linux/Documentation/networking/bonding.txt
<<
>>
Prefs
   1
   2                Linux Ethernet Bonding Driver HOWTO
   3
   4                Latest update: 12 November 2007
   5
   6Initial release : Thomas Davis <tadavis at lbl.gov>
   7Corrections, HA extensions : 2000/10/03-15 :
   8  - Willy Tarreau <willy at meta-x.org>
   9  - Constantine Gavrilov <const-g at xpert.com>
  10  - Chad N. Tindel <ctindel at ieee dot org>
  11  - Janice Girouard <girouard at us dot ibm dot com>
  12  - Jay Vosburgh <fubar at us dot ibm dot com>
  13
  14Reorganized and updated Feb 2005 by Jay Vosburgh
  15Added Sysfs information: 2006/04/24
  16  - Mitch Williams <mitch.a.williams at intel.com>
  17
  18Introduction
  19============
  20
  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.
  26        
  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.
  31
  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.
  34
  35Table of Contents
  36=================
  37
  381. Bonding Driver Installation
  39
  402. Bonding Driver Options
  41
  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
  52
  534. Querying Bonding Configuration
  544.1     Bonding Configuration
  554.2     Network Configuration
  56
  575. Switch Configuration
  58
  596. 802.1q VLAN Support
  60
  617. Link Monitoring
  627.1     ARP Monitor Operation
  637.2     Configuring Multiple ARP Targets
  647.3     MII Monitor Operation
  65
  668. Potential Trouble Sources
  678.1     Adventures in Routing
  688.2     Ethernet Device Renaming
  698.3     Painfully Slow Or No Failed Link Detection By Miimon
  70
  719. SNMP agents
  72
  7310. Promiscuous mode
  74
  7511. Configuring Bonding for High Availability
  7611.1    High Availability in a Single Switch Topology
  7711.2    High Availability in a Multiple Switch Topology
  7811.2.1          HA Bonding Mode Selection for Multiple Switch Topology
  7911.2.2          HA Link Monitoring for Multiple Switch Topology
  80
  8112. Configuring Bonding for Maximum Throughput
  8212.1    Maximum Throughput in a Single Switch Topology
  8312.1.1          MT Bonding Mode Selection for Single Switch Topology
  8412.1.2          MT Link Monitoring for Single Switch Topology
  8512.2    Maximum Throughput in a Multiple Switch Topology
  8612.2.1          MT Bonding Mode Selection for Multiple Switch Topology
  8712.2.2          MT Link Monitoring for Multiple Switch Topology
  88
  8913. Switch Behavior Issues
  9013.1    Link Establishment and Failover Delays
  9113.2    Duplicated Incoming Packets
  92
  9314. Hardware Specific Considerations
  9414.1    IBM BladeCenter
  95
  9615. Frequently Asked Questions
  97
  9816. Resources and Links
  99
 100
 1011. Bonding Driver Installation
 102==============================
 103
 104        Most popular distro kernels ship with the bonding driver
 105already available as a module and the ifenslave user level control
 106program installed and ready for use. If your distro does not, or you
 107have need to compile bonding from source (e.g., configuring and
 108installing a mainline kernel from kernel.org), you'll need to perform
 109the following steps:
 110
 1111.1 Configure and build the kernel with bonding
 112-----------------------------------------------
 113
 114        The current version of the bonding driver is available in the
 115drivers/net/bonding subdirectory of the most recent kernel source
 116(which is available on http://kernel.org).  Most users "rolling their
 117own" will want to use the most recent kernel from kernel.org.
 118
 119        Configure kernel with "make menuconfig" (or "make xconfig" or
 120"make config"), then select "Bonding driver support" in the "Network
 121device support" section.  It is recommended that you configure the
 122driver as module since it is currently the only way to pass parameters
 123to the driver or configure more than one bonding device.
 124
 125        Build and install the new kernel and modules, then continue
 126below to install ifenslave.
 127
 1281.2 Install ifenslave Control Utility
 129-------------------------------------
 130
 131        The ifenslave user level control program is included in the
 132kernel source tree, in the file Documentation/networking/ifenslave.c.
 133It is generally recommended that you use the ifenslave that
 134corresponds to the kernel that you are using (either from the same
 135source tree or supplied with the distro), however, ifenslave
 136executables from older kernels should function (but features newer
 137than the ifenslave release are not supported).  Running an ifenslave
 138that is newer than the kernel is not supported, and may or may not
 139work.
 140
 141        To install ifenslave, do the following:
 142
 143# gcc -Wall -O -I/usr/src/linux/include ifenslave.c -o ifenslave
 144# cp ifenslave /sbin/ifenslave
 145
 146        If your kernel source is not in "/usr/src/linux," then replace
 147"/usr/src/linux/include" in the above with the location of your kernel
 148source include directory.
 149
 150        You may wish to back up any existing /sbin/ifenslave, or, for
 151testing or informal use, tag the ifenslave to the kernel version
 152(e.g., name the ifenslave executable /sbin/ifenslave-2.6.10).
 153
 154IMPORTANT NOTE:
 155
 156        If you omit the "-I" or specify an incorrect directory, you
 157may end up with an ifenslave that is incompatible with the kernel
 158you're trying to build it for.  Some distros (e.g., Red Hat from 7.1
 159onwards) do not have /usr/include/linux symbolically linked to the
 160default kernel source include directory.
 161
 162SECOND IMPORTANT NOTE:
 163        If you plan to configure bonding using sysfs, you do not need
 164to use ifenslave.
 165
 1662. Bonding Driver Options
 167=========================
 168
 169        Options for the bonding driver are supplied as parameters to the
 170bonding module at load time, or are specified via sysfs.
 171
 172        Module options may be given as command line arguments to the
 173insmod or modprobe command, but are usually specified in either the
 174/etc/modules.conf or /etc/modprobe.conf configuration file, or in a
 175distro-specific configuration file (some of which are detailed in the next
 176section).
 177
 178        Details on bonding support for sysfs is provided in the
 179"Configuring Bonding Manually via Sysfs" section, below.
 180
 181        The available bonding driver parameters are listed below. If a
 182parameter is not specified the default value is used.  When initially
 183configuring a bond, it is recommended "tail -f /var/log/messages" be
 184run in a separate window to watch for bonding driver error messages.
 185
 186        It is critical that either the miimon or arp_interval and
 187arp_ip_target parameters be specified, otherwise serious network
 188degradation will occur during link failures.  Very few devices do not
 189support at least miimon, so there is really no reason not to use it.
 190
 191        Options with textual values will accept either the text name
 192or, for backwards compatibility, the option value.  E.g.,
 193"mode=802.3ad" and "mode=4" set the same mode.
 194
 195        The parameters are as follows:
 196
 197ad_select
 198
 199        Specifies the 802.3ad aggregation selection logic to use.  The
 200        possible values and their effects are:
 201
 202        stable or 0
 203
 204                The active aggregator is chosen by largest aggregate
 205                bandwidth.
 206
 207                Reselection of the active aggregator occurs only when all
 208                slaves of the active aggregator are down or the active
 209                aggregator has no slaves.
 210
 211                This is the default value.
 212
 213        bandwidth or 1
 214
 215                The active aggregator is chosen by largest aggregate
 216                bandwidth.  Reselection occurs if:
 217
 218                - A slave is added to or removed from the bond
 219
 220                - Any slave's link state changes
 221
 222                - Any slave's 802.3ad association state changes
 223
 224                - The bond's adminstrative state changes to up
 225
 226        count or 2
 227
 228                The active aggregator is chosen by the largest number of
 229                ports (slaves).  Reselection occurs as described under the
 230                "bandwidth" setting, above.
 231
 232        The bandwidth and count selection policies permit failover of
 233        802.3ad aggregations when partial failure of the active aggregator
 234        occurs.  This keeps the aggregator with the highest availability
 235        (either in bandwidth or in number of ports) active at all times.
 236
 237        This option was added in bonding version 3.4.0.
 238
 239arp_interval
 240
 241        Specifies the ARP link monitoring frequency in milliseconds.
 242
 243        The ARP monitor works by periodically checking the slave
 244        devices to determine whether they have sent or received
 245        traffic recently (the precise criteria depends upon the
 246        bonding mode, and the state of the slave).  Regular traffic is
 247        generated via ARP probes issued for the addresses specified by
 248        the arp_ip_target option.
 249
 250        This behavior can be modified by the arp_validate option,
 251        below.
 252
 253        If ARP monitoring is used in an etherchannel compatible mode
 254        (modes 0 and 2), the switch should be configured in a mode
 255        that evenly distributes packets across all links. If the
 256        switch is configured to distribute the packets in an XOR
 257        fashion, all replies from the ARP targets will be received on
 258        the same link which could cause the other team members to
 259        fail.  ARP monitoring should not be used in conjunction with
 260        miimon.  A value of 0 disables ARP monitoring.  The default
 261        value is 0.
 262
 263arp_ip_target
 264
 265        Specifies the IP addresses to use as ARP monitoring peers when
 266        arp_interval is > 0.  These are the targets of the ARP request
 267        sent to determine the health of the link to the targets.
 268        Specify these values in ddd.ddd.ddd.ddd format.  Multiple IP
 269        addresses must be separated by a comma.  At least one IP
 270        address must be given for ARP monitoring to function.  The
 271        maximum number of targets that can be specified is 16.  The
 272        default value is no IP addresses.
 273
 274arp_validate
 275
 276        Specifies whether or not ARP probes and replies should be
 277        validated in the active-backup mode.  This causes the ARP
 278        monitor to examine the incoming ARP requests and replies, and
 279        only consider a slave to be up if it is receiving the
 280        appropriate ARP traffic.
 281
 282        Possible values are:
 283
 284        none or 0
 285
 286                No validation is performed.  This is the default.
 287
 288        active or 1
 289
 290                Validation is performed only for the active slave.
 291
 292        backup or 2
 293
 294                Validation is performed only for backup slaves.
 295
 296        all or 3
 297
 298                Validation is performed for all slaves.
 299
 300        For the active slave, the validation checks ARP replies to
 301        confirm that they were generated by an arp_ip_target.  Since
 302        backup slaves do not typically receive these replies, the
 303        validation performed for backup slaves is on the ARP request
 304        sent out via the active slave.  It is possible that some
 305        switch or network configurations may result in situations
 306        wherein the backup slaves do not receive the ARP requests; in
 307        such a situation, validation of backup slaves must be
 308        disabled.
 309
 310        This option is useful in network configurations in which
 311        multiple bonding hosts are concurrently issuing ARPs to one or
 312        more targets beyond a common switch.  Should the link between
 313        the switch and target fail (but not the switch itself), the
 314        probe traffic generated by the multiple bonding instances will
 315        fool the standard ARP monitor into considering the links as
 316        still up.  Use of the arp_validate option can resolve this, as
 317        the ARP monitor will only consider ARP requests and replies
 318        associated with its own instance of bonding.
 319
 320        This option was added in bonding version 3.1.0.
 321
 322downdelay
 323
 324        Specifies the time, in milliseconds, to wait before disabling
 325        a slave after a link failure has been detected.  This option
 326        is only valid for the miimon link monitor.  The downdelay
 327        value should be a multiple of the miimon value; if not, it
 328        will be rounded down to the nearest multiple.  The default
 329        value is 0.
 330
 331fail_over_mac
 332
 333        Specifies whether active-backup mode should set all slaves to
 334        the same MAC address at enslavement (the traditional
 335        behavior), or, when enabled, perform special handling of the
 336        bond's MAC address in accordance with the selected policy.
 337
 338        Possible values are:
 339
 340        none or 0
 341
 342                This setting disables fail_over_mac, and causes
 343                bonding to set all slaves of an active-backup bond to
 344                the same MAC address at enslavement time.  This is the
 345                default.
 346
 347        active or 1
 348
 349                The "active" fail_over_mac policy indicates that the
 350                MAC address of the bond should always be the MAC
 351                address of the currently active slave.  The MAC
 352                address of the slaves is not changed; instead, the MAC
 353                address of the bond changes during a failover.
 354
 355                This policy is useful for devices that cannot ever
 356                alter their MAC address, or for devices that refuse
 357                incoming broadcasts with their own source MAC (which
 358                interferes with the ARP monitor).
 359
 360                The down side of this policy is that every device on
 361                the network must be updated via gratuitous ARP,
 362                vs. just updating a switch or set of switches (which
 363                often takes place for any traffic, not just ARP
 364                traffic, if the switch snoops incoming traffic to
 365                update its tables) for the traditional method.  If the
 366                gratuitous ARP is lost, communication may be
 367                disrupted.
 368
 369                When this policy is used in conjuction with the mii
 370                monitor, devices which assert link up prior to being
 371                able to actually transmit and receive are particularly
 372                susecptible to loss of the gratuitous ARP, and an
 373                appropriate updelay setting may be required.
 374
 375        follow or 2
 376
 377                The "follow" fail_over_mac policy causes the MAC
 378                address of the bond to be selected normally (normally
 379                the MAC address of the first slave added to the bond).
 380                However, the second and subsequent slaves are not set
 381                to this MAC address while they are in a backup role; a
 382                slave is programmed with the bond's MAC address at
 383                failover time (and the formerly active slave receives
 384                the newly active slave's MAC address).
 385
 386                This policy is useful for multiport devices that
 387                either become confused or incur a performance penalty
 388                when multiple ports are programmed with the same MAC
 389                address.
 390
 391
 392        The default policy is none, unless the first slave cannot
 393        change its MAC address, in which case the active policy is
 394        selected by default.
 395
 396        This option may be modified via sysfs only when no slaves are
 397        present in the bond.
 398
 399        This option was added in bonding version 3.2.0.  The "follow"
 400        policy was added in bonding version 3.3.0.
 401
 402lacp_rate
 403
 404        Option specifying the rate in which we'll ask our link partner
 405        to transmit LACPDU packets in 802.3ad mode.  Possible values
 406        are:
 407
 408        slow or 0
 409                Request partner to transmit LACPDUs every 30 seconds
 410
 411        fast or 1
 412                Request partner to transmit LACPDUs every 1 second
 413
 414        The default is slow.
 415
 416max_bonds
 417
 418        Specifies the number of bonding devices to create for this
 419        instance of the bonding driver.  E.g., if max_bonds is 3, and
 420        the bonding driver is not already loaded, then bond0, bond1
 421        and bond2 will be created.  The default value is 1.  Specifying
 422        a value of 0 will load bonding, but will not create any devices.
 423
 424miimon
 425
 426        Specifies the MII link monitoring frequency in milliseconds.
 427        This determines how often the link state of each slave is
 428        inspected for link failures.  A value of zero disables MII
 429        link monitoring.  A value of 100 is a good starting point.
 430        The use_carrier option, below, affects how the link state is
 431        determined.  See the High Availability section for additional
 432        information.  The default value is 0.
 433
 434mode
 435
 436        Specifies one of the bonding policies. The default is
 437        balance-rr (round robin).  Possible values are:
 438
 439        balance-rr or 0
 440
 441                Round-robin policy: Transmit packets in sequential
 442                order from the first available slave through the
 443                last.  This mode provides load balancing and fault
 444                tolerance.
 445
 446        active-backup or 1
 447
 448                Active-backup policy: Only one slave in the bond is
 449                active.  A different slave becomes active if, and only
 450                if, the active slave fails.  The bond's MAC address is
 451                externally visible on only one port (network adapter)
 452                to avoid confusing the switch.
 453
 454                In bonding version 2.6.2 or later, when a failover
 455                occurs in active-backup mode, bonding will issue one
 456                or more gratuitous ARPs on the newly active slave.
 457                One gratuitous ARP is issued for the bonding master
 458                interface and each VLAN interfaces configured above
 459                it, provided that the interface has at least one IP
 460                address configured.  Gratuitous ARPs issued for VLAN
 461                interfaces are tagged with the appropriate VLAN id.
 462
 463                This mode provides fault tolerance.  The primary
 464                option, documented below, affects the behavior of this
 465                mode.
 466
 467        balance-xor or 2
 468
 469                XOR policy: Transmit based on the selected transmit
 470                hash policy.  The default policy is a simple [(source
 471                MAC address XOR'd with destination MAC address) modulo
 472                slave count].  Alternate transmit policies may be
 473                selected via the xmit_hash_policy option, described
 474                below.
 475
 476                This mode provides load balancing and fault tolerance.
 477
 478        broadcast or 3
 479
 480                Broadcast policy: transmits everything on all slave
 481                interfaces.  This mode provides fault tolerance.
 482
 483        802.3ad or 4
 484
 485                IEEE 802.3ad Dynamic link aggregation.  Creates
 486                aggregation groups that share the same speed and
 487                duplex settings.  Utilizes all slaves in the active
 488                aggregator according to the 802.3ad specification.
 489
 490                Slave selection for outgoing traffic is done according
 491                to the transmit hash policy, which may be changed from
 492                the default simple XOR policy via the xmit_hash_policy
 493                option, documented below.  Note that not all transmit
 494                policies may be 802.3ad compliant, particularly in
 495                regards to the packet mis-ordering requirements of
 496                section 43.2.4 of the 802.3ad standard.  Differing
 497                peer implementations will have varying tolerances for
 498                noncompliance.
 499
 500                Prerequisites:
 501
 502                1. Ethtool support in the base drivers for retrieving
 503                the speed and duplex of each slave.
 504
 505                2. A switch that supports IEEE 802.3ad Dynamic link
 506                aggregation.
 507
 508                Most switches will require some type of configuration
 509                to enable 802.3ad mode.
 510
 511        balance-tlb or 5
 512
 513                Adaptive transmit load balancing: channel bonding that
 514                does not require any special switch support.  The
 515                outgoing traffic is distributed according to the
 516                current load (computed relative to the speed) on each
 517                slave.  Incoming traffic is received by the current
 518                slave.  If the receiving slave fails, another slave
 519                takes over the MAC address of the failed receiving
 520                slave.
 521
 522                Prerequisite:
 523
 524                Ethtool support in the base drivers for retrieving the
 525                speed of each slave.
 526
 527        balance-alb or 6
 528
 529                Adaptive load balancing: includes balance-tlb plus
 530                receive load balancing (rlb) for IPV4 traffic, and
 531                does not require any special switch support.  The
 532                receive load balancing is achieved by ARP negotiation.
 533                The bonding driver intercepts the ARP Replies sent by
 534                the local system on their way out and overwrites the
 535                source hardware address with the unique hardware
 536                address of one of the slaves in the bond such that
 537                different peers use different hardware addresses for
 538                the server.
 539
 540                Receive traffic from connections created by the server
 541                is also balanced.  When the local system sends an ARP
 542                Request the bonding driver copies and saves the peer's
 543                IP information from the ARP packet.  When the ARP
 544                Reply arrives from the peer, its hardware address is
 545                retrieved and the bonding driver initiates an ARP
 546                reply to this peer assigning it to one of the slaves
 547                in the bond.  A problematic outcome of using ARP
 548                negotiation for balancing is that each time that an
 549                ARP request is broadcast it uses the hardware address
 550                of the bond.  Hence, peers learn the hardware address
 551                of the bond and the balancing of receive traffic
 552                collapses to the current slave.  This is handled by
 553                sending updates (ARP Replies) to all the peers with
 554                their individually assigned hardware address such that
 555                the traffic is redistributed.  Receive traffic is also
 556                redistributed when a new slave is added to the bond
 557                and when an inactive slave is re-activated.  The
 558                receive load is distributed sequentially (round robin)
 559                among the group of highest speed slaves in the bond.
 560
 561                When a link is reconnected or a new slave joins the
 562                bond the receive traffic is redistributed among all
 563                active slaves in the bond by initiating ARP Replies
 564                with the selected MAC address to each of the
 565                clients. The updelay parameter (detailed below) must
 566                be set to a value equal or greater than the switch's
 567                forwarding delay so that the ARP Replies sent to the
 568                peers will not be blocked by the switch.
 569
 570                Prerequisites:
 571
 572                1. Ethtool support in the base drivers for retrieving
 573                the speed of each slave.
 574
 575                2. Base driver support for setting the hardware
 576                address of a device while it is open.  This is
 577                required so that there will always be one slave in the
 578                team using the bond hardware address (the
 579                curr_active_slave) while having a unique hardware
 580                address for each slave in the bond.  If the
 581                curr_active_slave fails its hardware address is
 582                swapped with the new curr_active_slave that was
 583                chosen.
 584
 585num_grat_arp
 586
 587        Specifies the number of gratuitous ARPs to be issued after a
 588        failover event.  One gratuitous ARP is issued immediately after
 589        the failover, subsequent ARPs are sent at a rate of one per link
 590        monitor interval (arp_interval or miimon, whichever is active).
 591
 592        The valid range is 0 - 255; the default value is 1.  This option
 593        affects only the active-backup mode.  This option was added for
 594        bonding version 3.3.0.
 595
 596num_unsol_na
 597
 598        Specifies the number of unsolicited IPv6 Neighbor Advertisements
 599        to be issued after a failover event.  One unsolicited NA is issued
 600        immediately after the failover.
 601
 602        The valid range is 0 - 255; the default value is 1.  This option
 603        affects only the active-backup mode.  This option was added for
 604        bonding version 3.4.0.
 605
 606primary
 607
 608        A string (eth0, eth2, etc) specifying which slave is the
 609        primary device.  The specified device will always be the
 610        active slave while it is available.  Only when the primary is
 611        off-line will alternate devices be used.  This is useful when
 612        one slave is preferred over another, e.g., when one slave has
 613        higher throughput than another.
 614
 615        The primary option is only valid for active-backup mode.
 616
 617updelay
 618
 619        Specifies the time, in milliseconds, to wait before enabling a
 620        slave after a link recovery has been detected.  This option is
 621        only valid for the miimon link monitor.  The updelay value
 622        should be a multiple of the miimon value; if not, it will be
 623        rounded down to the nearest multiple.  The default value is 0.
 624
 625use_carrier
 626
 627        Specifies whether or not miimon should use MII or ETHTOOL
 628        ioctls vs. netif_carrier_ok() to determine the link
 629        status. The MII or ETHTOOL ioctls are less efficient and
 630        utilize a deprecated calling sequence within the kernel.  The
 631        netif_carrier_ok() relies on the device driver to maintain its
 632        state with netif_carrier_on/off; at this writing, most, but
 633        not all, device drivers support this facility.
 634
 635        If bonding insists that the link is up when it should not be,
 636        it may be that your network device driver does not support
 637        netif_carrier_on/off.  The default state for netif_carrier is
 638        "carrier on," so if a driver does not support netif_carrier,
 639        it will appear as if the link is always up.  In this case,
 640        setting use_carrier to 0 will cause bonding to revert to the
 641        MII / ETHTOOL ioctl method to determine the link state.
 642
 643        A value of 1 enables the use of netif_carrier_ok(), a value of
 644        0 will use the deprecated MII / ETHTOOL ioctls.  The default
 645        value is 1.
 646
 647xmit_hash_policy
 648
 649        Selects the transmit hash policy to use for slave selection in
 650        balance-xor and 802.3ad modes.  Possible values are:
 651
 652        layer2
 653
 654                Uses XOR of hardware MAC addresses to generate the
 655                hash.  The formula is
 656
 657                (source MAC XOR destination MAC) modulo slave count
 658
 659                This algorithm will place all traffic to a particular
 660                network peer on the same slave.
 661
 662                This algorithm is 802.3ad compliant.
 663
 664        layer2+3
 665
 666                This policy uses a combination of layer2 and layer3
 667                protocol information to generate the hash.
 668
 669                Uses XOR of hardware MAC addresses and IP addresses to
 670                generate the hash.  The formula is
 671
 672                (((source IP XOR dest IP) AND 0xffff) XOR
 673                        ( source MAC XOR destination MAC ))
 674                                modulo slave count
 675
 676                This algorithm will place all traffic to a particular
 677                network peer on the same slave.  For non-IP traffic,
 678                the formula is the same as for the layer2 transmit
 679                hash policy.
 680
 681                This policy is intended to provide a more balanced
 682                distribution of traffic than layer2 alone, especially
 683                in environments where a layer3 gateway device is
 684                required to reach most destinations.
 685
 686                This algorithm is 802.3ad compliant.
 687
 688        layer3+4
 689
 690                This policy uses upper layer protocol information,
 691                when available, to generate the hash.  This allows for
 692                traffic to a particular network peer to span multiple
 693                slaves, although a single connection will not span
 694                multiple slaves.
 695
 696                The formula for unfragmented TCP and UDP packets is
 697
 698                ((source port XOR dest port) XOR
 699                         ((source IP XOR dest IP) AND 0xffff)
 700                                modulo slave count
 701
 702                For fragmented TCP or UDP packets and all other IP
 703                protocol traffic, the source and destination port
 704                information is omitted.  For non-IP traffic, the
 705                formula is the same as for the layer2 transmit hash
 706                policy.
 707
 708                This policy is intended to mimic the behavior of
 709                certain switches, notably Cisco switches with PFC2 as
 710                well as some Foundry and IBM products.
 711
 712                This algorithm is not fully 802.3ad compliant.  A
 713                single TCP or UDP conversation containing both
 714                fragmented and unfragmented packets will see packets
 715                striped across two interfaces.  This may result in out
 716                of order delivery.  Most traffic types will not meet
 717                this criteria, as TCP rarely fragments traffic, and
 718                most UDP traffic is not involved in extended
 719                conversations.  Other implementations of 802.3ad may
 720                or may not tolerate this noncompliance.
 721
 722        The default value is layer2.  This option was added in bonding
 723        version 2.6.3.  In earlier versions of bonding, this parameter
 724        does not exist, and the layer2 policy is the only policy.  The
 725        layer2+3 value was added for bonding version 3.2.2.
 726
 727
 7283. Configuring Bonding Devices
 729==============================
 730
 731        You can configure bonding using either your distro's network
 732initialization scripts, or manually using either ifenslave or the
 733sysfs interface.  Distros generally use one of two packages for the
 734network initialization scripts: initscripts or sysconfig.  Recent
 735versions of these packages have support for bonding, while older
 736versions do not.
 737
 738        We will first describe the options for configuring bonding for
 739distros using versions of initscripts and sysconfig with full or
 740partial support for bonding, then provide information on enabling
 741bonding without support from the network initialization scripts (i.e.,
 742older versions of initscripts or sysconfig).
 743
 744        If you're unsure whether your distro uses sysconfig or
 745initscripts, or don't know if it's new enough, have no fear.
 746Determining this is fairly straightforward.
 747
 748        First, issue the command:
 749
 750$ rpm -qf /sbin/ifup
 751
 752        It will respond with a line of text starting with either
 753"initscripts" or "sysconfig," followed by some numbers.  This is the
 754package that provides your network initialization scripts.
 755
 756        Next, to determine if your installation supports bonding,
 757issue the command:
 758
 759$ grep ifenslave /sbin/ifup
 760
 761        If this returns any matches, then your initscripts or
 762sysconfig has support for bonding.
 763
 7643.1 Configuration with Sysconfig Support
 765----------------------------------------
 766
 767        This section applies to distros using a version of sysconfig
 768with bonding support, for example, SuSE Linux Enterprise Server 9.
 769
 770        SuSE SLES 9's networking configuration system does support
 771bonding, however, at this writing, the YaST system configuration
 772front end does not provide any means to work with bonding devices.
 773Bonding devices can be managed by hand, however, as follows.
 774
 775        First, if they have not already been configured, configure the
 776slave devices.  On SLES 9, this is most easily done by running the
 777yast2 sysconfig configuration utility.  The goal is for to create an
 778ifcfg-id file for each slave device.  The simplest way to accomplish
 779this is to configure the devices for DHCP (this is only to get the
 780file ifcfg-id file created; see below for some issues with DHCP).  The
 781name of the configuration file for each device will be of the form:
 782
 783ifcfg-id-xx:xx:xx:xx:xx:xx
 784
 785        Where the "xx" portion will be replaced with the digits from
 786the device's permanent MAC address.
 787
 788        Once the set of ifcfg-id-xx:xx:xx:xx:xx:xx files has been
 789created, it is necessary to edit the configuration files for the slave
 790devices (the MAC addresses correspond to those of the slave devices).
 791Before editing, the file will contain multiple lines, and will look
 792something like this:
 793
 794BOOTPROTO='dhcp'
 795STARTMODE='on'
 796USERCTL='no'
 797UNIQUE='XNzu.WeZGOGF+4wE'
 798_nm_name='bus-pci-0001:61:01.0'
 799
 800        Change the BOOTPROTO and STARTMODE lines to the following:
 801
 802BOOTPROTO='none'
 803STARTMODE='off'
 804
 805        Do not alter the UNIQUE or _nm_name lines.  Remove any other
 806lines (USERCTL, etc).
 807
 808        Once the ifcfg-id-xx:xx:xx:xx:xx:xx files have been modified,
 809it's time to create the configuration file for the bonding device
 810itself.  This file is named ifcfg-bondX, where X is the number of the
 811bonding device to create, starting at 0.  The first such file is
 812ifcfg-bond0, the second is ifcfg-bond1, and so on.  The sysconfig
 813network configuration system will correctly start multiple instances
 814of bonding.
 815
 816        The contents of the ifcfg-bondX file is as follows:
 817
 818BOOTPROTO="static"
 819BROADCAST="10.0.2.255"
 820IPADDR="10.0.2.10"
 821NETMASK="255.255.0.0"
 822NETWORK="10.0.2.0"
 823REMOTE_IPADDR=""
 824STARTMODE="onboot"
 825BONDING_MASTER="yes"
 826BONDING_MODULE_OPTS="mode=active-backup miimon=100"
 827BONDING_SLAVE0="eth0"
 828BONDING_SLAVE1="bus-pci-0000:06:08.1"
 829
 830        Replace the sample BROADCAST, IPADDR, NETMASK and NETWORK
 831values with the appropriate values for your network.
 832
 833        The STARTMODE specifies when the device is brought online.
 834The possible values are:
 835
 836        onboot:  The device is started at boot time.  If you're not
 837                 sure, this is probably what you want.
 838
 839        manual:  The device is started only when ifup is called
 840                 manually.  Bonding devices may be configured this
 841                 way if you do not wish them to start automatically
 842                 at boot for some reason.
 843
 844        hotplug: The device is started by a hotplug event.  This is not
 845                 a valid choice for a bonding device.
 846
 847        off or ignore: The device configuration is ignored.
 848
 849        The line BONDING_MASTER='yes' indicates that the device is a
 850bonding master device.  The only useful value is "yes."
 851
 852        The contents of BONDING_MODULE_OPTS are supplied to the
 853instance of the bonding module for this device.  Specify the options
 854for the bonding mode, link monitoring, and so on here.  Do not include
 855the max_bonds bonding parameter; this will confuse the configuration
 856system if you have multiple bonding devices.
 857
 858        Finally, supply one BONDING_SLAVEn="slave device" for each
 859slave.  where "n" is an increasing value, one for each slave.  The
 860"slave device" is either an interface name, e.g., "eth0", or a device
 861specifier for the network device.  The interface name is easier to
 862find, but the ethN names are subject to change at boot time if, e.g.,
 863a device early in the sequence has failed.  The device specifiers
 864(bus-pci-0000:06:08.1 in the example above) specify the physical
 865network device, and will not change unless the device's bus location
 866changes (for example, it is moved from one PCI slot to another).  The
 867example above uses one of each type for demonstration purposes; most
 868configurations will choose one or the other for all slave devices.
 869
 870        When all configuration files have been modified or created,
 871networking must be restarted for the configuration changes to take
 872effect.  This can be accomplished via the following:
 873
 874# /etc/init.d/network restart
 875
 876        Note that the network control script (/sbin/ifdown) will
 877remove the bonding module as part of the network shutdown processing,
 878so it is not necessary to remove the module by hand if, e.g., the
 879module parameters have changed.
 880
 881        Also, at this writing, YaST/YaST2 will not manage bonding
 882devices (they do not show bonding interfaces on its list of network
 883devices).  It is necessary to edit the configuration file by hand to
 884change the bonding configuration.
 885
 886        Additional general options and details of the ifcfg file
 887format can be found in an example ifcfg template file:
 888
 889/etc/sysconfig/network/ifcfg.template
 890
 891        Note that the template does not document the various BONDING_
 892settings described above, but does describe many of the other options.
 893
 8943.1.1 Using DHCP with Sysconfig
 895-------------------------------
 896
 897        Under sysconfig, configuring a device with BOOTPROTO='dhcp'
 898will cause it to query DHCP for its IP address information.  At this
 899writing, this does not function for bonding devices; the scripts
 900attempt to obtain the device address from DHCP prior to adding any of
 901the slave devices.  Without active slaves, the DHCP requests are not
 902sent to the network.
 903
 9043.1.2 Configuring Multiple Bonds with Sysconfig
 905-----------------------------------------------
 906
 907        The sysconfig network initialization system is capable of
 908handling multiple bonding devices.  All that is necessary is for each
 909bonding instance to have an appropriately configured ifcfg-bondX file
 910(as described above).  Do not specify the "max_bonds" parameter to any
 911instance of bonding, as this will confuse sysconfig.  If you require
 912multiple bonding devices with identical parameters, create multiple
 913ifcfg-bondX files.
 914
 915        Because the sysconfig scripts supply the bonding module
 916options in the ifcfg-bondX file, it is not necessary to add them to
 917the system /etc/modules.conf or /etc/modprobe.conf configuration file.
 918
 9193.2 Configuration with Initscripts Support
 920------------------------------------------
 921
 922        This section applies to distros using a recent version of
 923initscripts with bonding support, for example, Red Hat Enterprise Linux
 924version 3 or later, Fedora, etc.  On these systems, the network
 925initialization scripts have knowledge of bonding, and can be configured to
 926control bonding devices.  Note that older versions of the initscripts
 927package have lower levels of support for bonding; this will be noted where
 928applicable.
 929
 930        These distros will not automatically load the network adapter
 931driver unless the ethX device is configured with an IP address.
 932Because of this constraint, users must manually configure a
 933network-script file for all physical adapters that will be members of
 934a bondX link.  Network script files are located in the directory:
 935
 936/etc/sysconfig/network-scripts
 937
 938        The file name must be prefixed with "ifcfg-eth" and suffixed
 939with the adapter's physical adapter number.  For example, the script
 940for eth0 would be named /etc/sysconfig/network-scripts/ifcfg-eth0.
 941Place the following text in the file:
 942
 943DEVICE=eth0
 944USERCTL=no
 945ONBOOT=yes
 946MASTER=bond0
 947SLAVE=yes
 948BOOTPROTO=none
 949
 950        The DEVICE= line will be different for every ethX device and
 951must correspond with the name of the file, i.e., ifcfg-eth1 must have
 952a device line of DEVICE=eth1.  The setting of the MASTER= line will
 953also depend on the final bonding interface name chosen for your bond.
 954As with other network devices, these typically start at 0, and go up
 955one for each device, i.e., the first bonding instance is bond0, the
 956second is bond1, and so on.
 957
 958        Next, create a bond network script.  The file name for this
 959script will be /etc/sysconfig/network-scripts/ifcfg-bondX where X is
 960the number of the bond.  For bond0 the file is named "ifcfg-bond0",
 961for bond1 it is named "ifcfg-bond1", and so on.  Within that file,
 962place the following text:
 963
 964DEVICE=bond0
 965IPADDR=192.168.1.1
 966NETMASK=255.255.255.0
 967NETWORK=192.168.1.0
 968BROADCAST=192.168.1.255
 969ONBOOT=yes
 970BOOTPROTO=none
 971USERCTL=no
 972
 973        Be sure to change the networking specific lines (IPADDR,
 974NETMASK, NETWORK and BROADCAST) to match your network configuration.
 975
 976        For later versions of initscripts, such as that found with Fedora
 9777 (or later) and Red Hat Enterprise Linux version 5 (or later), it is possible,
 978and, indeed, preferable, to specify the bonding options in the ifcfg-bond0
 979file, e.g. a line of the format:
 980
 981BONDING_OPTS="mode=active-backup arp_interval=60 arp_ip_target=192.168.1.254"
 982
 983        will configure the bond with the specified options.  The options
 984specified in BONDING_OPTS are identical to the bonding module parameters
 985except for the arp_ip_target field when using versions of initscripts older
 986than and 8.57 (Fedora 8) and 8.45.19 (Red Hat Enterprise Linux 5.2).  When
 987using older versions each target should be included as a separate option and
 988should be preceded by a '+' to indicate it should be added to the list of
 989queried targets, e.g.,
 990
 991        arp_ip_target=+192.168.1.1 arp_ip_target=+192.168.1.2
 992
 993        is the proper syntax to specify multiple targets.  When specifying
 994options via BONDING_OPTS, it is not necessary to edit /etc/modules.conf or
 995/etc/modprobe.conf.
 996
 997        For even older versions of initscripts that do not support
 998BONDING_OPTS, it is necessary to edit /etc/modules.conf (or
 999/etc/modprobe.conf, depending upon your distro) to load the bonding module
1000with your desired options when the bond0 interface is brought up.  The
1001following lines in /etc/modules.conf (or modprobe.conf) will load the
1002bonding module, and select its options:
1003
1004alias bond0 bonding
1005options bond0 mode=balance-alb miimon=100
1006
1007        Replace the sample parameters with the appropriate set of
1008options for your configuration.
1009
1010        Finally run "/etc/rc.d/init.d/network restart" as root.  This
1011will restart the networking subsystem and your bond link should be now
1012up and running.
1013
10143.2.1 Using DHCP with Initscripts
1015---------------------------------
1016
1017        Recent versions of initscripts (the versions supplied with Fedora
1018Core 3 and Red Hat Enterprise Linux 4, or later versions, are reported to
1019work) have support for assigning IP information to bonding devices via
1020DHCP.
1021
1022        To configure bonding for DHCP, configure it as described
1023above, except replace the line "BOOTPROTO=none" with "BOOTPROTO=dhcp"
1024and add a line consisting of "TYPE=Bonding".  Note that the TYPE value
1025is case sensitive.
1026
10273.2.2 Configuring Multiple Bonds with Initscripts
1028-------------------------------------------------
1029
1030        Initscripts packages that are included with Fedora 7 and Red Hat
1031Enterprise Linux 5 support multiple bonding interfaces by simply
1032specifying the appropriate BONDING_OPTS= in ifcfg-bondX where X is the
1033number of the bond.  This support requires sysfs support in the kernel,
1034and a bonding driver of version 3.0.0 or later.  Other configurations may
1035not support this method for specifying multiple bonding interfaces; for
1036those instances, see the "Configuring Multiple Bonds Manually" section,
1037below.
1038
10393.3 Configuring Bonding Manually with Ifenslave
1040-----------------------------------------------
1041
1042        This section applies to distros whose network initialization
1043scripts (the sysconfig or initscripts package) do not have specific
1044knowledge of bonding.  One such distro is SuSE Linux Enterprise Server
1045version 8.
1046
1047        The general method for these systems is to place the bonding
1048module parameters into /etc/modules.conf or /etc/modprobe.conf (as
1049appropriate for the installed distro), then add modprobe and/or
1050ifenslave commands to the system's global init script.  The name of
1051the global init script differs; for sysconfig, it is
1052/etc/init.d/boot.local and for initscripts it is /etc/rc.d/rc.local.
1053
1054        For example, if you wanted to make a simple bond of two e100
1055devices (presumed to be eth0 and eth1), and have it persist across
1056reboots, edit the appropriate file (/etc/init.d/boot.local or
1057/etc/rc.d/rc.local), and add the following:
1058
1059modprobe bonding mode=balance-alb miimon=100
1060modprobe e100
1061ifconfig bond0 192.168.1.1 netmask 255.255.255.0 up
1062ifenslave bond0 eth0
1063ifenslave bond0 eth1
1064
1065        Replace the example bonding module parameters and bond0
1066network configuration (IP address, netmask, etc) with the appropriate
1067values for your configuration.
1068
1069        Unfortunately, this method will not provide support for the
1070ifup and ifdown scripts on the bond devices.  To reload the bonding
1071configuration, it is necessary to run the initialization script, e.g.,
1072
1073# /etc/init.d/boot.local
1074
1075        or
1076
1077# /etc/rc.d/rc.local
1078
1079        It may be desirable in such a case to create a separate script
1080which only initializes the bonding configuration, then call that
1081separate script from within boot.local.  This allows for bonding to be
1082enabled without re-running the entire global init script.
1083
1084        To shut down the bonding devices, it is necessary to first
1085mark the bonding device itself as being down, then remove the
1086appropriate device driver modules.  For our example above, you can do
1087the following:
1088
1089# ifconfig bond0 down
1090# rmmod bonding
1091# rmmod e100
1092
1093        Again, for convenience, it may be desirable to create a script
1094with these commands.
1095
1096
10973.3.1 Configuring Multiple Bonds Manually
1098-----------------------------------------
1099
1100        This section contains information on configuring multiple
1101bonding devices with differing options for those systems whose network
1102initialization scripts lack support for configuring multiple bonds.
1103
1104        If you require multiple bonding devices, but all with the same
1105options, you may wish to use the "max_bonds" module parameter,
1106documented above.
1107
1108        To create multiple bonding devices with differing options, it is
1109preferrable to use bonding parameters exported by sysfs, documented in the
1110section below.
1111
1112        For versions of bonding without sysfs support, the only means to
1113provide multiple instances of bonding with differing options is to load
1114the bonding driver multiple times.  Note that current versions of the
1115sysconfig network initialization scripts handle this automatically; if
1116your distro uses these scripts, no special action is needed.  See the
1117section Configuring Bonding Devices, above, if you're not sure about your
1118network initialization scripts.
1119
1120        To load multiple instances of the module, it is necessary to
1121specify a different name for each instance (the module loading system
1122requires that every loaded module, even multiple instances of the same
1123module, have a unique name).  This is accomplished by supplying multiple
1124sets of bonding options in /etc/modprobe.conf, for example:
1125
1126alias bond0 bonding
1127options bond0 -o bond0 mode=balance-rr miimon=100
1128
1129alias bond1 bonding
1130options bond1 -o bond1 mode=balance-alb miimon=50
1131
1132        will load the bonding module two times.  The first instance is
1133named "bond0" and creates the bond0 device in balance-rr mode with an
1134miimon of 100.  The second instance is named "bond1" and creates the
1135bond1 device in balance-alb mode with an miimon of 50.
1136
1137        In some circumstances (typically with older distributions),
1138the above does not work, and the second bonding instance never sees
1139its options.  In that case, the second options line can be substituted
1140as follows:
1141
1142install bond1 /sbin/modprobe --ignore-install bonding -o bond1 \
1143        mode=balance-alb miimon=50
1144
1145        This may be repeated any number of times, specifying a new and
1146unique name in place of bond1 for each subsequent instance.
1147
1148        It has been observed that some Red Hat supplied kernels are unable
1149to rename modules at load time (the "-o bond1" part).  Attempts to pass
1150that option to modprobe will produce an "Operation not permitted" error.
1151This has been reported on some Fedora Core kernels, and has been seen on
1152RHEL 4 as well.  On kernels exhibiting this problem, it will be impossible
1153to configure multiple bonds with differing parameters (as they are older
1154kernels, and also lack sysfs support).
1155
11563.4 Configuring Bonding Manually via Sysfs
1157------------------------------------------
1158
1159        Starting with version 3.0.0, Channel Bonding may be configured
1160via the sysfs interface.  This interface allows dynamic configuration
1161of all bonds in the system without unloading the module.  It also
1162allows for adding and removing bonds at runtime.  Ifenslave is no
1163longer required, though it is still supported.
1164
1165        Use of the sysfs interface allows you to use multiple bonds
1166with different configurations without having to reload the module.
1167It also allows you to use multiple, differently configured bonds when
1168bonding is compiled into the kernel.
1169
1170        You must have the sysfs filesystem mounted to configure
1171bonding this way.  The examples in this document assume that you
1172are using the standard mount point for sysfs, e.g. /sys.  If your
1173sysfs filesystem is mounted elsewhere, you will need to adjust the
1174example paths accordingly.
1175
1176Creating and Destroying Bonds
1177-----------------------------
1178To add a new bond foo:
1179# echo +foo > /sys/class/net/bonding_masters
1180
1181To remove an existing bond bar:
1182# echo -bar > /sys/class/net/bonding_masters
1183
1184To show all existing bonds:
1185# cat /sys/class/net/bonding_masters
1186
1187NOTE: due to 4K size limitation of sysfs files, this list may be
1188truncated if you have more than a few hundred bonds.  This is unlikely
1189to occur under normal operating conditions.
1190
1191Adding and Removing Slaves
1192--------------------------
1193        Interfaces may be enslaved to a bond using the file
1194/sys/class/net/<bond>/bonding/slaves.  The semantics for this file
1195are the same as for the bonding_masters file.
1196
1197To enslave interface eth0 to bond bond0:
1198# ifconfig bond0 up
1199# echo +eth0 > /sys/class/net/bond0/bonding/slaves
1200
1201To free slave eth0 from bond bond0:
1202# echo -eth0 > /sys/class/net/bond0/bonding/slaves
1203
1204        When an interface is enslaved to a bond, symlinks between the
1205two are created in the sysfs filesystem.  In this case, you would get
1206/sys/class/net/bond0/slave_eth0 pointing to /sys/class/net/eth0, and
1207/sys/class/net/eth0/master pointing to /sys/class/net/bond0.
1208
1209        This means that you can tell quickly whether or not an
1210interface is enslaved by looking for the master symlink.  Thus:
1211# echo -eth0 > /sys/class/net/eth0/master/bonding/slaves
1212will free eth0 from whatever bond it is enslaved to, regardless of
1213the name of the bond interface.
1214
1215Changing a Bond's Configuration
1216-------------------------------
1217        Each bond may be configured individually by manipulating the
1218files located in /sys/class/net/<bond name>/bonding
1219
1220        The names of these files correspond directly with the command-
1221line parameters described elsewhere in this file, and, with the
1222exception of arp_ip_target, they accept the same values.  To see the
1223current setting, simply cat the appropriate file.
1224
1225        A few examples will be given here; for specific usage
1226guidelines for each parameter, see the appropriate section in this
1227document.
1228
1229To configure bond0 for balance-alb mode:
1230# ifconfig bond0 down
1231# echo 6 > /sys/class/net/bond0/bonding/mode
1232 - or -
1233# echo balance-alb > /sys/class/net/bond0/bonding/mode
1234        NOTE: The bond interface must be down before the mode can be
1235changed.
1236
1237To enable MII monitoring on bond0 with a 1 second interval:
1238# echo 1000 > /sys/class/net/bond0/bonding/miimon
1239        NOTE: If ARP monitoring is enabled, it will disabled when MII
1240monitoring is enabled, and vice-versa.
1241
1242To add ARP targets:
1243# echo +192.168.0.100 > /sys/class/net/bond0/bonding/arp_ip_target
1244# echo +192.168.0.101 > /sys/class/net/bond0/bonding/arp_ip_target
1245        NOTE:  up to 16 target addresses may be specified.
1246
1247To remove an ARP target:
1248# echo -192.168.0.100 > /sys/class/net/bond0/bonding/arp_ip_target
1249
1250Example Configuration
1251---------------------
1252        We begin with the same example that is shown in section 3.3,
1253executed with sysfs, and without using ifenslave.
1254
1255        To make a simple bond of two e100 devices (presumed to be eth0
1256and eth1), and have it persist across reboots, edit the appropriate
1257file (/etc/init.d/boot.local or /etc/rc.d/rc.local), and add the
1258following:
1259
1260modprobe bonding
1261modprobe e100
1262echo balance-alb > /sys/class/net/bond0/bonding/mode
1263ifconfig bond0 192.168.1.1 netmask 255.255.255.0 up
1264echo 100 > /sys/class/net/bond0/bonding/miimon
1265echo +eth0 > /sys/class/net/bond0/bonding/slaves
1266echo +eth1 > /sys/class/net/bond0/bonding/slaves
1267
1268        To add a second bond, with two e1000 interfaces in
1269active-backup mode, using ARP monitoring, add the following lines to
1270your init script:
1271
1272modprobe e1000
1273echo +bond1 > /sys/class/net/bonding_masters
1274echo active-backup > /sys/class/net/bond1/bonding/mode
1275ifconfig bond1 192.168.2.1 netmask 255.255.255.0 up
1276echo +192.168.2.100 /sys/class/net/bond1/bonding/arp_ip_target
1277echo 2000 > /sys/class/net/bond1/bonding/arp_interval
1278echo +eth2 > /sys/class/net/bond1/bonding/slaves
1279echo +eth3 > /sys/class/net/bond1/bonding/slaves
1280
1281
12824. Querying Bonding Configuration 
1283=================================
1284
12854.1 Bonding Configuration
1286-------------------------
1287
1288        Each bonding device has a read-only file residing in the
1289/proc/net/bonding directory.  The file contents include information
1290about the bonding configuration, options and state of each slave.
1291
1292        For example, the contents of /proc/net/bonding/bond0 after the
1293driver is loaded with parameters of mode=0 and miimon=1000 is
1294generally as follows:
1295
1296        Ethernet Channel Bonding Driver: 2.6.1 (October 29, 2004)
1297        Bonding Mode: load balancing (round-robin)
1298        Currently Active Slave: eth0
1299        MII Status: up
1300        MII Polling Interval (ms): 1000
1301        Up Delay (ms): 0
1302        Down Delay (ms): 0
1303
1304        Slave Interface: eth1
1305        MII Status: up
1306        Link Failure Count: 1
1307
1308        Slave Interface: eth0
1309        MII Status: up
1310        Link Failure Count: 1
1311
1312        The precise format and contents will change depending upon the
1313bonding configuration, state, and version of the bonding driver.
1314
13154.2 Network configuration
1316-------------------------
1317
1318        The network configuration can be inspected using the ifconfig
1319command.  Bonding devices will have the MASTER flag set; Bonding slave
1320devices will have the SLAVE flag set.  The ifconfig output does not
1321contain information on which slaves are associated with which masters.
1322
1323        In the example below, the bond0 interface is the master
1324(MASTER) while eth0 and eth1 are slaves (SLAVE). Notice all slaves of
1325bond0 have the same MAC address (HWaddr) as bond0 for all modes except
1326TLB and ALB that require a unique MAC address for each slave.
1327
1328# /sbin/ifconfig
1329bond0     Link encap:Ethernet  HWaddr 00:C0:F0:1F:37:B4
1330          inet addr:XXX.XXX.XXX.YYY  Bcast:XXX.XXX.XXX.255  Mask:255.255.252.0
1331          UP BROADCAST RUNNING MASTER MULTICAST  MTU:1500  Metric:1
1332          RX packets:7224794 errors:0 dropped:0 overruns:0 frame:0
1333          TX packets:3286647 errors:1 dropped:0 overruns:1 carrier:0
1334          collisions:0 txqueuelen:0
1335
1336eth0      Link encap:Ethernet  HWaddr 00:C0:F0:1F:37:B4
1337          UP BROADCAST RUNNING SLAVE MULTICAST  MTU:1500  Metric:1
1338          RX packets:3573025 errors:0 dropped:0 overruns:0 frame:0
1339          TX packets:1643167 errors:1 dropped:0 overruns:1 carrier:0
1340          collisions:0 txqueuelen:100
1341          Interrupt:10 Base address:0x1080
1342
1343eth1      Link encap:Ethernet  HWaddr 00:C0:F0:1F:37:B4
1344          UP BROADCAST RUNNING SLAVE MULTICAST  MTU:1500  Metric:1
1345          RX packets:3651769 errors:0 dropped:0 overruns:0 frame:0
1346          TX packets:1643480 errors:0 dropped:0 overruns:0 carrier:0
1347          collisions:0 txqueuelen:100
1348          Interrupt:9 Base address:0x1400
1349
13505. Switch Configuration
1351=======================
1352
1353        For this section, "switch" refers to whatever system the
1354bonded devices are directly connected to (i.e., where the other end of
1355the cable plugs into).  This may be an actual dedicated switch device,
1356or it may be another regular system (e.g., another computer running
1357Linux),
1358
1359        The active-backup, balance-tlb and balance-alb modes do not
1360require any specific configuration of the switch.
1361
1362        The 802.3ad mode requires that the switch have the appropriate
1363ports configured as an 802.3ad aggregation.  The precise method used
1364to configure this varies from switch to switch, but, for example, a
1365Cisco 3550 series switch requires that the appropriate ports first be
1366grouped together in a single etherchannel instance, then that
1367etherchannel is set to mode "lacp" to enable 802.3ad (instead of
1368standard EtherChannel).
1369
1370        The balance-rr, balance-xor and broadcast modes generally
1371require that the switch have the appropriate ports grouped together.
1372The nomenclature for such a group differs between switches, it may be
1373called an "etherchannel" (as in the Cisco example, above), a "trunk
1374group" or some other similar variation.  For these modes, each switch
1375will also have its own configuration options for the switch's transmit
1376policy to the bond.  Typical choices include XOR of either the MAC or
1377IP addresses.  The transmit policy of the two peers does not need to
1378match.  For these three modes, the bonding mode really selects a
1379transmit policy for an EtherChannel group; all three will interoperate
1380with another EtherChannel group.
1381
1382
13836. 802.1q VLAN Support
1384======================
1385
1386        It is possible to configure VLAN devices over a bond interface
1387using the 8021q driver.  However, only packets coming from the 8021q
1388driver and passing through bonding will be tagged by default.  Self
1389generated packets, for example, bonding's learning packets or ARP
1390packets generated by either ALB mode or the ARP monitor mechanism, are
1391tagged internally by bonding itself.  As a result, bonding must
1392"learn" the VLAN IDs configured above it, and use those IDs to tag
1393self generated packets.
1394
1395        For reasons of simplicity, and to support the use of adapters
1396that can do VLAN hardware acceleration offloading, the bonding
1397interface declares itself as fully hardware offloading capable, it gets
1398the add_vid/kill_vid notifications to gather the necessary
1399information, and it propagates those actions to the slaves.  In case
1400of mixed adapter types, hardware accelerated tagged packets that
1401should go through an adapter that is not offloading capable are
1402"un-accelerated" by the bonding driver so the VLAN tag sits in the
1403regular location.
1404
1405        VLAN interfaces *must* be added on top of a bonding interface
1406only after enslaving at least one slave.  The bonding interface has a
1407hardware address of 00:00:00:00:00:00 until the first slave is added.
1408If the VLAN interface is created prior to the first enslavement, it
1409would pick up the all-zeroes hardware address.  Once the first slave
1410is attached to the bond, the bond device itself will pick up the
1411slave's hardware address, which is then available for the VLAN device.
1412
1413        Also, be aware that a similar problem can occur if all slaves
1414are released from a bond that still has one or more VLAN interfaces on
1415top of it.  When a new slave is added, the bonding interface will
1416obtain its hardware address from the first slave, which might not
1417match the hardware address of the VLAN interfaces (which was
1418ultimately copied from an earlier slave).
1419
1420        There are two methods to insure that the VLAN device operates
1421with the correct hardware address if all slaves are removed from a
1422bond interface:
1423
1424        1. Remove all VLAN interfaces then recreate them
1425
1426        2. Set the bonding interface's hardware address so that it
1427matches the hardware address of the VLAN interfaces.
1428
1429        Note that changing a VLAN interface's HW address would set the
1430underlying device -- i.e. the bonding interface -- to promiscuous
1431mode, which might not be what you want.
1432
1433
14347. Link Monitoring
1435==================
1436
1437        The bonding driver at present supports two schemes for
1438monitoring a slave device's link state: the ARP monitor and the MII
1439monitor.
1440
1441        At the present time, due to implementation restrictions in the
1442bonding driver itself, it is not possible to enable both ARP and MII
1443monitoring simultaneously.
1444
14457.1 ARP Monitor Operation
1446-------------------------
1447
1448        The ARP monitor operates as its name suggests: it sends ARP
1449queries to one or more designated peer systems on the network, and
1450uses the response as an indication that the link is operating.  This
1451gives some assurance that traffic is actually flowing to and from one
1452or more peers on the local network.
1453
1454        The ARP monitor relies on the device driver itself to verify
1455that traffic is flowing.  In particular, the driver must keep up to
1456date the last receive time, dev->last_rx, and transmit start time,
1457dev->trans_start.  If these are not updated by the driver, then the
1458ARP monitor will immediately fail any slaves using that driver, and
1459those slaves will stay down.  If networking monitoring (tcpdump, etc)
1460shows the ARP requests and replies on the network, then it may be that
1461your device driver is not updating last_rx and trans_start.
1462
14637.2 Configuring Multiple ARP Targets
1464------------------------------------
1465
1466        While ARP monitoring can be done with just one target, it can
1467be useful in a High Availability setup to have several targets to
1468monitor.  In the case of just one target, the target itself may go
1469down or have a problem making it unresponsive to ARP requests.  Having
1470an additional target (or several) increases the reliability of the ARP
1471monitoring.
1472
1473        Multiple ARP targets must be separated by commas as follows:
1474
1475# example options for ARP monitoring with three targets
1476alias bond0 bonding
1477options bond0 arp_interval=60 arp_ip_target=192.168.0.1,192.168.0.3,192.168.0.9
1478
1479        For just a single target the options would resemble:
1480
1481# example options for ARP monitoring with one target
1482alias bond0 bonding
1483options bond0 arp_interval=60 arp_ip_target=192.168.0.100
1484
1485
14867.3 MII Monitor Operation
1487-------------------------
1488
1489        The MII monitor monitors only the carrier state of the local
1490network interface.  It accomplishes this in one of three ways: by
1491depending upon the device driver to maintain its carrier state, by
1492querying the device's MII registers, or by making an ethtool query to
1493the device.
1494
1495        If the use_carrier module parameter is 1 (the default value),
1496then the MII monitor will rely on the driver for carrier state
1497information (via the netif_carrier subsystem).  As explained in the
1498use_carrier parameter information, above, if the MII monitor fails to
1499detect carrier loss on the device (e.g., when the cable is physically
1500disconnected), it may be that the driver does not support
1501netif_carrier.
1502
1503        If use_carrier is 0, then the MII monitor will first query the
1504device's (via ioctl) MII registers and check the link state.  If that
1505request fails (not just that it returns carrier down), then the MII
1506monitor will make an ethtool ETHOOL_GLINK request to attempt to obtain
1507the same information.  If both methods fail (i.e., the driver either
1508does not support or had some error in processing both the MII register
1509and ethtool requests), then the MII monitor will assume the link is
1510up.
1511
15128. Potential Sources of Trouble
1513===============================
1514
15158.1 Adventures in Routing
1516-------------------------
1517
1518        When bonding is configured, it is important that the slave
1519devices not have routes that supersede routes of the master (or,
1520generally, not have routes at all).  For example, suppose the bonding
1521device bond0 has two slaves, eth0 and eth1, and the routing table is
1522as follows:
1523
1524Kernel IP routing table
1525Destination     Gateway         Genmask         Flags   MSS Window  irtt Iface
152610.0.0.0        0.0.0.0         255.255.0.0     U        40 0          0 eth0
152710.0.0.0        0.0.0.0         255.255.0.0     U        40 0          0 eth1
152810.0.0.0        0.0.0.0         255.255.0.0     U        40 0          0 bond0
1529127.0.0.0       0.0.0.0         255.0.0.0       U        40 0          0 lo
1530
1531        This routing configuration will likely still update the
1532receive/transmit times in the driver (needed by the ARP monitor), but
1533may bypass the bonding driver (because outgoing traffic to, in this
1534case, another host on network 10 would use eth0 or eth1 before bond0).
1535
1536        The ARP monitor (and ARP itself) may become confused by this
1537configuration, because ARP requests (generated by the ARP monitor)
1538will be sent on one interface (bond0), but the corresponding reply
1539will arrive on a different interface (eth0).  This reply looks to ARP
1540as an unsolicited ARP reply (because ARP matches replies on an
1541interface basis), and is discarded.  The MII monitor is not affected
1542by the state of the routing table.
1543
1544        The solution here is simply to insure that slaves do not have
1545routes of their own, and if for some reason they must, those routes do
1546not supersede routes of their master.  This should generally be the
1547case, but unusual configurations or errant manual or automatic static
1548route additions may cause trouble.
1549
15508.2 Ethernet Device Renaming
1551----------------------------
1552
1553        On systems with network configuration scripts that do not
1554associate physical devices directly with network interface names (so
1555that the same physical device always has the same "ethX" name), it may
1556be necessary to add some special logic to either /etc/modules.conf or
1557/etc/modprobe.conf (depending upon which is installed on the system).
1558
1559        For example, given a modules.conf containing the following:
1560
1561alias bond0 bonding
1562options bond0 mode=some-mode miimon=50
1563alias eth0 tg3
1564alias eth1 tg3
1565alias eth2 e1000
1566alias eth3 e1000
1567
1568        If neither eth0 and eth1 are slaves to bond0, then when the
1569bond0 interface comes up, the devices may end up reordered.  This
1570happens because bonding is loaded first, then its slave device's
1571drivers are loaded next.  Since no other drivers have been loaded,
1572when the e1000 driver loads, it will receive eth0 and eth1 for its
1573devices, but the bonding configuration tries to enslave eth2 and eth3
1574(which may later be assigned to the tg3 devices).
1575
1576        Adding the following:
1577
1578add above bonding e1000 tg3
1579
1580        causes modprobe to load e1000 then tg3, in that order, when
1581bonding is loaded.  This command is fully documented in the
1582modules.conf manual page.
1583
1584        On systems utilizing modprobe.conf (or modprobe.conf.local),
1585an equivalent problem can occur.  In this case, the following can be
1586added to modprobe.conf (or modprobe.conf.local, as appropriate), as
1587follows (all on one line; it has been split here for clarity):
1588
1589install bonding /sbin/modprobe tg3; /sbin/modprobe e1000;
1590        /sbin/modprobe --ignore-install bonding
1591
1592        This will, when loading the bonding module, rather than
1593performing the normal action, instead execute the provided command.
1594This command loads the device drivers in the order needed, then calls
1595modprobe with --ignore-install to cause the normal action to then take
1596place.  Full documentation on this can be found in the modprobe.conf
1597and modprobe manual pages.
1598
15998.3. Painfully Slow Or No Failed Link Detection By Miimon
1600---------------------------------------------------------
1601
1602        By default, bonding enables the use_carrier option, which
1603instructs bonding to trust the driver to maintain carrier state.
1604
1605        As discussed in the options section, above, some drivers do
1606not support the netif_carrier_on/_off link state tracking system.
1607With use_carrier enabled, bonding will always see these links as up,
1608regardless of their actual state.
1609
1610        Additionally, other drivers do support netif_carrier, but do
1611not maintain it in real time, e.g., only polling the link state at
1612some fixed interval.  In this case, miimon will detect failures, but
1613only after some long period of time has expired.  If it appears that
1614miimon is very slow in detecting link failures, try specifying
1615use_carrier=0 to see if that improves the failure detection time.  If
1616it does, then it may be that the driver checks the carrier state at a
1617fixed interval, but does not cache the MII register values (so the
1618use_carrier=0 method of querying the registers directly works).  If
1619use_carrier=0 does not improve the failover, then the driver may cache
1620the registers, or the problem may be elsewhere.
1621
1622        Also, remember that miimon only checks for the device's
1623carrier state.  It has no way to determine the state of devices on or
1624beyond other ports of a switch, or if a switch is refusing to pass
1625traffic while still maintaining carrier on.
1626
16279. SNMP agents
1628===============
1629
1630        If running SNMP agents, the bonding driver should be loaded
1631before any network drivers participating in a bond.  This requirement
1632is due to the interface index (ipAdEntIfIndex) being associated to
1633the first interface found with a given IP address.  That is, there is
1634only one ipAdEntIfIndex for each IP address.  For example, if eth0 and
1635eth1 are slaves of bond0 and the driver for eth0 is loaded before the
1636bonding driver, the interface for the IP address will be associated
1637with the eth0 interface.  This configuration is shown below, the IP
1638address 192.168.1.1 has an interface index of 2 which indexes to eth0
1639in the ifDescr table (ifDescr.2).
1640
1641     interfaces.ifTable.ifEntry.ifDescr.1 = lo
1642     interfaces.ifTable.ifEntry.ifDescr.2 = eth0
1643     interfaces.ifTable.ifEntry.ifDescr.3 = eth1
1644     interfaces.ifTable.ifEntry.ifDescr.4 = eth2
1645     interfaces.ifTable.ifEntry.ifDescr.5 = eth3
1646     interfaces.ifTable.ifEntry.ifDescr.6 = bond0
1647     ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.10.10.10 = 5
1648     ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.192.168.1.1 = 2
1649     ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.74.20.94 = 4
1650     ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.127.0.0.1 = 1
1651
1652        This problem is avoided by loading the bonding driver before
1653any network drivers participating in a bond.  Below is an example of
1654loading the bonding driver first, the IP address 192.168.1.1 is
1655correctly associated with ifDescr.2.
1656
1657     interfaces.ifTable.ifEntry.ifDescr.1 = lo
1658     interfaces.ifTable.ifEntry.ifDescr.2 = bond0
1659     interfaces.ifTable.ifEntry.ifDescr.3 = eth0
1660     interfaces.ifTable.ifEntry.ifDescr.4 = eth1
1661     interfaces.ifTable.ifEntry.ifDescr.5 = eth2
1662     interfaces.ifTable.ifEntry.ifDescr.6 = eth3
1663     ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.10.10.10 = 6
1664     ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.192.168.1.1 = 2
1665     ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.74.20.94 = 5
1666     ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.127.0.0.1 = 1
1667
1668        While some distributions may not report the interface name in
1669ifDescr, the association between the IP address and IfIndex remains
1670and SNMP functions such as Interface_Scan_Next will report that
1671association.
1672
167310. Promiscuous mode
1674====================
1675
1676        When running network monitoring tools, e.g., tcpdump, it is
1677common to enable promiscuous mode on the device, so that all traffic
1678is seen (instead of seeing only traffic destined for the local host).
1679The bonding driver handles promiscuous mode changes to the bonding
1680master device (e.g., bond0), and propagates the setting to the slave
1681devices.
1682
1683        For the balance-rr, balance-xor, broadcast, and 802.3ad modes,
1684the promiscuous mode setting is propagated to all slaves.
1685
1686        For the active-backup, balance-tlb and balance-alb modes, the
1687promiscuous mode setting is propagated only to the active slave.
1688
1689        For balance-tlb mode, the active slave is the slave currently
1690receiving inbound traffic.
1691
1692        For balance-alb mode, the active slave is the slave used as a
1693"primary."  This slave is used for mode-specific control traffic, for
1694sending to peers that are unassigned or if the load is unbalanced.
1695
1696        For the active-backup, balance-tlb and balance-alb modes, when
1697the active slave changes (e.g., due to a link failure), the
1698promiscuous setting will be propagated to the new active slave.
1699
170011. Configuring Bonding for High Availability
1701=============================================
1702
1703        High Availability refers to configurations that provide
1704maximum network availability by having redundant or backup devices,
1705links or switches between the host and the rest of the world.  The
1706goal is to provide the maximum availability of network connectivity
1707(i.e., the network always works), even though other configurations
1708could provide higher throughput.
1709
171011.1 High Availability in a Single Switch Topology
1711--------------------------------------------------
1712
1713        If two hosts (or a host and a single switch) are directly
1714connected via multiple physical links, then there is no availability
1715penalty to optimizing for maximum bandwidth.  In this case, there is
1716only one switch (or peer), so if it fails, there is no alternative
1717access to fail over to.  Additionally, the bonding load balance modes
1718support link monitoring of their members, so if individual links fail,
1719the load will be rebalanced across the remaining devices.
1720
1721        See Section 13, "Configuring Bonding for Maximum Throughput"
1722for information on configuring bonding with one peer device.
1723
172411.2 High Availability in a Multiple Switch Topology
1725----------------------------------------------------
1726
1727        With multiple switches, the configuration of bonding and the
1728network changes dramatically.  In multiple switch topologies, there is
1729a trade off between network availability and usable bandwidth.
1730
1731        Below is a sample network, configured to maximize the
1732availability of the network:
1733
1734                |                                     |
1735                |port3                           port3|
1736          +-----+----+                          +-----+----+
1737          |          |port2       ISL      port2|          |
1738          | switch A +--------------------------+ switch B |
1739          |          |                          |          |
1740          +-----+----+                          +-----++---+
1741                |port1                           port1|
1742                |             +-------+               |
1743                +-------------+ host1 +---------------+
1744                         eth0 +-------+ eth1
1745
1746        In this configuration, there is a link between the two
1747switches (ISL, or inter switch link), and multiple ports connecting to
1748the outside world ("port3" on each switch).  There is no technical
1749reason that this could not be extended to a third switch.
1750
175111.2.1 HA Bonding Mode Selection for Multiple Switch Topology
1752-------------------------------------------------------------
1753
1754        In a topology such as the example above, the active-backup and
1755broadcast modes are the only useful bonding modes when optimizing for
1756availability; the other modes require all links to terminate on the
1757same peer for them to behave rationally.
1758
1759active-backup: This is generally the preferred mode, particularly if
1760        the switches have an ISL and play together well.  If the
1761        network configuration is such that one switch is specifically
1762        a backup switch (e.g., has lower capacity, higher cost, etc),
1763        then the primary option can be used to insure that the
1764        preferred link is always used when it is available.
1765
1766broadcast: This mode is really a special purpose mode, and is suitable
1767        only for very specific needs.  For example, if the two
1768        switches are not connected (no ISL), and the networks beyond
1769        them are totally independent.  In this case, if it is
1770        necessary for some specific one-way traffic to reach both
1771        independent networks, then the broadcast mode may be suitable.
1772
177311.2.2 HA Link Monitoring Selection for Multiple Switch Topology
1774----------------------------------------------------------------
1775
1776        The choice of link monitoring ultimately depends upon your
1777switch.  If the switch can reliably fail ports in response to other
1778failures, then either the MII or ARP monitors should work.  For
1779example, in the above example, if the "port3" link fails at the remote
1780end, the MII monitor has no direct means to detect this.  The ARP
1781monitor could be configured with a target at the remote end of port3,
1782thus detecting that failure without switch support.
1783
1784        In general, however, in a multiple switch topology, the ARP
1785monitor can provide a higher level of reliability in detecting end to
1786end connectivity failures (which may be caused by the failure of any
1787individual component to pass traffic for any reason).  Additionally,
1788the ARP monitor should be configured with multiple targets (at least
1789one for each switch in the network).  This will insure that,
1790regardless of which switch is active, the ARP monitor has a suitable
1791target to query.
1792
1793        Note, also, that of late many switches now support a functionality
1794generally referred to as "trunk failover."  This is a feature of the
1795switch that causes the link state of a particular switch port to be set
1796down (or up) when the state of another switch port goes down (or up).
1797It's purpose is to propogate link failures from logically "exterior" ports
1798to the logically "interior" ports that bonding is able to monitor via
1799miimon.  Availability and configuration for trunk failover varies by
1800switch, but this can be a viable alternative to the ARP monitor when using
1801suitable switches.
1802
180312. Configuring Bonding for Maximum Throughput
1804==============================================
1805
180612.1 Maximizing Throughput in a Single Switch Topology
1807------------------------------------------------------
1808
1809        In a single switch configuration, the best method to maximize
1810throughput depends upon the application and network environment.  The
1811various load balancing modes each have strengths and weaknesses in
1812different environments, as detailed below.
1813
1814        For this discussion, we will break down the topologies into
1815two categories.  Depending upon the destination of most traffic, we
1816categorize them into either "gatewayed" or "local" configurations.
1817
1818        In a gatewayed configuration, the "switch" is acting primarily
1819as a router, and the majority of traffic passes through this router to
1820other networks.  An example would be the following:
1821
1822
1823     +----------+                     +----------+
1824     |          |eth0            port1|          | to other networks
1825     | Host A   +---------------------+ router   +------------------->
1826     |          +---------------------+          | Hosts B and C are out
1827     |          |eth1            port2|          | here somewhere
1828     +----------+                     +----------+
1829
1830        The router may be a dedicated router device, or another host
1831acting as a gateway.  For our discussion, the important point is that
1832the majority of traffic from Host A will pass through the router to
1833some other network before reaching its final destination.
1834
1835        In a gatewayed network configuration, although Host A may
1836communicate with many other systems, all of its traffic will be sent
1837and received via one other peer on the local network, the router.
1838
1839        Note that the case of two systems connected directly via
1840multiple physical links is, for purposes of configuring bonding, the
1841same as a gatewayed configuration.  In that case, it happens that all
1842traffic is destined for the "gateway" itself, not some other network
1843beyond the gateway.
1844
1845        In a local configuration, the "switch" is acting primarily as
1846a switch, and the majority of traffic passes through this switch to
1847reach other stations on the same network.  An example would be the
1848following:
1849
1850    +----------+            +----------+       +--------+
1851    |          |eth0   port1|          +-------+ Host B |
1852    |  Host A  +------------+  switch  |port3  +--------+
1853    |          +------------+          |                  +--------+
1854    |          |eth1   port2|          +------------------+ Host C |
1855    +----------+            +----------+port4             +--------+
1856
1857
1858        Again, the switch may be a dedicated switch device, or another
1859host acting as a gateway.  For our discussion, the important point is
1860that the majority of traffic from Host A is destined for other hosts
1861on the same local network (Hosts B and C in the above example).
1862
1863        In summary, in a gatewayed configuration, traffic to and from
1864the bonded device will be to the same MAC level peer on the network
1865(the gateway itself, i.e., the router), regardless of its final
1866destination.  In a local configuration, traffic flows directly to and
1867from the final destinations, thus, each destination (Host B, Host C)
1868will be addressed directly by their individual MAC addresses.
1869
1870        This distinction between a gatewayed and a local network
1871configuration is important because many of the load balancing modes
1872available use the MAC addresses of the local network source and
1873destination to make load balancing decisions.  The behavior of each
1874mode is described below.
1875
1876
187712.1.1 MT Bonding Mode Selection for Single Switch Topology
1878-----------------------------------------------------------
1879
1880        This configuration is the easiest to set up and to understand,
1881although you will have to decide which bonding mode best suits your
1882needs.  The trade offs for each mode are detailed below:
1883
1884balance-rr: This mode is the only mode that will permit a single
1885        TCP/IP connection to stripe traffic across multiple
1886        interfaces. It is therefore the only mode that will allow a
1887        single TCP/IP stream to utilize more than one interface's
1888        worth of throughput.  This comes at a cost, however: the
1889        striping generally results in peer systems receiving packets out
1890        of order, causing TCP/IP's congestion control system to kick
1891        in, often by retransmitting segments.
1892
1893        It is possible to adjust TCP/IP's congestion limits by
1894        altering the net.ipv4.tcp_reordering sysctl parameter.  The
1895        usual default value is 3, and the maximum useful value is 127.
1896        For a four interface balance-rr bond, expect that a single
1897        TCP/IP stream will utilize no more than approximately 2.3
1898        interface's worth of throughput, even after adjusting
1899        tcp_reordering.
1900
1901        Note that the fraction of packets that will be delivered out of
1902        order is highly variable, and is unlikely to be zero.  The level
1903        of reordering depends upon a variety of factors, including the
1904        networking interfaces, the switch, and the topology of the
1905        configuration.  Speaking in general terms, higher speed network
1906        cards produce more reordering (due to factors such as packet
1907        coalescing), and a "many to many" topology will reorder at a
1908        higher rate than a "many slow to one fast" configuration.
1909
1910        Many switches do not support any modes that stripe traffic
1911        (instead choosing a port based upon IP or MAC level addresses);
1912        for those devices, traffic for a particular connection flowing
1913        through the switch to a balance-rr bond will not utilize greater
1914        than one interface's worth of bandwidth.
1915
1916        If you are utilizing protocols other than TCP/IP, UDP for
1917        example, and your application can tolerate out of order
1918        delivery, then this mode can allow for single stream datagram
1919        performance that scales near linearly as interfaces are added
1920        to the bond.
1921
1922        This mode requires the switch to have the appropriate ports
1923        configured for "etherchannel" or "trunking."
1924
1925active-backup: There is not much advantage in this network topology to
1926        the active-backup mode, as the inactive backup devices are all
1927        connected to the same peer as the primary.  In this case, a
1928        load balancing mode (with link monitoring) will provide the
1929        same level of network availability, but with increased
1930        available bandwidth.  On the plus side, active-backup mode
1931        does not require any configuration of the switch, so it may
1932        have value if the hardware available does not support any of
1933        the load balance modes.
1934
1935balance-xor: This mode will limit traffic such that packets destined
1936        for specific peers will always be sent over the same
1937        interface.  Since the destination is determined by the MAC
1938        addresses involved, this mode works best in a "local" network
1939        configuration (as described above), with destinations all on
1940        the same local network.  This mode is likely to be suboptimal
1941        if all your traffic is passed through a single router (i.e., a
1942        "gatewayed" network configuration, as described above).
1943
1944        As with balance-rr, the switch ports need to be configured for
1945        "etherchannel" or "trunking."
1946
1947broadcast: Like active-backup, there is not much advantage to this
1948        mode in this type of network topology.
1949
1950802.3ad: This mode can be a good choice for this type of network
1951        topology.  The 802.3ad mode is an IEEE standard, so all peers
1952        that implement 802.3ad should interoperate well.  The 802.3ad
1953        protocol includes automatic configuration of the aggregates,
1954        so minimal manual configuration of the switch is needed
1955        (typically only to designate that some set of devices is
1956        available for 802.3ad).  The 802.3ad standard also mandates
1957        that frames be delivered in order (within certain limits), so
1958        in general single connections will not see misordering of
1959        packets.  The 802.3ad mode does have some drawbacks: the
1960        standard mandates that all devices in the aggregate operate at
1961        the same speed and duplex.  Also, as with all bonding load
1962        balance modes other than balance-rr, no single connection will
1963        be able to utilize more than a single interface's worth of
1964        bandwidth.  
1965
1966        Additionally, the linux bonding 802.3ad implementation
1967        distributes traffic by peer (using an XOR of MAC addresses),
1968        so in a "gatewayed" configuration, all outgoing traffic will
1969        generally use the same device.  Incoming traffic may also end
1970        up on a single device, but that is dependent upon the
1971        balancing policy of the peer's 8023.ad implementation.  In a
1972        "local" configuration, traffic will be distributed across the
1973        devices in the bond.
1974
1975        Finally, the 802.3ad mode mandates the use of the MII monitor,
1976        therefore, the ARP monitor is not available in this mode.
1977
1978balance-tlb: The balance-tlb mode balances outgoing traffic by peer.
1979        Since the balancing is done according to MAC address, in a
1980        "gatewayed" configuration (as described above), this mode will
1981        send all traffic across a single device.  However, in a
1982        "local" network configuration, this mode balances multiple
1983        local network peers across devices in a vaguely intelligent
1984        manner (not a simple XOR as in balance-xor or 802.3ad mode),
1985        so that mathematically unlucky MAC addresses (i.e., ones that
1986        XOR to the same value) will not all "bunch up" on a single
1987        interface.
1988
1989        Unlike 802.3ad, interfaces may be of differing speeds, and no
1990        special switch configuration is required.  On the down side,
1991        in this mode all incoming traffic arrives over a single
1992        interface, this mode requires certain ethtool support in the
1993        network device driver of the slave interfaces, and the ARP
1994        monitor is not available.
1995
1996balance-alb: This mode is everything that balance-tlb is, and more.
1997        It has all of the features (and restrictions) of balance-tlb,
1998        and will also balance incoming traffic from local network
1999        peers (as described in the Bonding Module Options section,
2000        above).
2001
2002        The only additional down side to this mode is that the network
2003        device driver must support changing the hardware address while
2004        the device is open.
2005
200612.1.2 MT Link Monitoring for Single Switch Topology
2007----------------------------------------------------
2008
2009        The choice of link monitoring may largely depend upon which
2010mode you choose to use.  The more advanced load balancing modes do not
2011support the use of the ARP monitor, and are thus restricted to using
2012the MII monitor (which does not provide as high a level of end to end
2013assurance as the ARP monitor).
2014
201512.2 Maximum Throughput in a Multiple Switch Topology
2016-----------------------------------------------------
2017
2018        Multiple switches may be utilized to optimize for throughput
2019when they are configured in parallel as part of an isolated network
2020between two or more systems, for example:
2021
2022                       +-----------+
2023                       |  Host A   | 
2024                       +-+---+---+-+
2025                         |   |   |
2026                +--------+   |   +---------+
2027                |            |             |
2028         +------+---+  +-----+----+  +-----+----+
2029         | Switch A |  | Switch B |  | Switch C |
2030         +------+---+  +-----+----+  +-----+----+
2031                |            |             |
2032                +--------+   |   +---------+
2033                         |   |   |
2034                       +-+---+---+-+
2035                       |  Host B   | 
2036                       +-----------+
2037
2038        In this configuration, the switches are isolated from one
2039another.  One reason to employ a topology such as this is for an
2040isolated network with many hosts (a cluster configured for high
2041performance, for example), using multiple smaller switches can be more
2042cost effective than a single larger switch, e.g., on a network with 24
2043hosts, three 24 port switches can be significantly less expensive than
2044a single 72 port switch.
2045
2046        If access beyond the network is required, an individual host
2047can be equipped with an additional network device connected to an
2048external network; this host then additionally acts as a gateway.
2049
205012.2.1 MT Bonding Mode Selection for Multiple Switch Topology
2051-------------------------------------------------------------
2052
2053        In actual practice, the bonding mode typically employed in
2054configurations of this type is balance-rr.  Historically, in this
2055network configuration, the usual caveats about out of order packet
2056delivery are mitigated by the use of network adapters that do not do
2057any kind of packet coalescing (via the use of NAPI, or because the
2058device itself does not generate interrupts until some number of
2059packets has arrived).  When employed in this fashion, the balance-rr
2060mode allows individual connections between two hosts to effectively
2061utilize greater than one interface's bandwidth.
2062
206312.2.2 MT Link Monitoring for Multiple Switch Topology
2064------------------------------------------------------
2065
2066        Again, in actual practice, the MII monitor is most often used
2067in this configuration, as performance is given preference over
2068availability.  The ARP monitor will function in this topology, but its
2069advantages over the MII monitor are mitigated by the volume of probes
2070needed as the number of systems involved grows (remember that each
2071host in the network is configured with bonding).
2072
207313. Switch Behavior Issues
2074==========================
2075
207613.1 Link Establishment and Failover Delays
2077-------------------------------------------
2078
2079        Some switches exhibit undesirable behavior with regard to the
2080timing of link up and down reporting by the switch.
2081
2082        First, when a link comes up, some switches may indicate that
2083the link is up (carrier available), but not pass traffic over the
2084interface for some period of time.  This delay is typically due to
2085some type of autonegotiation or routing protocol, but may also occur
2086during switch initialization (e.g., during recovery after a switch
2087failure).  If you find this to be a problem, specify an appropriate
2088value to the updelay bonding module option to delay the use of the
2089relevant interface(s).
2090
2091        Second, some switches may "bounce" the link state one or more
2092times while a link is changing state.  This occurs most commonly while
2093the switch is initializing.  Again, an appropriate updelay value may
2094help.
2095
2096        Note that when a bonding interface has no active links, the
2097driver will immediately reuse the first link that goes up, even if the
2098updelay parameter has been specified (the updelay is ignored in this
2099case).  If there are slave interfaces waiting for the updelay timeout
2100to expire, the interface that first went into that state will be
2101immediately reused.  This reduces down time of the network if the
2102value of updelay has been overestimated, and since this occurs only in
2103cases with no connectivity, there is no additional penalty for
2104ignoring the updelay.
2105
2106        In addition to the concerns about switch timings, if your
2107switches take a long time to go into backup mode, it may be desirable
2108to not activate a backup interface immediately after a link goes down.
2109Failover may be delayed via the downdelay bonding module option.
2110
211113.2 Duplicated Incoming Packets
2112--------------------------------
2113
2114        NOTE: Starting with version 3.0.2, the bonding driver has logic to
2115suppress duplicate packets, which should largely eliminate this problem.
2116The following description is kept for reference.
2117
2118        It is not uncommon to observe a short burst of duplicated
2119traffic when the bonding device is first used, or after it has been
2120idle for some period of time.  This is most easily observed by issuing
2121a "ping" to some other host on the network, and noticing that the
2122output from ping flags duplicates (typically one per slave).
2123
2124        For example, on a bond in active-backup mode with five slaves
2125all connected to one switch, the output may appear as follows:
2126
2127# ping -n 10.0.4.2
2128PING 10.0.4.2 (10.0.4.2) from 10.0.3.10 : 56(84) bytes of data.
212964 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.7 ms
213064 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
213164 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
213264 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
213364 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
213464 bytes from 10.0.4.2: icmp_seq=2 ttl=64 time=0.216 ms
213564 bytes from 10.0.4.2: icmp_seq=3 ttl=64 time=0.267 ms
213664 bytes from 10.0.4.2: icmp_seq=4 ttl=64 time=0.222 ms
2137
2138        This is not due to an error in the bonding driver, rather, it
2139is a side effect of how many switches update their MAC forwarding
2140tables.  Initially, the switch does not associate the MAC address in
2141the packet with a particular switch port, and so it may send the
2142traffic to all ports until its MAC forwarding table is updated.  Since
2143the interfaces attached to the bond may occupy multiple ports on a
2144single switch, when the switch (temporarily) floods the traffic to all
2145ports, the bond device receives multiple copies of the same packet
2146(one per slave device).
2147
2148        The duplicated packet behavior is switch dependent, some
2149switches exhibit this, and some do not.  On switches that display this
2150behavior, it can be induced by clearing the MAC forwarding table (on
2151most Cisco switches, the privileged command "clear mac address-table
2152dynamic" will accomplish this).
2153
215414. Hardware Specific Considerations
2155====================================
2156
2157        This section contains additional information for configuring
2158bonding on specific hardware platforms, or for interfacing bonding
2159with particular switches or other devices.
2160
216114.1 IBM BladeCenter
2162--------------------
2163
2164        This applies to the JS20 and similar systems.
2165
2166        On the JS20 blades, the bonding driver supports only
2167balance-rr, active-backup, balance-tlb and balance-alb modes.  This is
2168largely due to the network topology inside the BladeCenter, detailed
2169below.
2170
2171JS20 network adapter information
2172--------------------------------
2173
2174        All JS20s come with two Broadcom Gigabit Ethernet ports
2175integrated on the planar (that's "motherboard" in IBM-speak).  In the
2176BladeCenter chassis, the eth0 port of all JS20 blades is hard wired to
2177I/O Module #1; similarly, all eth1 ports are wired to I/O Module #2.
2178An add-on Broadcom daughter card can be installed on a JS20 to provide
2179two more Gigabit Ethernet ports.  These ports, eth2 and eth3, are
2180wired to I/O Modules 3 and 4, respectively.
2181
2182        Each I/O Module may contain either a switch or a passthrough
2183module (which allows ports to be directly connected to an external
2184switch).  Some bonding modes require a specific BladeCenter internal
2185network topology in order to function; these are detailed below.
2186
2187        Additional BladeCenter-specific networking information can be
2188found in two IBM Redbooks (www.ibm.com/redbooks):
2189
2190"IBM eServer BladeCenter Networking Options"
2191"IBM eServer BladeCenter Layer 2-7 Network Switching"
2192
2193BladeCenter networking configuration
2194------------------------------------
2195
2196        Because a BladeCenter can be configured in a very large number
2197of ways, this discussion will be confined to describing basic
2198configurations.
2199
2200        Normally, Ethernet Switch Modules (ESMs) are used in I/O
2201modules 1 and 2.  In this configuration, the eth0 and eth1 ports of a
2202JS20 will be connected to different internal switches (in the
2203respective I/O modules).
2204
2205        A passthrough module (OPM or CPM, optical or copper,
2206passthrough module) connects the I/O module directly to an external
2207switch.  By using PMs in I/O module #1 and #2, the eth0 and eth1
2208interfaces of a JS20 can be redirected to the outside world and
2209connected to a common external switch.
2210
2211        Depending upon the mix of ESMs and PMs, the network will
2212appear to bonding as either a single switch topology (all PMs) or as a
2213multiple switch topology (one or more ESMs, zero or more PMs).  It is
2214also possible to connect ESMs together, resulting in a configuration
2215much like the example in "High Availability in a Multiple Switch
2216Topology," above.
2217
2218Requirements for specific modes
2219-------------------------------
2220
2221        The balance-rr mode requires the use of passthrough modules
2222for devices in the bond, all connected to an common external switch.
2223That switch must be configured for "etherchannel" or "trunking" on the
2224appropriate ports, as is usual for balance-rr.
2225
2226        The balance-alb and balance-tlb modes will function with
2227either switch modules or passthrough modules (or a mix).  The only
2228specific requirement for these modes is that all network interfaces
2229must be able to reach all destinations for traffic sent over the
2230bonding device (i.e., the network must converge at some point outside
2231the BladeCenter).
2232
2233        The active-backup mode has no additional requirements.
2234
2235Link monitoring issues
2236----------------------
2237
2238        When an Ethernet Switch Module is in place, only the ARP
2239monitor will reliably detect link loss to an external switch.  This is
2240nothing unusual, but examination of the BladeCenter cabinet would
2241suggest that the "external" network ports are the ethernet ports for
2242the system, when it fact there is a switch between these "external"
2243ports and the devices on the JS20 system itself.  The MII monitor is
2244only able to detect link failures between the ESM and the JS20 system.
2245
2246        When a passthrough module is in place, the MII monitor does
2247detect failures to the "external" port, which is then directly
2248connected to the JS20 system.
2249
2250Other concerns
2251--------------
2252
2253        The Serial Over LAN (SoL) link is established over the primary
2254ethernet (eth0) only, therefore, any loss of link to eth0 will result
2255in losing your SoL connection.  It will not fail over with other
2256network traffic, as the SoL system is beyond the control of the
2257bonding driver.
2258
2259        It may be desirable to disable spanning tree on the switch
2260(either the internal Ethernet Switch Module, or an external switch) to
2261avoid fail-over delay issues when using bonding.
2262
2263        
226415. Frequently Asked Questions
2265==============================
2266
22671.  Is it SMP safe?
2268
2269        Yes. The old 2.0.xx channel bonding patch was not SMP safe.
2270The new driver was designed to be SMP safe from the start.
2271
22722.  What type of cards will work with it?
2273
2274        Any Ethernet type cards (you can even mix cards - a Intel
2275EtherExpress PRO/100 and a 3com 3c905b, for example).  For most modes,
2276devices need not be of the same speed.
2277
2278        Starting with version 3.2.1, bonding also supports Infiniband
2279slaves in active-backup mode.
2280
22813.  How many bonding devices can I have?
2282
2283        There is no limit.
2284
22854.  How many slaves can a bonding device have?
2286
2287        This is limited only by the number of network interfaces Linux
2288supports and/or the number of network cards you can place in your
2289system.
2290
22915.  What happens when a slave link dies?
2292
2293        If link monitoring is enabled, then the failing device will be
2294disabled.  The active-backup mode will fail over to a backup link, and
2295other modes will ignore the failed link.  The link will continue to be
2296monitored, and should it recover, it will rejoin the bond (in whatever
2297manner is appropriate for the mode). See the sections on High
2298Availability and the documentation for each mode for additional
2299information.
2300        
2301        Link monitoring can be enabled via either the miimon or
2302arp_interval parameters (described in the module parameters section,
2303above).  In general, miimon monitors the carrier state as sensed by
2304the underlying network device, and the arp monitor (arp_interval)
2305monitors connectivity to another host on the local network.
2306
2307        If no link monitoring is configured, the bonding driver will
2308be unable to detect link failures, and will assume that all links are
2309always available.  This will likely result in lost packets, and a
2310resulting degradation of performance.  The precise performance loss
2311depends upon the bonding mode and network configuration.
2312
23136.  Can bonding be used for High Availability?
2314
2315        Yes.  See the section on High Availability for details.
2316
23177.  Which switches/systems does it work with?
2318
2319        The full answer to this depends upon the desired mode.
2320
2321        In the basic balance modes (balance-rr and balance-xor), it
2322works with any system that supports etherchannel (also called
2323trunking).  Most managed switches currently available have such
2324support, and many unmanaged switches as well.
2325
2326        The advanced balance modes (balance-tlb and balance-alb) do
2327not have special switch requirements, but do need device drivers that
2328support specific features (described in the appropriate section under
2329module parameters, above).
2330
2331        In 802.3ad mode, it works with systems that support IEEE
2332802.3ad Dynamic Link Aggregation.  Most managed and many unmanaged
2333switches currently available support 802.3ad.
2334
2335        The active-backup mode should work with any Layer-II switch.
2336
23378.  Where does a bonding device get its MAC address from?
2338
2339        When using slave devices that have fixed MAC addresses, or when
2340the fail_over_mac option is enabled, the bonding device's MAC address is
2341the MAC address of the active slave.
2342
2343        For other configurations, if not explicitly configured (with
2344ifconfig or ip link), the MAC address of the bonding device is taken from
2345its first slave device.  This MAC address is then passed to all following
2346slaves and remains persistent (even if the first slave is removed) until
2347the bonding device is brought down or reconfigured.
2348
2349        If you wish to change the MAC address, you can set it with
2350ifconfig or ip link:
2351
2352# ifconfig bond0 hw ether 00:11:22:33:44:55
2353
2354# ip link set bond0 address 66:77:88:99:aa:bb
2355
2356        The MAC address can be also changed by bringing down/up the
2357device and then changing its slaves (or their order):
2358
2359# ifconfig bond0 down ; modprobe -r bonding
2360# ifconfig bond0 .... up
2361# ifenslave bond0 eth...
2362
2363        This method will automatically take the address from the next
2364slave that is added.
2365
2366        To restore your slaves' MAC addresses, you need to detach them
2367from the bond (`ifenslave -d bond0 eth0'). The bonding driver will
2368then restore the MAC addresses that the slaves had before they were
2369enslaved.
2370
237116. Resources and Links
2372=======================
2373
2374The latest version of the bonding driver can be found in the latest
2375version of the linux kernel, found on http://kernel.org
2376
2377The latest version of this document can be found in either the latest
2378kernel source (named Documentation/networking/bonding.txt), or on the
2379bonding sourceforge site:
2380
2381http://www.sourceforge.net/projects/bonding
2382
2383Discussions regarding the bonding driver take place primarily on the
2384bonding-devel mailing list, hosted at sourceforge.net.  If you have
2385questions or problems, post them to the list.  The list address is:
2386
2387bonding-devel@lists.sourceforge.net
2388
2389        The administrative interface (to subscribe or unsubscribe) can
2390be found at:
2391
2392https://lists.sourceforge.net/lists/listinfo/bonding-devel
2393
2394Donald Becker's Ethernet Drivers and diag programs may be found at :
2395 - http://www.scyld.com/network/
2396
2397You will also find a lot of information regarding Ethernet, NWay, MII,
2398etc. at www.scyld.com.
2399
2400-- END --
2401
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.