enslaved to a bond using the file
1298/sys/class/net/<bond>/bonding/slaves.  The semantics for this file
1299are the same as for the bonding_masters file.
1301To enslave interface eth0 to bond bond0:
1302# ifconfig bond0 up
1303# echo +eth0 > /sys/class/net/bond0/bonding/slaves
1305To free slave eth0 from bond bond0:
1306# echo -eth0 > /sys/class/net/bond0/bonding/slaves
1308        When an interface is enslaved to a bond, symlinks between the
1309two are created in the sysfs filesystem.  In this case, you would get
1310/sys/class/net/bond0/slave_eth0 pointing to /sys/class/net/eth0, and
1311/sys/class/net/eth0/master pointing to /sys/class/net/bond0.
1313        This means that you can tell quickly whether or not an
1314interface is enslaved by looking for the master symlink.  Thus:
1315# echo -eth0 > /sys/class/net/eth0/master/bonding/slaves
1316will free eth0 from whatever bond it is enslaved to, regardless of
1317the name of the bond interface.
1319Changing a Bond's Configuration
1321        Each bond may be configured individually by manipulating the
1322files located in /sys/class/net/<bond name>/bonding
1324        The names of these files correspond directly with the command-
1325line parameters described elsewhere in this file, and, with the
1326exception of arp_ip_target, they accept the same values.  To see the
1327current setting, simply cat the appropriate file.
1329        A few examples will be given here; for specific usage
1330guidelines for each parameter, see the appropriate section in this
1333To configure bond0 for balance-alb mode:
1334# ifconfig bond0 down
1335# echo 6 > /sys/class/net/bond0/bonding/mode
1336 - or -
1337# echo balance-alb > /sys/class/net/bond0/bonding/mode
1338        NOTE: The bond interface must be down before the mode can be
1341To enable MII monitoring on bond0 with a 1 second interval:
1342# echo 1000 > /sys/class/net/bond0/bonding/miimon
1343        NOTE: If ARP monitoring is enabled, it will disabled when MII
1344monitoring is enabled, and vice-versa.
1346To add ARP targets:
1347# echo + > /sys/class/net/bond0/bonding/arp_ip_target
1348# echo + > /sys/class/net/bond0/bonding/arp_ip_target
1349        NOTE:  up to 16 target addresses may be specified.
1351To remove an ARP target:
1352# echo - > /sys/class/net/bond0/bonding/arp_ip_target
1354Example Configuration
1356        We begin with the same example that is shown in section 3.3,
1357executed with sysfs, and without using ifenslave.
1359        To make a simple bond of two e100 devices (presumed to be eth0
1360and eth1), and have it persist across reboots, edit the appropriate
1361file (/etc/init.d/boot.local or /etc/rc.d/rc.local), and add the
1364modprobe bonding
1365modprobe e100
1366echo balance-alb > /sys/class/net/bond0/bonding/mode
1367ifconfig bond0 netmask up
1368echo 100 > /sys/class/net/bond0/bonding/miimon
1369echo +eth0 > /sys/class/net/bond0/bonding/slaves
1370echo +eth1 > /sys/class/net/bond0/bonding/slaves
1372        To add a second bond, with two e1000 interfaces in
1373active-backup mode, using ARP monitoring, add the following lines to
1374your init script:
1376modprobe e1000
1377echo +bond1 > /sys/class/net/bonding_masters
1378echo active-backup > /sys/class/net/bond1/bonding/mode
1379ifconfig bond1 netmask up
1380echo + /sys/class/net/bond1/bonding/arp_ip_target
1381echo 2000 > /sys/class/net/bond1/bonding/arp_interval
1382echo +eth2 > /sys/class/net/bond1/bonding/slaves
1383echo +eth3 > /sys/class/net/bond1/bonding/slaves
13853.5 Configuration with Interfaces Support
1388        This section applies to distros which use /etc/network/interfaces file
1389to describe network interface configuration, most notably Debian and it's
1392        The ifup and ifdown commands on Debian don't support bonding out of
1393the box. The ifenslave-2.6 package should be installed to provide bonding
1394support.  Once installed, this package will provide bond-* options to be used
1395into /etc/network/interfaces.
1397        Note that ifenslave-2.6 package will load the bonding module and use
1398the ifenslave command when appropriate.
1400Example Configurations
1403In /etc/network/interfaces, the following stanza will configure bond0, in
1404active-backup mode, with eth0 and eth1 as slaves.
1406auto bond0
1407iface bond0 inet dhcp
1408        bond-slaves eth0 eth1
1409        bond-mode active-backup
1410        bond-miimon 100
1411        bond-primary eth0 eth1
1413If the above configuration doesn't work, you might have a system using
1414upstart for system startup. This is most notably true for recent
1415Ubuntu versions. The following stanza in /etc/network/interfaces will
1416produce the same result on those systems.
1418auto bond0
1419iface bond0 inet dhcp
1420        bond-slaves none
1421        bond-mode active-backup
1422        bond-miimon 100
1424auto eth0
1425iface eth0 inet manual
1426        bond-master bond0
1427        bond-primary eth0 eth1
1429auto eth1
1430iface eth1 inet manual
1431        bond-master bond0
1432        bond-primary eth0 eth1
1434For a full list of bond-* supported options in /etc/network/interfaces and some
1435more advanced examples tailored to you particular distros, see the files in
14383.6 Overriding Configuration for Special Cases
1441When using the bonding driver, the physical port which transmits a frame is
1442typically selected by the bonding driver, and is not relevant to the user or
1443system administrator.  The output port is simply selected using the policies of
1444the selected bonding mode.  On occasion however, it is helpful to direct certain
1445classes of traffic to certain physical interfaces on output to implement
1446slightly more complex policies.  For example, to reach a web server over a
1447bonded interface in which eth0 connects to a private network, while eth1
1448connects via a public network, it may be desirous to bias the bond to send said
1449traffic over eth0 first, using eth1 only as a fall back, while all other traffic
1450can safely be sent over either interface.  Such configurations may be achieved
1451using the traffic control utilities inherent in linux.
1453By default the bonding driver is multiqueue aware and 16 queues are created
1454when the driver initializes (see Documentation/networking/multiqueue.txt
1455for details).  If more or less queues are desired the module parameter
1456tx_queues can be used to change this value.  There is no sysfs parameter
1457available as the allocation is done at module init time.
1459The output of the file /proc/net/bonding/bondX has changed so the output Queue
1460ID is now printed for each slave:
1462Bonding Mode: fault-tolerance (active-backup)
1463Primary Slave: None
1464Currently Active Slave: eth0
1465MII Status: up
1466MII Polling Interval (ms): 0
1467Up Delay (ms): 0
1468Down Delay (ms): 0
1470Slave Interface: eth0
1471MII Status: up
1472Link Failure Count: 0
1473Permanent HW addr: 00:1a:a0:12:8f:cb
1474Slave queue ID: 0
1476Slave Interface: eth1
1477MII Status: up
1478Link Failure Count: 0
1479Permanent HW addr: 00:1a:a0:12:8f:cc
1480Slave queue ID: 2
1482The queue_id for a slave can be set using the command:
1484# echo "eth1:2" > /sys/class/net/bond0/bonding/queue_id
1486Any interface that needs a queue_id set should set it with multiple calls
1487like the one above until proper priorities are set for all interfaces.  On
1488distributions that allow configuration via initscripts, multiple 'queue_id'
1489arguments can be added to BONDING_OPTS to set all needed slave queues.
1491These queue id's can be used in conjunction with the tc utility to configure
1492a multiqueue qdisc and filters to bias certain traffic to transmit on certain
1493slave devices.  For instance, say we wanted, in the above configuration to
1494force all traffic bound to to use eth1 in the bond as its output
1495device. The following commands would accomplish this:
1497# tc qdisc add dev bond0 handle 1 root multiq
1499# tc filter add dev bond0 protocol ip parent 1: prio 1 u32 match ip dst \
1500 action skbedit queue_mapping 2
1502These commands tell the kernel to attach a multiqueue queue discipline to the
1503bond0 interface and filter traffic enqueued to it, such that packets with a dst
1504ip of have their output queue mapping value overwritten to 2.
1505This value is then passed into the driver, causing the normal output path
1506selection policy to be overridden, selecting instead qid 2, which maps to eth1.
1508Note that qid values begin at 1.  Qid 0 is reserved to initiate to the driver
1509that normal output policy selection should take place.  One benefit to simply
1510leaving the qid for a slave to 0 is the multiqueue awareness in the bonding
1511driver that is now present.  This awareness allows tc filters to be placed on
1512slave devices as well as bond devices and the bonding driver will simply act as
1513a pass-through for selecting output queues on the slave device rather than 
1514output port selection.
1516This feature first appeared in bonding driver version 3.7.0 and support for
1517output slave selection was limited to round-robin and active-backup modes.
15194 Querying Bonding Configuration
15224.1 Bonding Configuration
1525        Each bonding device has a read-only file residing in the
1526/proc/net/bonding directory.  The file contents include information
1527about the bonding configuration, options and state of each slave.
1529        For example, the contents of /proc/net/bonding/bond0 after the
1530driver is loaded with parameters of mode=0 and miimon=1000 is
1531generally as follows:
1533        Ethernet Channel Bonding Driver: 2.6.1 (October 29, 2004)
1534        Bonding Mode: load balancing (round-robin)
1535        Currently Active Slave: eth0
1536        MII Status: up
1537        MII Polling Interval (ms): 1000
1538        Up Delay (ms): 0
1539        Down Delay (ms): 0
1541        Slave Interface: eth1
1542        MII Status: up
1543        Link Failure Count: 1
1545        Slave Interface: eth0
1546        MII Status: up
1547        Link Failure Count: 1
1549        The precise format and contents will change depending upon the
1550bonding configuration, state, and version of the bonding driver.
15524.2 Network configuration
1555        The network configuration can be inspected using the ifconfig
1556command.  Bonding devices will have the MASTER flag set; Bonding slave
1557devices will have the SLAVE flag set.  The ifconfig output does not
1558contain information on which slaves are associated with which masters.
1560        In the example below, the bond0 interface is the master
1561(MASTER) while eth0 and eth1 are slaves (SLAVE). Notice all slaves of
1562bond0 have the same MAC address (HWaddr) as bond0 for all modes except
1563TLB and ALB that require a unique MAC address for each slave.
1565# /sbin/ifconfig
1566bond0     Link encap:Ethernet  HWaddr 00:C0:F0:1F:37:B4
1567          inet addr:XXX.XXX.XXX.YYY  Bcast:XXX.XXX.XXX.255  Mask:
1569          RX packets:7224794 errors:0 dropped:0 overruns:0 frame:0
1570          TX packets:3286647 errors:1 dropped:0 overruns:1 carrier:0
1571          collisions:0 txqueuelen:0
1573eth0      Link encap:Ethernet  HWaddr 00:C0:F0:1F:37:B4
1575          RX packets:3573025 errors:0 dropped:0 overruns:0 frame:0
1576          TX packets:1643167 errors:1 dropped:0 overruns:1 carrier:0
1577          collisions:0 txqueuelen:100
1578          Interrupt:10 Base address:0x1080
1580eth1      Link encap:Ethernet  HWaddr 00:C0:F0:1F:37:B4
1582          RX packets:3651769 errors:0 dropped:0 overruns:0 frame:0
1583          TX packets:1643480 errors:0 dropped:0 overruns:0 carrier:0
1584          collisions:0 txqueuelen:100
1585          Interrupt:9 Base address:0x1400
15875. Switch Configuration
1590        For this section, "switch" refers to whatever system the
1591bonded devices are directly connected to (i.e., where the other end of
1592the cable plugs into).  This may be an actual dedicated switch device,
1593or it may be another regular system (e.g., another computer running
1596        The active-backup, balance-tlb and balance-alb modes do not
1597require any specific configuration of the switch.
1599        The 802.3ad mode requires that the switch have the appropriate
1600ports configured as an 802.3ad aggregation.  The precise method used
1601to configure this varies from switch to switch, but, for example, a
1602Cisco 3550 series switch requires that the appropriate ports first be
1603grouped together in a single etherchannel instance, then that
1604etherchannel is set to mode "lacp" to enable 802.3ad (instead of
1605standard EtherChannel).
1607        The balance-rr, balance-xor and broadcast modes generally
1608require that the switch have the appropriate ports grouped together.
1609The nomenclature for such a group differs between switches, it may be
1610called an "etherchannel" (as in the Cisco example, above), a "trunk
1611group" or some other similar variation.  For these modes, each switch
1612will also have its own configuration options for the switch's transmit
1613policy to the bond.  Typical choices include XOR of either the MAC or
1614IP addresses.  The transmit policy of the two peers does not need to
1615match.  For these three modes, the bonding mode really selects a
1616transmit policy for an EtherChannel group; all three will interoperate
1617with another EtherChannel group.
16206. 802.1q VLAN Support
1623        It is possible to configure VLAN devices over a bond interface
1624using the 8021q driver.  However, only packets coming from the 8021q
1625driver and passing through bonding will be tagged by default.  Self
1626generated packets, for example, bonding's learning packets or ARP
1627packets generated by either ALB mode or the ARP monitor mechanism, are
1628tagged internally by bonding itself.  As a result, bonding must
1629"learn" the VLAN IDs configured above it, and use those IDs to tag
1630self generated packets.
1632        For reasons of simplicity, and to support the use of adapters
1633that can do VLAN hardware acceleration offloading, the bonding
1634interface declares itself as fully hardware offloading capable, it gets
1635the add_vid/kill_vid notifications to gather the necessary
1636information, and it propagates those actions to the slaves.  In case
1637of mixed adapter types, hardware accelerated tagged packets that
1638should go through an adapter that is not offloading capable are
1639"un-accelerated" by the bonding driver so the VLAN tag sits in the
1640regular location.
1642        VLAN interfaces *must* be added on top of a bonding interface
1643only after enslaving at least one slave.  The bonding interface has a
1644hardware address of 00:00:00:00:00:00 until the first slave is added.
1645If the VLAN interface is created prior to the first enslavement, it
1646would pick up the all-zeroes hardware address.  Once the first slave
1647is attached to the bond, the bond device itself will pick up the
1648slave's hardware address, which is then available for the VLAN device.
1650        Also, be aware that a similar problem can occur if all slaves
1651are released from a bond that still has one or more VLAN interfaces on
1652top of it.  When a new slave is added, the bonding interface will
1653obtain its hardware address from the first slave, which might not
1654match the hardware address of the VLAN interfaces (which was
1655ultimately copied from an earlier slave).
1657        There are two methods to insure that the VLAN device operates
1658with the correct hardware address if all slaves are removed from a
1659bond interface:
1661        1. Remove all VLAN interfaces then recreate them
1663        2. Set the bonding interface's hardware address so that it
1664matches the hardware address of the VLAN interfaces.
1666        Note that changing a VLAN interface's HW address would set the
1667underlying device -- i.e. the bonding interface -- to promiscuous
1668mode, which might not be what you want.
16717. Link Monitoring
1674        The bonding driver at present supports two schemes for
1675monitoring a slave device's link state: the ARP monitor and the MII
1678        At the present time, due to implementation restrictions in the
1679bonding driver itself, it is not possible to enable both ARP and MII
1680monitoring simultaneously.
16827.1 ARP Monitor Operation
1685        The ARP monitor operates as its name suggests: it sends ARP
1686queries to one or more designated peer systems on the network, and
1687uses the response as an indication that the link is operating.  This
1688gives some assurance that traffic is actually flowing to and from one
1689or more peers on the local network.
1691        The ARP monitor relies on the device driver itself to verify
1692that traffic is flowing.  In particular, the driver must keep up to
1693date the last receive time, dev->last_rx, and transmit start time,
1694dev->trans_start.  If these are not updated by the driver, then the
1695ARP monitor will immediately fail any slaves using that driver, and
1696those slaves will stay down.  If networking monitoring (tcpdump, etc)
1697shows the ARP requests and replies on the network, then it may be that
1698your device driver is not updating last_rx and trans_start.
17007.2 Configuring Multiple ARP Targets
1703        While ARP monitoring can be done with just one target, it can
1704be useful in a High Availability setup to have several targets to
1705monitor.  In the case of just one target, the target itself may go
1706down or have a problem making it unresponsive to ARP requests.  Having
1707an additional target (or several) increases the reliability of the ARP
1710        Multiple ARP targets must be separated by commas as follows:
1712# example options for ARP monitoring with three targets
1713alias bond0 bonding
1714options bond0 arp_interval=60 arp_ip_target=,,
1716        For just a single target the options would resemble:
1718# example options for ARP monitoring with one target
1719alias bond0 bonding
1720options bond0 arp_interval=60 arp_ip_target=
17237.3 MII Monitor Operation
1726        The MII monitor monitors only the carrier state of the local
1727network interface.  It accomplishes this in one of three ways: by
1728depending upon the device driver to maintain its carrier state, by
1729querying the device's MII registers, or by making an ethtool query to
1730the device.
1732        If the use_carrier module parameter is 1 (the default value),
1733then the MII monitor will rely on the driver for carrier state
1734information (via the netif_carrier subsystem).  As explained in the
1735use_carrier parameter information, above, if the MII monitor fails to
1736detect carrier loss on the device (e.g., when the cable is physically
1737disconnected), it may be that the driver does not support
1740        If use_carrier is 0, then the MII monitor will first query the
1741device's (via ioctl) MII registers and check the link state.  If that
1742request fails (not just that it returns carrier down), then the MII
1743monitor will make an ethtool ETHOOL_GLINK request to attempt to obtain
1744the same information.  If both methods fail (i.e., the driver either
1745does not support or had some error in processing both the MII register
1746and ethtool requests), then the MII monitor will assume the link is
17498. Potential Sources of Trouble
17528.1 Adventures in Routing
1755        When bonding is configured, it is important that the slave
1756devices not have routes that supersede routes of the master (or,
1757generally, not have routes at all).  For example, suppose the bonding
1758device bond0 has two slaves, eth0 and eth1, and the routing table is
1759as follows:
1761Kernel IP routing table
1762Destination     Gateway         Genmask         Flags   MSS Window  irtt Iface
176310.0.0.0     U        40 0          0 eth0
176410.0.0.0     U        40 0          0 eth1
176510.0.0.0     U        40 0          0 bond0
1766127.0.0.0       U        40 0          0 lo
1768        This routing configuration will likely still update the
1769receive/transmit times in the driver (needed by the ARP monitor), but
1770may bypass the bonding driver (because outgoing traffic to, in this
1771case, another host on network 10 would use eth0 or eth1 before bond0).
1773        The ARP monitor (and ARP itself) may become confused by this
1774configuration, because ARP requests (generated by the ARP monitor)
1775will be sent on one interface (bond0), but the corresponding reply
1776will arrive on a different interface (eth0).  This reply looks to ARP
1777as an unsolicited ARP reply (because ARP matches replies on an
1778interface basis), and is discarded.  The MII monitor is not affected
1779by the state of the routing table.
1781        The solution here is simply to insure that slaves do not have
1782routes of their own, and if for some reason they must, those routes do
1783not supersede routes of their master.  This should generally be the
1784case, but unusual configurations or errant manual or automatic static
1785route additions may cause trouble.
17878.2 Ethernet Device Renaming
1790        On systems with network configuration scripts that do not
1791associate physical devices directly with network interface names (so
1792that the same physical device always has the same "ethX" name), it may
1793be necessary to add some special logic to config files in
1796        For example, given a modules.conf containing the following:
1798alias bond0 bonding
1799options bond0 mode=some-mode miimon=50
1800alias eth0 tg3
1801alias eth1 tg3
1802alias eth2 e1000
1803alias eth3 e1000
1805        If neither eth0 and eth1 are slaves to bond0, then when the
1806bond0 interface comes up, the devices may end up reordered.  This
1807happens because bonding is loaded first, then its slave device's
1808drivers are loaded next.  Since no other drivers have been loaded,
1809when the e1000 driver loads, it will receive eth0 and eth1 for its
1810devices, but the bonding configuration tries to enslave eth2 and eth3
1811(which may later be assigned to the tg3 devices).
1813        Adding the following:
1815add above bonding e1000 tg3
1817        causes modprobe to load e1000 then tg3, in that order, when
1818bonding is loaded.  This command is fully documented in the
1819modules.conf manual page.
1821        On systems utilizing modprobe an equivalent problem can occur.
1822In this case, the following can be added to config files in
1823/etc/modprobe.d/ as:
1825softdep bonding pre: tg3 e1000
1827        This will load tg3 and e1000 modules before loading the bonding one.
1828Full documentation on this can be found in the modprobe.d and modprobe
1829manual pages.
18318.3. Painfully Slow Or No Failed Link Detection By Miimon
1834        By default, bonding enables the use_carrier option, which
1835instructs bonding to trust the driver to maintain carrier state.
1837        As discussed in the options section, above, some drivers do
1838not support the netif_carrier_on/_off link state tracking system.
1839With use_carrier enabled, bonding will always see these links as up,
1840regardless of their actual state.
1842        Additionally, other drivers do support netif_carrier, but do
1843not maintain it in real time, e.g., only polling the link state at
1844some fixed interval.  In this case, miimon will detect failures, but
1845only after some long period of time has expired.  If it appears that
1846miimon is very slow in detecting link failures, try specifying
1847use_carrier=0 to see if that improves the failure detection time.  If
1848it does, then it may be that the driver checks the carrier state at a
1849fixed interval, but does not cache the MII register values (so the
1850use_carrier=0 method of querying the registers directly works).  If
1851use_carrier=0 does not improve the failover, then the driver may cache
1852the registers, or the problem may be elsewhere.
1854        Also, remember that miimon only checks for the device's
1855carrier state.  It has no way to determine the state of devices on or
1856beyond other ports of a switch, or if a switch is refusing to pass
1857traffic while still maintaining carrier on.
18599. SNMP agents
1862        If running SNMP agents, the bonding driver should be loaded
1863before any network drivers participating in a bond.  This requirement
1864is due to the interface index (ipAdEntIfIndex) being associated to
1865the first interface found with a given IP address.  That is, there is
1866only one ipAdEntIfIndex for each IP address.  For example, if eth0 and
1867eth1 are slaves of bond0 and the driver for eth0 is loaded before the
1868bonding driver, the interface for the IP address will be associated
1869with the eth0 interface.  This configuration is shown below, the IP
1870address has an interface index of 2 which indexes to eth0
1871in the ifDescr table (ifDescr.2).
1873     interfaces.ifTable.ifEntry.ifDescr.1 = lo
1874     interfaces.ifTable.ifEntry.ifDescr.2 = eth0
1875     interfaces.ifTable.ifEntry.ifDescr.3 = eth1
1876     interfaces.ifTable.ifEntry.ifDescr.4 = eth2
1877     interfaces.ifTable.ifEntry.ifDescr.5 = eth3
1878     interfaces.ifTable.ifEntry.ifDescr.6 = bond0
1879     ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex. = 5
1880     ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex. = 2
1881     ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex. = 4
1882     ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex. = 1
1884        This problem is avoided by loading the bonding driver before
1885any network drivers participating in a bond.  Below is an example of
1886loading the bonding driver first, the IP address is
1887correctly associated with ifDescr.2.
1889     interfaces.ifTable.ifEntry.ifDescr.1 = lo
1890     interfaces.ifTable.ifEntry.ifDescr.2 = bond0
1891     interfaces.ifTable.ifEntry.ifDescr.3 = eth0
1892     interfaces.ifTable.ifEntry.ifDescr.4 = eth1
1893     interfaces.ifTable.ifEntry.ifDescr.5 = eth2
1894     interfaces.ifTable.ifEntry.ifDescr.6 = eth3
1895     ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex. = 6
1896     ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex. = 2
1897     ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex. = 5
1898     ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex. = 1
1900        While some distributions may not report the interface name in
1901ifDescr, the association between the IP address and IfIndex remains
1902and SNMP functions such as Interface_Scan_Next will report that
190510. Promiscuous mode
1908        When running network monitoring tools, e.g., tcpdump, it is
1909common to enable promiscuous mode on the device, so that all traffic
1910is seen (instead of seeing only traffic destined for the local host).
1911The bonding driver handles promiscuous mode changes to the bonding
1912master device (e.g., bond0), and propagates the setting to the slave
1915        For the balance-rr, balance-xor, broadcast, and 802.3ad modes,
1916the promiscuous mode setting is propagated to all slaves.
1918        For the active-backup, balance-tlb and balance-alb modes, the
1919promiscuous mode setting is propagated only to the active slave.
1921        For balance-tlb mode, the active slave is the slave currently
1922receiving inbound traffic.
1924        For balance-alb mode, the active slave is the slave used as a
1925"primary."  This slave is used for mode-specific control traffic, for
1926sending to peers that are unassigned or if the load is unbalanced.
1928        For the active-backup, balance-tlb and balance-alb modes, when
1929the active slave changes (e.g., due to a link failure), the
1930promiscuous setting will be propagated to the new active slave.
193211. Configuring Bonding for High Availability
1935        High Availability refers to configurations that provide
1936maximum network availability by having redundant or backup devices,
1937links or switches between the host and the rest of the world.  The
1938goal is to provide the maximum availability of network connectivity
1939(i.e., the network always works), even though other configurations
1940could provide higher throughput.
194211.1 High Availability in a Single Switch Topology
1945        If two hosts (or a host and a single switch) are directly
1946connected via multiple physical links, then there is no availability
1947penalty to optimizing for maximum bandwidth.  In this case, there is
1948only one switch (or peer), so if it fails, there is no alternative
1949access to fail over to.  Additionally, the bonding load balance modes
1950support link monitoring of their members, so if individual links fail,
1951the load will be rebalanced across the remaining devices.
1953        See Section 12, "Configuring Bonding for Maximum Throughput"
1954for information on configuring bonding with one peer device.
195611.2 High Availability in a Multiple Switch Topology
1959        With multiple switches, the configuration of bonding and the
1960network changes dramatically.  In multiple switch topologies, there is
1961a trade off between network availability and usable bandwidth.
1963        Below is a sample network, configured to maximize the
1964availability of the network:
1966                |                                     |
1967                |port3                           port3|
1968          +-----+----+                          +-----+----+
1969          |          |port2       ISL      port2|          |
1970          | switch A +--------------------------+ switch B |
1971          |          |                          |          |
1972          +-----+----+                          +-----++---+
1973                |port1                           port1|
1974                |             +-------+               |
1975                +-------------+ host1 +---------------+
1976                         eth0 +-------+ eth1
1978        In this configuration, there is a link between the two
1979switches (ISL, or inter switch link), and multiple ports connecting to
1980the outside world ("port3" on each switch).  There is no technical
1981reason that this could not be extended to a third switch.
198311.2.1 HA Bonding Mode Selection for Multiple Switch Topology
1986        In a topology such as the example above, the active-backup and
1987broadcast modes are the only useful bonding modes when optimizing for
1988availability; the other modes require all links to terminate on the
1989same peer for them to behave rationally.
1991active-backup: This is generally the preferred mode, particularly if
1992        the switches have an ISL and play together well.  If the
1993        network configuration is such that one switch is specifically
1994        a backup switch (e.g., has lower capacity, higher cost, etc),
1995        then the primary option can be used to insure that the
1996        preferred link is always used when it is available.
1998broadcast: This mode is really a special purpose mode, and is suitable
1999        only for very specific needs.  For example, if the two
2000        switches are not connected (no ISL), and the networks beyond
2001        them are totally independent.  In this case, if it is
2002        necessary for some specific one-way traffic to reach both
2003        independent networks, then the broadcast mode may be suitable.
200511.2.2 HA Link Monitoring Selection for Multiple Switch Topology
2008        The choice of link monitoring ultimately depends upon your
2009switch.  If the switch can reliably fail ports in response to other
2010failures, then either the MII or ARP monitors should work.  For
2011example, in the above example, if the "port3" link fails at the remote
2012end, the MII monitor has no direct means to detect this.  The ARP
2013monitor could be configured with a target at the remote end of port3,
2014thus detecting that failure without switch support.
2016        In general, however, in a multiple switch topology, the ARP
2017monitor can provide a higher level of reliability in detecting end to
2018end connectivity failures (which may be caused by the failure of any
2019individual component to pass traffic for any reason).  Additionally,
2020the ARP monitor should be configured with multiple targets (at least
2021one for each switch in the network).  This will insure that,
2022regardless of which switch is active, the ARP monitor has a suitable
2023target to query.
2025        Note, also, that of late many switches now support a functionality
2026generally referred to as "trunk failover."  This is a feature of the
2027switch that causes the link state of a particular switch port to be set
2028down (or up) when the state of another switch port goes down (or up).
2029Its purpose is to propagate link failures from logically "exterior" ports
2030to the logically "interior" ports that bonding is able to monitor via
2031miimon.  Availability and configuration for trunk failover varies by
2032switch, but this can be a viable alternative to the ARP monitor when using
2033suitable switches.
203512. Configuring Bonding for Maximum Throughput
203812.1 Maximizing Throughput in a Single Switch Topology
2041        In a single switch configuration, the best method to maximize
2042throughput depends upon the application and network environment.  The
2043various load balancing modes each have strengths and weaknesses in
2044different environments, as detailed below.
2046        For this discussion, we will break down the topologies into
2047two categories.  Depending upon the destination of most traffic, we
2048categorize them into either "gatewayed" or "local" configurations.
2050        In a gatewayed configuration, the "switch" is acting primarily
2051as a router, and the majority of traffic passes through this router to
2052other networks.  An example would be the following:
2055     +----------+                     +----------+
2056     |          |eth0            port1|          | to other networks
2057     | Host A   +---------------------+ router   +------------------->
2058     |          +---------------------+          | Hosts B and C are out
2059     |          |eth1            port2|          | here somewhere
2060     +----------+                     +----------+
2062        The router may be a dedicated router device, or another host
2063acting as a gateway.  For our discussion, the important point is that
2064the majority of traffic from Host A will pass through the router to
2065some other network before reaching its final destination.
2067        In a gatewayed network configuration, although Host A may
2068communicate with many other systems, all of its traffic will be sent
2069and received via one other peer on the local network, the router.
2071        Note that the case of two systems connected directly via
2072multiple physical links is, for purposes of configuring bonding, the
2073same as a gatewayed configuration.  In that case, it happens that all
2074traffic is destined for the "gateway" itself, not some other network
2075beyond the gateway.
2077        In a local configuration, the "switch" is acting primarily as
2078a switch, and the majority of traffic passes through this switch to
2079reach other stations on the same network.  An example would be the
2082    +----------+            +----------+       +--------+
2083    |          |eth0   port1|          +-------+ Host B |
2084    |  Host A  +------------+  switch  |port3  +--------+
2085    |          +------------+          |                  +--------+
2086    |          |eth1   port2|          +------------------+ Host C |
2087    +----------+            +----------+port4             +--------+
2090        Again, the switch may be a dedicated switch device, or another
2091host acting as a gateway.  For our discussion, the important point is
2092that the majority of traffic from Host A is destined for other hosts
2093on the same local network (Hosts B and C in the above example).
2095        In summary, in a gatewayed configuration, traffic to and from
2096the bonded device will be to the same MAC level peer on the network
2097(the gateway itself, i.e., the router), regardless of its final
2098destination.  In a local configuration, traffic flows directly to and
2099from the final destinations, thus, each destination (Host B, Host C)
2100will be addressed directly by their individual MAC addresses.
2102        This distinction between a gatewayed and a local network
2103configuration is important because many of the load balancing modes
2104available use the MAC addresses of the local network source and
2105destination to make load balancing decisions.  The behavior of each
2106mode is described below.
210912.1.1 MT Bonding Mode Selection for Single Switch Topology
2112        This configuration is the easiest to set up and to understand,
2113although you will have to decide which bonding mode best suits your
2114needs.  The trade offs for each mode are detailed below:
2116balance-rr: This mode is the only mode that will permit a single
2117        TCP/IP connection to stripe traffic across multiple
2118        interfaces. It is therefore the only mode that will allow a
2119        single TCP/IP stream to utilize more than one interface's
2120        worth of throughput.  This comes at a cost, however: the
2121        striping generally results in peer systems receiving packets out
2122        of order, causing TCP/IP's congestion control system to kick
2123        in, often by retransmitting segments.
2125        It is possible to adjust TCP/IP's congestion limits by
2126        altering the net.ipv4.tcp_reordering sysctl parameter.  The
2127        usual default value is 3, and the maximum useful value is 127.
2128        For a four interface balance-rr bond, expect that a single
2129        TCP/IP stream will utilize no more than approximately 2.3
2130        interface's worth of throughput, even after adjusting
2131        tcp_reordering.
2133        Note that the fraction of packets that will be delivered out of
2134        order is highly variable, and is unlikely to be zero.  The level
2135        of reordering depends upon a variety of factors, including the
2136        networking interfaces, the switch, and the topology of the
2137        configuration.  Speaking in general terms, higher speed network
2138        cards produce more reordering (due to factors such as packet
2139        coalescing), and a "many to many" topology will reorder at a
2140        higher rate than a "many slow to one fast" configuration.
2142        Many switches do not support any modes that stripe traffic
2143        (instead choosing a port based upon IP or MAC level addresses);
2144        for those devices, traffic for a particular connection flowing
2145        through the switch to a balance-rr bond will not utilize greater
2146        than one interface's worth of bandwidth.
2148        If you are utilizing protocols other than TCP/IP, UDP for
2149        example, and your application can tolerate out of order
2150        delivery, then this mode can allow for single stream datagram
2151        performance that scales near linearly as interfaces are added
2152        to the bond.
2154        This mode requires the switch to have the appropriate ports
2155        configured for "etherchannel" or "trunking."
2157active-backup: There is not much advantage in this network topology to
2158        the active-backup mode, as the inactive backup devices are all
2159        connected to the same peer as the primary.  In this case, a
2160        load balancing mode (with link monitoring) will provide the
2161        same level of network availability, but with increased
2162        available bandwidth.  On the plus side, active-backup mode
2163        does not require any configuration of the switch, so it may
2164        have value if the hardware available does not support any of
2165        the load balance modes.
2167balance-xor: This mode will limit traffic such that packets destined
2168        for specific peers will always be sent over the same
2169        interface.  Since the destination is determined by the MAC
2170        addresses involved, this mode works best in a "local" network
2171        configuration (as described above), with destinations all on
2172        the same local network.  This mode is likely to be suboptimal
2173        if all your traffic is passed through a single router (i.e., a
2174        "gatewayed" network configuration, as described above).
2176        As with balance-rr, the switch ports need to be configured for
2177        "etherchannel" or "trunking."
2179broadcast: Like active-backup, there is not much advantage to this
2180        mode in this type of network topology.
2182802.3ad: This mode can be a good choice for this type of network
2183        topology.  The 802.3ad mode is an IEEE standard, so all peers
2184        that implement 802.3ad should interoperate well.  The 802.3ad
2185        protocol includes automatic configuration of the aggregates,
2186        so minimal manual configuration of the switch is needed
2187        (typically only to designate that some set of devices is
2188        available for 802.3ad).  The 802.3ad standard also mandates
2189        that frames be delivered in order (within certain limits), so
2190        in general single connections will not see misordering of
2191        packets.  The 802.3ad mode does have some drawbacks: the
2192        standard mandates that all devices in the aggregate operate at
2193        the same speed and duplex.  Also, as with all bonding load
2194        balance modes other than balance-rr, no single connection will
2195        be able to utilize more than a single interface's worth of
2196        bandwidth.  
2198        Additionally, the linux bonding 802.3ad implementation
2199        distributes traffic by peer (using an XOR of MAC addresses),
2200        so in a "gatewayed" configuration, all outgoing traffic will
2201        generally use the same device.  Incoming traffic may also end
2202        up on a single device, but that is dependent upon the
2203        balancing policy of the peer's implementation.  In a
2204        "local" configuration, traffic will be distributed across the
2205        devices in the bond.
2207        Finally, the 802.3ad mode mandates the use of the MII monitor,
2208        therefore, the ARP monitor is not available in this mode.
2210balance-tlb: The balance-tlb mode balances outgoing traffic by peer.
2211        Since the balancing is done according to MAC address, in a
2212        "gatewayed" configuration (as described above), this mode will
2213        send all traffic across a single device.  However, in a
2214        "local" network configuration, this mode balances multiple
2215        local network peers across devices in a vaguely intelligent
2216        manner (not a simple XOR as in balance-xor or 802.3ad mode),
2217        so that mathematically unlucky MAC addresses (i.e., ones that
2218        XOR to the same value) will not all "bunch up" on a single
2219        interface.
2221        Unlike 802.3ad, interfaces may be of differing speeds, and no
2222        special switch configuration is required.  On the down side,
2223        in this mode all incoming traffic arrives over a single
2224        interface, this mode requires certain ethtool support in the
2225        network device driver of the slave interfaces, and the ARP
2226        monitor is not available.
2228balance-alb: This mode is everything that balance-tlb is, and more.
2229        It has all of the features (and restrictions) of balance-tlb,
2230        and will also balance incoming traffic from local network
2231        peers (as described in the Bonding Module Options section,
2232        above).
2234        The only additional down side to this mode is that the network
2235        device driver must support changing the hardware address while
2236        the device is open.
223812.1.2 MT Link Monitoring for Single Switch Topology
2241        The choice of link monitoring may largely depend upon which
2242mode you choose to use.  The more advanced load balancing modes do not
2243support the use of the ARP monitor, and are thus restricted to using
2244the MII monitor (which does not provide as high a level of end to end
2245assurance as the ARP monitor).
224712.2 Maximum Throughput in a Multiple Switch Topology
2250        Multiple switches may be utilized to optimize for throughput
2251when they are configured in parallel as part of an isolated network
2252between two or more systems, for example:
2254                       +-----------+
2255                       |  Host A   | 
2256                       +-+---+---+-+
2257                         |   |   |
2258                +--------+   |   +---------+
2259                |            |             |
2260         +------+---+  +-----+----+  +-----+----+
2261         | Switch A |  | Switch B |  | Switch C |
2262         +------+---+  +-----+----+  +-----+----+
2263                |            |             |
2264                +--------+   |   +---------+
2265                         |   |   |
2266                       +-+---+---+-+
2267                       |  Host B   | 
2268                       +-----------+
2270        In this configuration, the switches are isolated from one
2271another.  One reason to employ a topology such as this is for an
2272isolated network with many hosts (a cluster configured for high
2273performance, for example), using multiple smaller switches can be more
2274cost effective than a single larger switch, e.g., on a network with 24
2275hosts, three 24 port switches can be significantly less expensive than
2276a single 72 port switch.
2278        If access beyond the network is required, an individual host
2279can be equipped with an additional network device connected to an
2280external network; this host then additionally acts as a gateway.
228212.2.1 MT Bonding Mode Selection for Multiple Switch Topology
2285        In actual practice, the bonding mode typically employed in
2286configurations of this type is balance-rr.  Historically, in this
2287network configuration, the usual caveats about out of order packet
2288delivery are mitigated by the use of network adapters that do not do
2289any kind of packet coalescing (via the use of NAPI, or because the
2290device itself does not generate interrupts until some number of
2291packets has arrived).  When employed in this fashion, the balance-rr
2292mode allows individual connections between two hosts to effectively
2293utilize greater than one interface's bandwidth.
229512.2.2 MT Link Monitoring for Multiple Switch Topology
2298        Again, in actual practice, the MII monitor is most often used
2299in this configuration, as performance is given preference over
2300availability.  The ARP monitor will function in this topology, but its
2301advantages over the MII monitor are mitigated by the volume of probes
2302needed as the number of systems involved grows (remember that each
2303host in the network is configured with bonding).
230513. Switch Behavior Issues
230813.1 Link Establishment and Failover Delays
2311        Some switches exhibit undesirable behavior with regard to the
2312timing of link up and down reporting by the switch.
2314        First, when a link comes up, some switches may indicate that
2315the link is up (carrier available), but not pass traffic over the
2316interface for some period of time.  This delay is typically due to
2317some type of autonegotiation or routing protocol, but may also occur
2318during switch initialization (e.g., during recovery after a switch
2319failure).  If you find this to be a problem, specify an appropriate
2320value to the updelay bonding module option to delay the use of the
2321relevant interface(s).
2323        Second, some switches may "bounce" the link state one or more
2324times while a link is changing state.  This occurs most commonly while
2325the switch is initializing.  Again, an appropriate updelay value may
2328        Note that when a bonding interface has no active links, the
2329driver will immediately reuse the first link that goes up, even if the
2330updelay parameter has been specified (the updelay is ignored in this
2331case).  If there are slave interfaces waiting for the updelay timeout
2332to expire, the interface that first went into that state will be
2333immediately reused.  This reduces down time of the network if the
2334value of updelay has been overestimated, and since this occurs only in
2335cases with no connectivity, there is no additional penalty for
2336ignoring the updelay.
2338        In addition to the concerns about switch timings, if your
2339switches take a long time to go into backup mode, it may be desirable
2340to not activate a backup interface immediately after a link goes down.
2341Failover may be delayed via the downdelay bonding module option.
234313.2 Duplicated Incoming Packets
2346        NOTE: Starting with version 3.0.2, the bonding driver has logic to
2347suppress duplicate packets, which should largely eliminate this problem.
2348The following description is kept for reference.
2350        It is not uncommon to observe a short burst of duplicated
2351traffic when the bonding device is first used, or after it has been
2352idle for some period of time.  This is most easily observed by issuing
2353a "ping" to some other host on the network, and noticing that the
2354output from ping flags duplicates (typically one per slave).
2356        For example, on a bond in active-backup mode with five slaves
2357all connected to one switch, the output may appear as follows:
2359# ping -n
2360PING ( from : 56(84) bytes of data.
236164 bytes from icmp_seq=1 ttl=64 time=13.7 ms
236264 bytes from icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
236364 bytes from icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
236464 bytes from icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
236564 bytes from icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
236664 bytes from icmp_seq=2 ttl=64 time=0.216 ms
236764 bytes from icmp_seq=3 ttl=64 time=0.267 ms
236864 bytes from icmp_seq=4 ttl=64 time=0.222 ms
2370        This is not due to an error in the bonding driver, rather, it
2371is a side effect of how many switches update their MAC forwarding
2372tables.  Initially, the switch does not associate the MAC address in
2373the packet with a particular switch port, and so it may send the
2374traffic to all ports until its MAC forwarding table is updated.  Since
2375the interfaces attached to the bond may occupy multiple ports on a
2376single switch, when the switch (temporarily) floods the traffic to all
2377ports, the bond device receives multiple copies of the same packet
2378(one per slave device).
2380        The duplicated packet behavior is switch dependent, some
2381switches exhibit this, and some do not.  On switches that display this
2382behavior, it can be induced by clearing the MAC forwarding table (on
2383most Cisco switches, the privileged command "clear mac address-table
2384dynamic" will accomplish this).
238614. Hardware Specific Considerations
2389        This section contains additional information for configuring
2390bonding on specific hardware platforms, or for interfacing bonding
2391with particular switches or other devices.
239314.1 IBM BladeCenter
2396        This applies to the JS20 and similar systems.
2398        On the JS20 blades, the bonding driver supports only
2399balance-rr, active-backup, balance-tlb and balance-alb modes.  This is
2400largely due to the network topology inside the BladeCenter, detailed
2403JS20 network adapter information
2406        All JS20s come with two Broadcom Gigabit Ethernet ports
2407integrated on the planar (that's "motherboard" in IBM-speak).  In the
2408BladeCenter chassis, the eth0 port of all JS20 blades is hard wired to
2409I/O Module #1; similarly, all eth1 ports are wired to I/O Module #2.
2410An add-on Broadcom daughter card can be installed on a JS20 to provide
2411two more Gigabit Ethernet ports.  These ports, eth2 and eth3, are
2412wired to I/O Modules 3 and 4, respectively.
2414        Each I/O Module may contain either a switch or a passthrough
2415module (which allows ports to be directly connected to an external
2416switch).  Some bonding modes require a specific BladeCenter internal
2417network topology in order to function; these are detailed below.
2419        Additional BladeCenter-specific networking information can be
2420found in two IBM Redbooks (
2422"IBM eServer BladeCenter Networking Options"
2423"IBM eServer BladeCenter Layer 2-7 Network Switching"
2425BladeCenter networking configuration
2428        Because a BladeCenter can be configured in a very large number
2429of ways, this discussion will be confined to describing basic
2432        Normally, Ethernet Switch Modules (ESMs) are used in I/O
2433modules 1 and 2.  In this configuration, the eth0 and eth1 ports of a
2434JS20 will be connected to different internal switches (in the
2435respective I/O modules).
2437        A passthrough module (OPM or CPM, optical or copper,
2438passthrough module) connects the I/O module directly to an external
2439switch.  By using PMs in I/O module #1 and #2, the eth0 and eth1
2440interfaces of a JS20 can be redirected to the outside world and
2441connected to a common external switch.
2443        Depending upon the mix of ESMs and PMs, the network will
2444appear to bonding as either a single switch topology (all PMs) or as a
2445multiple switch topology (one or more ESMs, zero or more PMs).  It is
2446also possible to connect ESMs together, resulting in a configuration
2447much like the example in "High Availability in a Multiple Switch
2448Topology," above.
2450Requirements for specific modes
2453        The balance-rr mode requires the use of passthrough modules
2454for devices in the bond, all connected to an common external switch.
2455That switch must be configured for "etherchannel" or "trunking" on the
2456appropriate ports, as is usual for balance-rr.
2458        The balance-alb and balance-tlb modes will function with
2459either switch modules or passthrough modules (or a mix).  The only
2460specific requirement for these modes is that all network interfaces
2461must be able to reach all destinations for traffic sent over the
2462bonding device (i.e., the network must converge at some point outside
2463the BladeCenter).
2465        The active-backup mode has no additional requirements.
2467Link monitoring issues
2470        When an Ethernet Switch Module is in place, only the ARP
2471monitor will reliably detect link loss to an external switch.  This is
2472nothing unusual, but examination of the BladeCenter cabinet would
2473suggest that the "external" network ports are the ethernet ports for
2474the system, when it fact there is a switch between these "external"
2475ports and the devices on the JS20 system itself.  The MII monitor is
2476only able to detect link failures between the ESM and the JS20 system.
2478        When a passthrough module is in place, the MII monitor does
2479detect failures to the "external" port, which is then directly
2480connected to the JS20 system.
2482Other concerns
2485        The Serial Over LAN (SoL) link is established over the primary
2486ethernet (eth0) only, therefore, any loss of link to eth0 will result
2487in losing your SoL connection.  It will not fail over with other
2488network traffic, as the SoL system is beyond the control of the
2489bonding driver.
2491        It may be desirable to disable spanning tree on the switch
2492(either the internal Ethernet Switch Module, or an external switch) to
2493avoid fail-over delay issues when using bonding.
249615. Frequently Asked Questions
24991.  Is it SMP safe?
2501        Yes. The old 2.0.xx channel bonding patch was not SMP safe.
2502The new driver was designed to be SMP safe from the start.
25042.  What type of cards will work with it?
2506        Any Ethernet type cards (you can even mix cards - a Intel
2507EtherExpress PRO/100 and a 3com 3c905b, for example).  For most modes,
2508devices need not be of the same speed.
2510        Starting with version 3.2.1, bonding also supports Infiniband
2511slaves in active-backup mode.
25133.  How many bonding devices can I have?
2515        There is no limit.
25174.  How many slaves can a bonding device have?
2519        This is limited only by the number of network interfaces Linux
2520supports and/or the number of network cards you can place in your
25235.  What happens when a slave link dies?
2525        If link monitoring is enabled, then the failing device will be
2526disabled.  The active-backup mode will fail over to a backup link, and
2527other modes will ignore the failed link.  The link will continue to be
2528monitored, and should it recover, it will rejoin the bond (in whatever
2529manner is appropriate for the mode). See the sections on High
2530Availability and the documentation for each mode for additional
2533        Link monitoring can be enabled via either the miimon or
2534arp_interval parameters (described in the module parameters section,
2535above).  In general, miimon monitors the carrier state as sensed by
2536the underlying network device, and the arp monitor (arp_interval)
2537monitors connectivity to another host on the local network.
2539        If no link monitoring is configured, the bonding driver will
2540be unable to detect link failures, and will assume that all links are
2541always available.  This will likely result in lost packets, and a
2542resulting degradation of performance.  The precise performance loss
2543depends upon the bonding mode and network configuration.
25456.  Can bonding be used for High Availability?
2547        Yes.  See the section on High Availability for details.
25497.  Which switches/systems does it work with?
2551        The full answer to this depends upon the desired mode.
2553        In the basic balance modes (balance-rr and balance-xor), it
2554works with any system that supports etherchannel (also called
2555trunking).  Most managed switches currently available have such
2556support, and many unmanaged switches as well.
2558        The advanced balance modes (balance-tlb and balance-alb) do
2559not have special switch requirements, but do need device drivers that
2560support specific features (described in the appropriate section under
2561module parameters, above).
2563        In 802.3ad mode, it works with systems that support IEEE
2564802.3ad Dynamic Link Aggregation.  Most managed and many unmanaged
2565switches currently available support 802.3ad.
2567        The active-backup mode should work with any Layer-II switch.
25698.  Where does a bonding device get its MAC address from?
2571        When using slave devices that have fixed MAC addresses, or when
2572the fail_over_mac option is enabled, the bonding device's MAC address is
2573the MAC address of the active slave.
2575        For other configurations, if not explicitly configured (with
2576ifconfig or ip link), the MAC address of the bonding device is taken from
2577its first slave device.  This MAC address is then passed to all following
2578slaves and remains persistent (even if the first slave is removed) until
2579the bonding device is brought down or reconfigured.
2581        If you wish to change the MAC address, you can set it with
2582ifconfig or ip link:
2584# ifconfig bond0 hw ether 00:11:22:33:44:55
2586# ip link set bond0 address 66:77:88:99:aa:bb
2588        The MAC address can be also changed by bringing down/up the
2589device and then changing its slaves (or their order):
2591# ifconfig bond0 down ; modprobe -r bonding
2592# ifconfig bond0 .... up
2593# ifenslave bond0 eth...
2595        This method will automatically take the address from the next
2596slave that is added.
2598        To restore your slaves' MAC addresses, you need to detach them
2599from the bond (`ifenslave -d bond0 eth0'). The bonding driver will
2600then restore the MAC addresses that the slaves had before they were
260316. Resources and Links
2606        The latest version of the bonding driver can be found in the latest
2607version of the linux kernel, found on
2609        The latest version of this document can be found in the latest kernel
2610source (named Documentation/networking/bonding.txt).
2612        Discussions regarding the usage of the bonding driver take place on the
2613bonding-devel mailing list, hosted at If you have questions or
2614problems, post them to the list.  The list address is:
2618        The administrative interface (to subscribe or unsubscribe) can
2619be found at:
2623        Discussions regarding the development of the bonding driver take place
2624on the main Linux network mailing list, hosted at The list
2625address is:
2629        The administrative interface (to subscribe or unsubscribe) can
2630be found at:
2634Donald Becker's Ethernet Drivers and diag programs may be found at :
2635 -*/ 
2637You will also find a lot of information regarding Ethernet, NWay, MII,
2638etc. at
2640-- END --
2641 kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.