linux/Documentation/networking/bonding.txt
<<
>>
Prefs
   1
   2                Linux Ethernet Bonding Driver HOWTO
   3
   4                Latest update: 27 April 2011
   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
  523.5     Configuration with Interfaces Support
  533.6     Overriding Configuration for Special Cases
  543.7 Configuring LACP for 802.3ad mode in a more secure way
  55
  564. Querying Bonding Configuration
  574.1     Bonding Configuration
  584.2     Network Configuration
  59
  605. Switch Configuration
  61
  626. 802.1q VLAN Support
  63
  647. Link Monitoring
  657.1     ARP Monitor Operation
  667.2     Configuring Multiple ARP Targets
  677.3     MII Monitor Operation
  68
  698. Potential Trouble Sources
  708.1     Adventures in Routing
  718.2     Ethernet Device Renaming
  728.3     Painfully Slow Or No Failed Link Detection By Miimon
  73
  749. SNMP agents
  75
  7610. Promiscuous mode
  77
  7811. Configuring Bonding for High Availability
  7911.1    High Availability in a Single Switch Topology
  8011.2    High Availability in a Multiple Switch Topology
  8111.2.1          HA Bonding Mode Selection for Multiple Switch Topology
  8211.2.2          HA Link Monitoring for Multiple Switch Topology
  83
  8412. Configuring Bonding for Maximum Throughput
  8512.1    Maximum Throughput in a Single Switch Topology
  8612.1.1          MT Bonding Mode Selection for Single Switch Topology
  8712.1.2          MT Link Monitoring for Single Switch Topology
  8812.2    Maximum Throughput in a Multiple Switch Topology
  8912.2.1          MT Bonding Mode Selection for Multiple Switch Topology
  9012.2.2          MT Link Monitoring for Multiple Switch Topology
  91
  9213. Switch Behavior Issues
  9313.1    Link Establishment and Failover Delays
  9413.2    Duplicated Incoming Packets
  95
  9614. Hardware Specific Considerations
  9714.1    IBM BladeCenter
  98
  9915. Frequently Asked Questions
 100
 10116. Resources and Links
 102
 103
 1041. Bonding Driver Installation
 105==============================
 106
 107        Most popular distro kernels ship with the bonding driver
 108already available as a module. If your distro does not, or you
 109have need to compile bonding from source (e.g., configuring and
 110installing a mainline kernel from kernel.org), you'll need to perform
 111the following steps:
 112
 1131.1 Configure and build the kernel with bonding
 114-----------------------------------------------
 115
 116        The current version of the bonding driver is available in the
 117drivers/net/bonding subdirectory of the most recent kernel source
 118(which is available on http://kernel.org).  Most users "rolling their
 119own" will want to use the most recent kernel from kernel.org.
 120
 121        Configure kernel with "make menuconfig" (or "make xconfig" or
 122"make config"), then select "Bonding driver support" in the "Network
 123device support" section.  It is recommended that you configure the
 124driver as module since it is currently the only way to pass parameters
 125to the driver or configure more than one bonding device.
 126
 127        Build and install the new kernel and modules.
 128
 1291.2 Bonding Control Utility
 130-------------------------------------
 131
 132         It is recommended to configure bonding via iproute2 (netlink)
 133or sysfs, the old ifenslave control utility is obsolete.
 134
 1352. Bonding Driver Options
 136=========================
 137
 138        Options for the bonding driver are supplied as parameters to the
 139bonding module at load time, or are specified via sysfs.
 140
 141        Module options may be given as command line arguments to the
 142insmod or modprobe command, but are usually specified in either the
 143/etc/modrobe.d/*.conf configuration files, or in a distro-specific
 144configuration file (some of which are detailed in the next section).
 145
 146        Details on bonding support for sysfs is provided in the
 147"Configuring Bonding Manually via Sysfs" section, below.
 148
 149        The available bonding driver parameters are listed below. If a
 150parameter is not specified the default value is used.  When initially
 151configuring a bond, it is recommended "tail -f /var/log/messages" be
 152run in a separate window to watch for bonding driver error messages.
 153
 154        It is critical that either the miimon or arp_interval and
 155arp_ip_target parameters be specified, otherwise serious network
 156degradation will occur during link failures.  Very few devices do not
 157support at least miimon, so there is really no reason not to use it.
 158
 159        Options with textual values will accept either the text name
 160or, for backwards compatibility, the option value.  E.g.,
 161"mode=802.3ad" and "mode=4" set the same mode.
 162
 163        The parameters are as follows:
 164
 165active_slave
 166
 167        Specifies the new active slave for modes that support it
 168        (active-backup, balance-alb and balance-tlb).  Possible values
 169        are the name of any currently enslaved interface, or an empty
 170        string.  If a name is given, the slave and its link must be up in order
 171        to be selected as the new active slave.  If an empty string is
 172        specified, the current active slave is cleared, and a new active
 173        slave is selected automatically.
 174
 175        Note that this is only available through the sysfs interface. No module
 176        parameter by this name exists.
 177
 178        The normal value of this option is the name of the currently
 179        active slave, or the empty string if there is no active slave or
 180        the current mode does not use an active slave.
 181
 182ad_actor_sys_prio
 183
 184        In an AD system, this specifies the system priority. The allowed range
 185        is 1 - 65535. If the value is not specified, it takes 65535 as the
 186        default value.
 187
 188        This parameter has effect only in 802.3ad mode and is available through
 189        SysFs interface.
 190
 191ad_actor_system
 192
 193        In an AD system, this specifies the mac-address for the actor in
 194        protocol packet exchanges (LACPDUs). The value cannot be NULL or
 195        multicast. It is preferred to have the local-admin bit set for this
 196        mac but driver does not enforce it. If the value is not given then
 197        system defaults to using the masters' mac address as actors' system
 198        address.
 199
 200        This parameter has effect only in 802.3ad mode and is available through
 201        SysFs interface.
 202
 203ad_select
 204
 205        Specifies the 802.3ad aggregation selection logic to use.  The
 206        possible values and their effects are:
 207
 208        stable or 0
 209
 210                The active aggregator is chosen by largest aggregate
 211                bandwidth.
 212
 213                Reselection of the active aggregator occurs only when all
 214                slaves of the active aggregator are down or the active
 215                aggregator has no slaves.
 216
 217                This is the default value.
 218
 219        bandwidth or 1
 220
 221                The active aggregator is chosen by largest aggregate
 222                bandwidth.  Reselection occurs if:
 223
 224                - A slave is added to or removed from the bond
 225
 226                - Any slave's link state changes
 227
 228                - Any slave's 802.3ad association state changes
 229
 230                - The bond's administrative state changes to up
 231
 232        count or 2
 233
 234                The active aggregator is chosen by the largest number of
 235                ports (slaves).  Reselection occurs as described under the
 236                "bandwidth" setting, above.
 237
 238        The bandwidth and count selection policies permit failover of
 239        802.3ad aggregations when partial failure of the active aggregator
 240        occurs.  This keeps the aggregator with the highest availability
 241        (either in bandwidth or in number of ports) active at all times.
 242
 243        This option was added in bonding version 3.4.0.
 244
 245ad_user_port_key
 246
 247        In an AD system, the port-key has three parts as shown below -
 248
 249           Bits   Use
 250           00     Duplex
 251           01-05  Speed
 252           06-15  User-defined
 253
 254        This defines the upper 10 bits of the port key. The values can be
 255        from 0 - 1023. If not given, the system defaults to 0.
 256
 257        This parameter has effect only in 802.3ad mode and is available through
 258        SysFs interface.
 259
 260all_slaves_active
 261
 262        Specifies that duplicate frames (received on inactive ports) should be
 263        dropped (0) or delivered (1).
 264
 265        Normally, bonding will drop duplicate frames (received on inactive
 266        ports), which is desirable for most users. But there are some times
 267        it is nice to allow duplicate frames to be delivered.
 268
 269        The default value is 0 (drop duplicate frames received on inactive
 270        ports).
 271
 272arp_interval
 273
 274        Specifies the ARP link monitoring frequency in milliseconds.
 275
 276        The ARP monitor works by periodically checking the slave
 277        devices to determine whether they have sent or received
 278        traffic recently (the precise criteria depends upon the
 279        bonding mode, and the state of the slave).  Regular traffic is
 280        generated via ARP probes issued for the addresses specified by
 281        the arp_ip_target option.
 282
 283        This behavior can be modified by the arp_validate option,
 284        below.
 285
 286        If ARP monitoring is used in an etherchannel compatible mode
 287        (modes 0 and 2), the switch should be configured in a mode
 288        that evenly distributes packets across all links. If the
 289        switch is configured to distribute the packets in an XOR
 290        fashion, all replies from the ARP targets will be received on
 291        the same link which could cause the other team members to
 292        fail.  ARP monitoring should not be used in conjunction with
 293        miimon.  A value of 0 disables ARP monitoring.  The default
 294        value is 0.
 295
 296arp_ip_target
 297
 298        Specifies the IP addresses to use as ARP monitoring peers when
 299        arp_interval is > 0.  These are the targets of the ARP request
 300        sent to determine the health of the link to the targets.
 301        Specify these values in ddd.ddd.ddd.ddd format.  Multiple IP
 302        addresses must be separated by a comma.  At least one IP
 303        address must be given for ARP monitoring to function.  The
 304        maximum number of targets that can be specified is 16.  The
 305        default value is no IP addresses.
 306
 307arp_validate
 308
 309        Specifies whether or not ARP probes and replies should be
 310        validated in any mode that supports arp monitoring, or whether
 311        non-ARP traffic should be filtered (disregarded) for link
 312        monitoring purposes.
 313
 314        Possible values are:
 315
 316        none or 0
 317
 318                No validation or filtering is performed.
 319
 320        active or 1
 321
 322                Validation is performed only for the active slave.
 323
 324        backup or 2
 325
 326                Validation is performed only for backup slaves.
 327
 328        all or 3
 329
 330                Validation is performed for all slaves.
 331
 332        filter or 4
 333
 334                Filtering is applied to all slaves. No validation is
 335                performed.
 336
 337        filter_active or 5
 338
 339                Filtering is applied to all slaves, validation is performed
 340                only for the active slave.
 341
 342        filter_backup or 6
 343
 344                Filtering is applied to all slaves, validation is performed
 345                only for backup slaves.
 346
 347        Validation:
 348
 349        Enabling validation causes the ARP monitor to examine the incoming
 350        ARP requests and replies, and only consider a slave to be up if it
 351        is receiving the appropriate ARP traffic.
 352
 353        For an active slave, the validation checks ARP replies to confirm
 354        that they were generated by an arp_ip_target.  Since backup slaves
 355        do not typically receive these replies, the validation performed
 356        for backup slaves is on the broadcast ARP request sent out via the
 357        active slave.  It is possible that some switch or network
 358        configurations may result in situations wherein the backup slaves
 359        do not receive the ARP requests; in such a situation, validation
 360        of backup slaves must be disabled.
 361
 362        The validation of ARP requests on backup slaves is mainly helping
 363        bonding to decide which slaves are more likely to work in case of
 364        the active slave failure, it doesn't really guarantee that the
 365        backup slave will work if it's selected as the next active slave.
 366
 367        Validation is useful in network configurations in which multiple
 368        bonding hosts are concurrently issuing ARPs to one or more targets
 369        beyond a common switch.  Should the link between the switch and
 370        target fail (but not the switch itself), the probe traffic
 371        generated by the multiple bonding instances will fool the standard
 372        ARP monitor into considering the links as still up.  Use of
 373        validation can resolve this, as the ARP monitor will only consider
 374        ARP requests and replies associated with its own instance of
 375        bonding.
 376
 377        Filtering:
 378
 379        Enabling filtering causes the ARP monitor to only use incoming ARP
 380        packets for link availability purposes.  Arriving packets that are
 381        not ARPs are delivered normally, but do not count when determining
 382        if a slave is available.
 383
 384        Filtering operates by only considering the reception of ARP
 385        packets (any ARP packet, regardless of source or destination) when
 386        determining if a slave has received traffic for link availability
 387        purposes.
 388
 389        Filtering is useful in network configurations in which significant
 390        levels of third party broadcast traffic would fool the standard
 391        ARP monitor into considering the links as still up.  Use of
 392        filtering can resolve this, as only ARP traffic is considered for
 393        link availability purposes.
 394
 395        This option was added in bonding version 3.1.0.
 396
 397arp_all_targets
 398
 399        Specifies the quantity of arp_ip_targets that must be reachable
 400        in order for the ARP monitor to consider a slave as being up.
 401        This option affects only active-backup mode for slaves with
 402        arp_validation enabled.
 403
 404        Possible values are:
 405
 406        any or 0
 407
 408                consider the slave up only when any of the arp_ip_targets
 409                is reachable
 410
 411        all or 1
 412
 413                consider the slave up only when all of the arp_ip_targets
 414                are reachable
 415
 416downdelay
 417
 418        Specifies the time, in milliseconds, to wait before disabling
 419        a slave after a link failure has been detected.  This option
 420        is only valid for the miimon link monitor.  The downdelay
 421        value should be a multiple of the miimon value; if not, it
 422        will be rounded down to the nearest multiple.  The default
 423        value is 0.
 424
 425fail_over_mac
 426
 427        Specifies whether active-backup mode should set all slaves to
 428        the same MAC address at enslavement (the traditional
 429        behavior), or, when enabled, perform special handling of the
 430        bond's MAC address in accordance with the selected policy.
 431
 432        Possible values are:
 433
 434        none or 0
 435
 436                This setting disables fail_over_mac, and causes
 437                bonding to set all slaves of an active-backup bond to
 438                the same MAC address at enslavement time.  This is the
 439                default.
 440
 441        active or 1
 442
 443                The "active" fail_over_mac policy indicates that the
 444                MAC address of the bond should always be the MAC
 445                address of the currently active slave.  The MAC
 446                address of the slaves is not changed; instead, the MAC
 447                address of the bond changes during a failover.
 448
 449                This policy is useful for devices that cannot ever
 450                alter their MAC address, or for devices that refuse
 451                incoming broadcasts with their own source MAC (which
 452                interferes with the ARP monitor).
 453
 454                The down side of this policy is that every device on
 455                the network must be updated via gratuitous ARP,
 456                vs. just updating a switch or set of switches (which
 457                often takes place for any traffic, not just ARP
 458                traffic, if the switch snoops incoming traffic to
 459                update its tables) for the traditional method.  If the
 460                gratuitous ARP is lost, communication may be
 461                disrupted.
 462
 463                When this policy is used in conjunction with the mii
 464                monitor, devices which assert link up prior to being
 465                able to actually transmit and receive are particularly
 466                susceptible to loss of the gratuitous ARP, and an
 467                appropriate updelay setting may be required.
 468
 469        follow or 2
 470
 471                The "follow" fail_over_mac policy causes the MAC
 472                address of the bond to be selected normally (normally
 473                the MAC address of the first slave added to the bond).
 474                However, the second and subsequent slaves are not set
 475                to this MAC address while they are in a backup role; a
 476                slave is programmed with the bond's MAC address at
 477                failover time (and the formerly active slave receives
 478                the newly active slave's MAC address).
 479
 480                This policy is useful for multiport devices that
 481                either become confused or incur a performance penalty
 482                when multiple ports are programmed with the same MAC
 483                address.
 484
 485
 486        The default policy is none, unless the first slave cannot
 487        change its MAC address, in which case the active policy is
 488        selected by default.
 489
 490        This option may be modified via sysfs only when no slaves are
 491        present in the bond.
 492
 493        This option was added in bonding version 3.2.0.  The "follow"
 494        policy was added in bonding version 3.3.0.
 495
 496lacp_rate
 497
 498        Option specifying the rate in which we'll ask our link partner
 499        to transmit LACPDU packets in 802.3ad mode.  Possible values
 500        are:
 501
 502        slow or 0
 503                Request partner to transmit LACPDUs every 30 seconds
 504
 505        fast or 1
 506                Request partner to transmit LACPDUs every 1 second
 507
 508        The default is slow.
 509
 510max_bonds
 511
 512        Specifies the number of bonding devices to create for this
 513        instance of the bonding driver.  E.g., if max_bonds is 3, and
 514        the bonding driver is not already loaded, then bond0, bond1
 515        and bond2 will be created.  The default value is 1.  Specifying
 516        a value of 0 will load bonding, but will not create any devices.
 517
 518miimon
 519
 520        Specifies the MII link monitoring frequency in milliseconds.
 521        This determines how often the link state of each slave is
 522        inspected for link failures.  A value of zero disables MII
 523        link monitoring.  A value of 100 is a good starting point.
 524        The use_carrier option, below, affects how the link state is
 525        determined.  See the High Availability section for additional
 526        information.  The default value is 0.
 527
 528min_links
 529
 530        Specifies the minimum number of links that must be active before
 531        asserting carrier. It is similar to the Cisco EtherChannel min-links
 532        feature. This allows setting the minimum number of member ports that
 533        must be up (link-up state) before marking the bond device as up
 534        (carrier on). This is useful for situations where higher level services
 535        such as clustering want to ensure a minimum number of low bandwidth
 536        links are active before switchover. This option only affect 802.3ad
 537        mode.
 538
 539        The default value is 0. This will cause carrier to be asserted (for
 540        802.3ad mode) whenever there is an active aggregator, regardless of the
 541        number of available links in that aggregator. Note that, because an
 542        aggregator cannot be active without at least one available link,
 543        setting this option to 0 or to 1 has the exact same effect.
 544
 545mode
 546
 547        Specifies one of the bonding policies. The default is
 548        balance-rr (round robin).  Possible values are:
 549
 550        balance-rr or 0
 551
 552                Round-robin policy: Transmit packets in sequential
 553                order from the first available slave through the
 554                last.  This mode provides load balancing and fault
 555                tolerance.
 556
 557        active-backup or 1
 558
 559                Active-backup policy: Only one slave in the bond is
 560                active.  A different slave becomes active if, and only
 561                if, the active slave fails.  The bond's MAC address is
 562                externally visible on only one port (network adapter)
 563                to avoid confusing the switch.
 564
 565                In bonding version 2.6.2 or later, when a failover
 566                occurs in active-backup mode, bonding will issue one
 567                or more gratuitous ARPs on the newly active slave.
 568                One gratuitous ARP is issued for the bonding master
 569                interface and each VLAN interfaces configured above
 570                it, provided that the interface has at least one IP
 571                address configured.  Gratuitous ARPs issued for VLAN
 572                interfaces are tagged with the appropriate VLAN id.
 573
 574                This mode provides fault tolerance.  The primary
 575                option, documented below, affects the behavior of this
 576                mode.
 577
 578        balance-xor or 2
 579
 580                XOR policy: Transmit based on the selected transmit
 581                hash policy.  The default policy is a simple [(source
 582                MAC address XOR'd with destination MAC address XOR
 583                packet type ID) modulo slave count].  Alternate transmit
 584                policies may be selected via the xmit_hash_policy option,
 585                described below.
 586
 587                This mode provides load balancing and fault tolerance.
 588
 589        broadcast or 3
 590
 591                Broadcast policy: transmits everything on all slave
 592                interfaces.  This mode provides fault tolerance.
 593
 594        802.3ad or 4
 595
 596                IEEE 802.3ad Dynamic link aggregation.  Creates
 597                aggregation groups that share the same speed and
 598                duplex settings.  Utilizes all slaves in the active
 599                aggregator according to the 802.3ad specification.
 600
 601                Slave selection for outgoing traffic is done according
 602                to the transmit hash policy, which may be changed from
 603                the default simple XOR policy via the xmit_hash_policy
 604                option, documented below.  Note that not all transmit
 605                policies may be 802.3ad compliant, particularly in
 606                regards to the packet mis-ordering requirements of
 607                section 43.2.4 of the 802.3ad standard.  Differing
 608                peer implementations will have varying tolerances for
 609                noncompliance.
 610
 611                Prerequisites:
 612
 613                1. Ethtool support in the base drivers for retrieving
 614                the speed and duplex of each slave.
 615
 616                2. A switch that supports IEEE 802.3ad Dynamic link
 617                aggregation.
 618
 619                Most switches will require some type of configuration
 620                to enable 802.3ad mode.
 621
 622        balance-tlb or 5
 623
 624                Adaptive transmit load balancing: channel bonding that
 625                does not require any special switch support.
 626
 627                In tlb_dynamic_lb=1 mode; the outgoing traffic is
 628                distributed according to the current load (computed
 629                relative to the speed) on each slave.
 630
 631                In tlb_dynamic_lb=0 mode; the load balancing based on
 632                current load is disabled and the load is distributed
 633                only using the hash distribution.
 634
 635                Incoming traffic is received by the current slave.
 636                If the receiving slave fails, another slave takes over
 637                the MAC address of the failed receiving slave.
 638
 639                Prerequisite:
 640
 641                Ethtool support in the base drivers for retrieving the
 642                speed of each slave.
 643
 644        balance-alb or 6
 645
 646                Adaptive load balancing: includes balance-tlb plus
 647                receive load balancing (rlb) for IPV4 traffic, and
 648                does not require any special switch support.  The
 649                receive load balancing is achieved by ARP negotiation.
 650                The bonding driver intercepts the ARP Replies sent by
 651                the local system on their way out and overwrites the
 652                source hardware address with the unique hardware
 653                address of one of the slaves in the bond such that
 654                different peers use different hardware addresses for
 655                the server.
 656
 657                Receive traffic from connections created by the server
 658                is also balanced.  When the local system sends an ARP
 659                Request the bonding driver copies and saves the peer's
 660                IP information from the ARP packet.  When the ARP
 661                Reply arrives from the peer, its hardware address is
 662                retrieved and the bonding driver initiates an ARP
 663                reply to this peer assigning it to one of the slaves
 664                in the bond.  A problematic outcome of using ARP
 665                negotiation for balancing is that each time that an
 666                ARP request is broadcast it uses the hardware address
 667                of the bond.  Hence, peers learn the hardware address
 668                of the bond and the balancing of receive traffic
 669                collapses to the current slave.  This is handled by
 670                sending updates (ARP Replies) to all the peers with
 671                their individually assigned hardware address such that
 672                the traffic is redistributed.  Receive traffic is also
 673                redistributed when a new slave is added to the bond
 674                and when an inactive slave is re-activated.  The
 675                receive load is distributed sequentially (round robin)
 676                among the group of highest speed slaves in the bond.
 677
 678                When a link is reconnected or a new slave joins the
 679                bond the receive traffic is redistributed among all
 680                active slaves in the bond by initiating ARP Replies
 681                with the selected MAC address to each of the
 682                clients. The updelay parameter (detailed below) must
 683                be set to a value equal or greater than the switch's
 684                forwarding delay so that the ARP Replies sent to the
 685                peers will not be blocked by the switch.
 686
 687                Prerequisites:
 688
 689                1. Ethtool support in the base drivers for retrieving
 690                the speed of each slave.
 691
 692                2. Base driver support for setting the hardware
 693                address of a device while it is open.  This is
 694                required so that there will always be one slave in the
 695                team using the bond hardware address (the
 696                curr_active_slave) while having a unique hardware
 697                address for each slave in the bond.  If the
 698                curr_active_slave fails its hardware address is
 699                swapped with the new curr_active_slave that was
 700                chosen.
 701
 702num_grat_arp
 703num_unsol_na
 704
 705        Specify the number of peer notifications (gratuitous ARPs and
 706        unsolicited IPv6 Neighbor Advertisements) to be issued after a
 707        failover event.  As soon as the link is up on the new slave
 708        (possibly immediately) a peer notification is sent on the
 709        bonding device and each VLAN sub-device.  This is repeated at
 710        each link monitor interval (arp_interval or miimon, whichever
 711        is active) if the number is greater than 1.
 712
 713        The valid range is 0 - 255; the default value is 1.  These options
 714        affect only the active-backup mode.  These options were added for
 715        bonding versions 3.3.0 and 3.4.0 respectively.
 716
 717        From Linux 3.0 and bonding version 3.7.1, these notifications
 718        are generated by the ipv4 and ipv6 code and the numbers of
 719        repetitions cannot be set independently.
 720
 721packets_per_slave
 722
 723        Specify the number of packets to transmit through a slave before
 724        moving to the next one. When set to 0 then a slave is chosen at
 725        random.
 726
 727        The valid range is 0 - 65535; the default value is 1. This option
 728        has effect only in balance-rr mode.
 729
 730primary
 731
 732        A string (eth0, eth2, etc) specifying which slave is the
 733        primary device.  The specified device will always be the
 734        active slave while it is available.  Only when the primary is
 735        off-line will alternate devices be used.  This is useful when
 736        one slave is preferred over another, e.g., when one slave has
 737        higher throughput than another.
 738
 739        The primary option is only valid for active-backup(1),
 740        balance-tlb (5) and balance-alb (6) mode.
 741
 742primary_reselect
 743
 744        Specifies the reselection policy for the primary slave.  This
 745        affects how the primary slave is chosen to become the active slave
 746        when failure of the active slave or recovery of the primary slave
 747        occurs.  This option is designed to prevent flip-flopping between
 748        the primary slave and other slaves.  Possible values are:
 749
 750        always or 0 (default)
 751
 752                The primary slave becomes the active slave whenever it
 753                comes back up.
 754
 755        better or 1
 756
 757                The primary slave becomes the active slave when it comes
 758                back up, if the speed and duplex of the primary slave is
 759                better than the speed and duplex of the current active
 760                slave.
 761
 762        failure or 2
 763
 764                The primary slave becomes the active slave only if the
 765                current active slave fails and the primary slave is up.
 766
 767        The primary_reselect setting is ignored in two cases:
 768
 769                If no slaves are active, the first slave to recover is
 770                made the active slave.
 771
 772                When initially enslaved, the primary slave is always made
 773                the active slave.
 774
 775        Changing the primary_reselect policy via sysfs will cause an
 776        immediate selection of the best active slave according to the new
 777        policy.  This may or may not result in a change of the active
 778        slave, depending upon the circumstances.
 779
 780        This option was added for bonding version 3.6.0.
 781
 782tlb_dynamic_lb
 783
 784        Specifies if dynamic shuffling of flows is enabled in tlb
 785        mode. The value has no effect on any other modes.
 786
 787        The default behavior of tlb mode is to shuffle active flows across
 788        slaves based on the load in that interval. This gives nice lb
 789        characteristics but can cause packet reordering. If re-ordering is
 790        a concern use this variable to disable flow shuffling and rely on
 791        load balancing provided solely by the hash distribution.
 792        xmit-hash-policy can be used to select the appropriate hashing for
 793        the setup.
 794
 795        The sysfs entry can be used to change the setting per bond device
 796        and the initial value is derived from the module parameter. The
 797        sysfs entry is allowed to be changed only if the bond device is
 798        down.
 799
 800        The default value is "1" that enables flow shuffling while value "0"
 801        disables it. This option was added in bonding driver 3.7.1
 802
 803
 804updelay
 805
 806        Specifies the time, in milliseconds, to wait before enabling a
 807        slave after a link recovery has been detected.  This option is
 808        only valid for the miimon link monitor.  The updelay value
 809        should be a multiple of the miimon value; if not, it will be
 810        rounded down to the nearest multiple.  The default value is 0.
 811
 812use_carrier
 813
 814        Specifies whether or not miimon should use MII or ETHTOOL
 815        ioctls vs. netif_carrier_ok() to determine the link
 816        status. The MII or ETHTOOL ioctls are less efficient and
 817        utilize a deprecated calling sequence within the kernel.  The
 818        netif_carrier_ok() relies on the device driver to maintain its
 819        state with netif_carrier_on/off; at this writing, most, but
 820        not all, device drivers support this facility.
 821
 822        If bonding insists that the link is up when it should not be,
 823        it may be that your network device driver does not support
 824        netif_carrier_on/off.  The default state for netif_carrier is
 825        "carrier on," so if a driver does not support netif_carrier,
 826        it will appear as if the link is always up.  In this case,
 827        setting use_carrier to 0 will cause bonding to revert to the
 828        MII / ETHTOOL ioctl method to determine the link state.
 829
 830        A value of 1 enables the use of netif_carrier_ok(), a value of
 831        0 will use the deprecated MII / ETHTOOL ioctls.  The default
 832        value is 1.
 833
 834xmit_hash_policy
 835
 836        Selects the transmit hash policy to use for slave selection in
 837        balance-xor, 802.3ad, and tlb modes.  Possible values are:
 838
 839        layer2
 840
 841                Uses XOR of hardware MAC addresses and packet type ID
 842                field to generate the hash. The formula is
 843
 844                hash = source MAC XOR destination MAC XOR packet type ID
 845                slave number = hash modulo slave count
 846
 847                This algorithm will place all traffic to a particular
 848                network peer on the same slave.
 849
 850                This algorithm is 802.3ad compliant.
 851
 852        layer2+3
 853
 854                This policy uses a combination of layer2 and layer3
 855                protocol information to generate the hash.
 856
 857                Uses XOR of hardware MAC addresses and IP addresses to
 858                generate the hash.  The formula is
 859
 860                hash = source MAC XOR destination MAC XOR packet type ID
 861                hash = hash XOR source IP XOR destination IP
 862                hash = hash XOR (hash RSHIFT 16)
 863                hash = hash XOR (hash RSHIFT 8)
 864                And then hash is reduced modulo slave count.
 865
 866                If the protocol is IPv6 then the source and destination
 867                addresses are first hashed using ipv6_addr_hash.
 868
 869                This algorithm will place all traffic to a particular
 870                network peer on the same slave.  For non-IP traffic,
 871                the formula is the same as for the layer2 transmit
 872                hash policy.
 873
 874                This policy is intended to provide a more balanced
 875                distribution of traffic than layer2 alone, especially
 876                in environments where a layer3 gateway device is
 877                required to reach most destinations.
 878
 879                This algorithm is 802.3ad compliant.
 880
 881        layer3+4
 882
 883                This policy uses upper layer protocol information,
 884                when available, to generate the hash.  This allows for
 885                traffic to a particular network peer to span multiple
 886                slaves, although a single connection will not span
 887                multiple slaves.
 888
 889                The formula for unfragmented TCP and UDP packets is
 890
 891                hash = source port, destination port (as in the header)
 892                hash = hash XOR source IP XOR destination IP
 893                hash = hash XOR (hash RSHIFT 16)
 894                hash = hash XOR (hash RSHIFT 8)
 895                And then hash is reduced modulo slave count.
 896
 897                If the protocol is IPv6 then the source and destination
 898                addresses are first hashed using ipv6_addr_hash.
 899
 900                For fragmented TCP or UDP packets and all other IPv4 and
 901                IPv6 protocol traffic, the source and destination port
 902                information is omitted.  For non-IP traffic, the
 903                formula is the same as for the layer2 transmit hash
 904                policy.
 905
 906                This algorithm is not fully 802.3ad compliant.  A
 907                single TCP or UDP conversation containing both
 908                fragmented and unfragmented packets will see packets
 909                striped across two interfaces.  This may result in out
 910                of order delivery.  Most traffic types will not meet
 911                this criteria, as TCP rarely fragments traffic, and
 912                most UDP traffic is not involved in extended
 913                conversations.  Other implementations of 802.3ad may
 914                or may not tolerate this noncompliance.
 915
 916        encap2+3
 917
 918                This policy uses the same formula as layer2+3 but it
 919                relies on skb_flow_dissect to obtain the header fields
 920                which might result in the use of inner headers if an
 921                encapsulation protocol is used. For example this will
 922                improve the performance for tunnel users because the
 923                packets will be distributed according to the encapsulated
 924                flows.
 925
 926        encap3+4
 927
 928                This policy uses the same formula as layer3+4 but it
 929                relies on skb_flow_dissect to obtain the header fields
 930                which might result in the use of inner headers if an
 931                encapsulation protocol is used. For example this will
 932                improve the performance for tunnel users because the
 933                packets will be distributed according to the encapsulated
 934                flows.
 935
 936        The default value is layer2.  This option was added in bonding
 937        version 2.6.3.  In earlier versions of bonding, this parameter
 938        does not exist, and the layer2 policy is the only policy.  The
 939        layer2+3 value was added for bonding version 3.2.2.
 940
 941resend_igmp
 942
 943        Specifies the number of IGMP membership reports to be issued after
 944        a failover event. One membership report is issued immediately after
 945        the failover, subsequent packets are sent in each 200ms interval.
 946
 947        The valid range is 0 - 255; the default value is 1. A value of 0
 948        prevents the IGMP membership report from being issued in response
 949        to the failover event.
 950
 951        This option is useful for bonding modes balance-rr (0), active-backup
 952        (1), balance-tlb (5) and balance-alb (6), in which a failover can
 953        switch the IGMP traffic from one slave to another.  Therefore a fresh
 954        IGMP report must be issued to cause the switch to forward the incoming
 955        IGMP traffic over the newly selected slave.
 956
 957        This option was added for bonding version 3.7.0.
 958
 959lp_interval
 960
 961        Specifies the number of seconds between instances where the bonding
 962        driver sends learning packets to each slaves peer switch.
 963
 964        The valid range is 1 - 0x7fffffff; the default value is 1. This Option
 965        has effect only in balance-tlb and balance-alb modes.
 966
 9673. Configuring Bonding Devices
 968==============================
 969
 970        You can configure bonding using either your distro's network
 971initialization scripts, or manually using either iproute2 or the
 972sysfs interface.  Distros generally use one of three packages for the
 973network initialization scripts: initscripts, sysconfig or interfaces.
 974Recent versions of these packages have support for bonding, while older
 975versions do not.
 976
 977        We will first describe the options for configuring bonding for
 978distros using versions of initscripts, sysconfig and interfaces with full
 979or partial support for bonding, then provide information on enabling
 980bonding without support from the network initialization scripts (i.e.,
 981older versions of initscripts or sysconfig).
 982
 983        If you're unsure whether your distro uses sysconfig,
 984initscripts or interfaces, or don't know if it's new enough, have no fear.
 985Determining this is fairly straightforward.
 986
 987        First, look for a file called interfaces in /etc/network directory.
 988If this file is present in your system, then your system use interfaces. See
 989Configuration with Interfaces Support.
 990
 991        Else, issue the command:
 992
 993$ rpm -qf /sbin/ifup
 994
 995        It will respond with a line of text starting with either
 996"initscripts" or "sysconfig," followed by some numbers.  This is the
 997package that provides your network initialization scripts.
 998
 999        Next, to determine if your installation supports bonding,
1000issue the command:
1001
1002$ grep ifenslave /sbin/ifup
1003
1004        If this returns any matches, then your initscripts or
1005sysconfig has support for bonding.
1006
10073.1 Configuration with Sysconfig Support
1008----------------------------------------
1009
1010        This section applies to distros using a version of sysconfig
1011with bonding support, for example, SuSE Linux Enterprise Server 9.
1012
1013        SuSE SLES 9's networking configuration system does support
1014bonding, however, at this writing, the YaST system configuration
1015front end does not provide any means to work with bonding devices.
1016Bonding devices can be managed by hand, however, as follows.
1017
1018        First, if they have not already been configured, configure the
1019slave devices.  On SLES 9, this is most easily done by running the
1020yast2 sysconfig configuration utility.  The goal is for to create an
1021ifcfg-id file for each slave device.  The simplest way to accomplish
1022this is to configure the devices for DHCP (this is only to get the
1023file ifcfg-id file created; see below for some issues with DHCP).  The
1024name of the configuration file for each device will be of the form:
1025
1026ifcfg-id-xx:xx:xx:xx:xx:xx
1027
1028        Where the "xx" portion will be replaced with the digits from
1029the device's permanent MAC address.
1030
1031        Once the set of ifcfg-id-xx:xx:xx:xx:xx:xx files has been
1032created, it is necessary to edit the configuration files for the slave
1033devices (the MAC addresses correspond to those of the slave devices).
1034Before editing, the file will contain multiple lines, and will look
1035something like this:
1036
1037BOOTPROTO='dhcp'
1038STARTMODE='on'
1039USERCTL='no'
1040UNIQUE='XNzu.WeZGOGF+4wE'
1041_nm_name='bus-pci-0001:61:01.0'
1042
1043        Change the BOOTPROTO and STARTMODE lines to the following:
1044
1045BOOTPROTO='none'
1046STARTMODE='off'
1047
1048        Do not alter the UNIQUE or _nm_name lines.  Remove any other
1049lines (USERCTL, etc).
1050
1051        Once the ifcfg-id-xx:xx:xx:xx:xx:xx files have been modified,
1052it's time to create the configuration file for the bonding device
1053itself.  This file is named ifcfg-bondX, where X is the number of the
1054bonding device to create, starting at 0.  The first such file is
1055ifcfg-bond0, the second is ifcfg-bond1, and so on.  The sysconfig
1056network configuration system will correctly start multiple instances
1057of bonding.
1058
1059        The contents of the ifcfg-bondX file is as follows:
1060
1061BOOTPROTO="static"
1062BROADCAST="10.0.2.255"
1063IPADDR="10.0.2.10"
1064NETMASK="255.255.0.0"
1065NETWORK="10.0.2.0"
1066REMOTE_IPADDR=""
1067STARTMODE="onboot"
1068BONDING_MASTER="yes"
1069BONDING_MODULE_OPTS="mode=active-backup miimon=100"
1070BONDING_SLAVE0="eth0"
1071BONDING_SLAVE1="bus-pci-0000:06:08.1"
1072
1073        Replace the sample BROADCAST, IPADDR, NETMASK and NETWORK
1074values with the appropriate values for your network.
1075
1076        The STARTMODE specifies when the device is brought online.
1077The possible values are:
1078
1079        onboot:  The device is started at boot time.  If you're not
1080                 sure, this is probably what you want.
1081
1082        manual:  The device is started only when ifup is called
1083                 manually.  Bonding devices may be configured this
1084                 way if you do not wish them to start automatically
1085                 at boot for some reason.
1086
1087        hotplug: The device is started by a hotplug event.  This is not
1088                 a valid choice for a bonding device.
1089
1090        off or ignore: The device configuration is ignored.
1091
1092        The line BONDING_MASTER='yes' indicates that the device is a
1093bonding master device.  The only useful value is "yes."
1094
1095        The contents of BONDING_MODULE_OPTS are supplied to the
1096instance of the bonding module for this device.  Specify the options
1097for the bonding mode, link monitoring, and so on here.  Do not include
1098the max_bonds bonding parameter; this will confuse the configuration
1099system if you have multiple bonding devices.
1100
1101        Finally, supply one BONDING_SLAVEn="slave device" for each
1102slave.  where "n" is an increasing value, one for each slave.  The
1103"slave device" is either an interface name, e.g., "eth0", or a device
1104specifier for the network device.  The interface name is easier to
1105find, but the ethN names are subject to change at boot time if, e.g.,
1106a device early in the sequence has failed.  The device specifiers
1107(bus-pci-0000:06:08.1 in the example above) specify the physical
1108network device, and will not change unless the device's bus location
1109changes (for example, it is moved from one PCI slot to another).  The
1110example above uses one of each type for demonstration purposes; most
1111configurations will choose one or the other for all slave devices.
1112
1113        When all configuration files have been modified or created,
1114networking must be restarted for the configuration changes to take
1115effect.  This can be accomplished via the following:
1116
1117# /etc/init.d/network restart
1118
1119        Note that the network control script (/sbin/ifdown) will
1120remove the bonding module as part of the network shutdown processing,
1121so it is not necessary to remove the module by hand if, e.g., the
1122module parameters have changed.
1123
1124        Also, at this writing, YaST/YaST2 will not manage bonding
1125devices (they do not show bonding interfaces on its list of network
1126devices).  It is necessary to edit the configuration file by hand to
1127change the bonding configuration.
1128
1129        Additional general options and details of the ifcfg file
1130format can be found in an example ifcfg template file:
1131
1132/etc/sysconfig/network/ifcfg.template
1133
1134        Note that the template does not document the various BONDING_
1135settings described above, but does describe many of the other options.
1136
11373.1.1 Using DHCP with Sysconfig
1138-------------------------------
1139
1140        Under sysconfig, configuring a device with BOOTPROTO='dhcp'
1141will cause it to query DHCP for its IP address information.  At this
1142writing, this does not function for bonding devices; the scripts
1143attempt to obtain the device address from DHCP prior to adding any of
1144the slave devices.  Without active slaves, the DHCP requests are not
1145sent to the network.
1146
11473.1.2 Configuring Multiple Bonds with Sysconfig
1148-----------------------------------------------
1149
1150        The sysconfig network initialization system is capable of
1151handling multiple bonding devices.  All that is necessary is for each
1152bonding instance to have an appropriately configured ifcfg-bondX file
1153(as described above).  Do not specify the "max_bonds" parameter to any
1154instance of bonding, as this will confuse sysconfig.  If you require
1155multiple bonding devices with identical parameters, create multiple
1156ifcfg-bondX files.
1157
1158        Because the sysconfig scripts supply the bonding module
1159options in the ifcfg-bondX file, it is not necessary to add them to
1160the system /etc/modules.d/*.conf configuration files.
1161
11623.2 Configuration with Initscripts Support
1163------------------------------------------
1164
1165        This section applies to distros using a recent version of
1166initscripts with bonding support, for example, Red Hat Enterprise Linux
1167version 3 or later, Fedora, etc.  On these systems, the network
1168initialization scripts have knowledge of bonding, and can be configured to
1169control bonding devices.  Note that older versions of the initscripts
1170package have lower levels of support for bonding; this will be noted where
1171applicable.
1172
1173        These distros will not automatically load the network adapter
1174driver unless the ethX device is configured with an IP address.
1175Because of this constraint, users must manually configure a
1176network-script file for all physical adapters that will be members of
1177a bondX link.  Network script files are located in the directory:
1178
1179/etc/sysconfig/network-scripts
1180
1181        The file name must be prefixed with "ifcfg-eth" and suffixed
1182with the adapter's physical adapter number.  For example, the script
1183for eth0 would be named /etc/sysconfig/network-scripts/ifcfg-eth0.
1184Place the following text in the file:
1185
1186DEVICE=eth0
1187USERCTL=no
1188ONBOOT=yes
1189MASTER=bond0
1190SLAVE=yes
1191BOOTPROTO=none
1192
1193        The DEVICE= line will be different for every ethX device and
1194must correspond with the name of the file, i.e., ifcfg-eth1 must have
1195a device line of DEVICE=eth1.  The setting of the MASTER= line will
1196also depend on the final bonding interface name chosen for your bond.
1197As with other network devices, these typically start at 0, and go up
1198one for each device, i.e., the first bonding instance is bond0, the
1199second is bond1, and so on.
1200
1201        Next, create a bond network script.  The file name for this
1202script will be /etc/sysconfig/network-scripts/ifcfg-bondX where X is
1203the number of the bond.  For bond0 the file is named "ifcfg-bond0",
1204for bond1 it is named "ifcfg-bond1", and so on.  Within that file,
1205place the following text:
1206
1207DEVICE=bond0
1208IPADDR=192.168.1.1
1209NETMASK=255.255.255.0
1210NETWORK=192.168.1.0
1211BROADCAST=192.168.1.255
1212ONBOOT=yes
1213BOOTPROTO=none
1214USERCTL=no
1215
1216        Be sure to change the networking specific lines (IPADDR,
1217NETMASK, NETWORK and BROADCAST) to match your network configuration.
1218
1219        For later versions of initscripts, such as that found with Fedora
12207 (or later) and Red Hat Enterprise Linux version 5 (or later), it is possible,
1221and, indeed, preferable, to specify the bonding options in the ifcfg-bond0
1222file, e.g. a line of the format:
1223
1224BONDING_OPTS="mode=active-backup arp_interval=60 arp_ip_target=192.168.1.254"
1225
1226        will configure the bond with the specified options.  The options
1227specified in BONDING_OPTS are identical to the bonding module parameters
1228except for the arp_ip_target field when using versions of initscripts older
1229than and 8.57 (Fedora 8) and 8.45.19 (Red Hat Enterprise Linux 5.2).  When
1230using older versions each target should be included as a separate option and
1231should be preceded by a '+' to indicate it should be added to the list of
1232queried targets, e.g.,
1233
1234        arp_ip_target=+192.168.1.1 arp_ip_target=+192.168.1.2
1235
1236        is the proper syntax to specify multiple targets.  When specifying
1237options via BONDING_OPTS, it is not necessary to edit /etc/modprobe.d/*.conf.
1238
1239        For even older versions of initscripts that do not support
1240BONDING_OPTS, it is necessary to edit /etc/modprobe.d/*.conf, depending upon
1241your distro) to load the bonding module with your desired options when the
1242bond0 interface is brought up.  The following lines in /etc/modprobe.d/*.conf
1243will load the bonding module, and select its options:
1244
1245alias bond0 bonding
1246options bond0 mode=balance-alb miimon=100
1247
1248        Replace the sample parameters with the appropriate set of
1249options for your configuration.
1250
1251        Finally run "/etc/rc.d/init.d/network restart" as root.  This
1252will restart the networking subsystem and your bond link should be now
1253up and running.
1254
12553.2.1 Using DHCP with Initscripts
1256---------------------------------
1257
1258        Recent versions of initscripts (the versions supplied with Fedora
1259Core 3 and Red Hat Enterprise Linux 4, or later versions, are reported to
1260work) have support for assigning IP information to bonding devices via
1261DHCP.
1262
1263        To configure bonding for DHCP, configure it as described
1264above, except replace the line "BOOTPROTO=none" with "BOOTPROTO=dhcp"
1265and add a line consisting of "TYPE=Bonding".  Note that the TYPE value
1266is case sensitive.
1267
12683.2.2 Configuring Multiple Bonds with Initscripts
1269-------------------------------------------------
1270
1271        Initscripts packages that are included with Fedora 7 and Red Hat
1272Enterprise Linux 5 support multiple bonding interfaces by simply
1273specifying the appropriate BONDING_OPTS= in ifcfg-bondX where X is the
1274number of the bond.  This support requires sysfs support in the kernel,
1275and a bonding driver of version 3.0.0 or later.  Other configurations may
1276not support this method for specifying multiple bonding interfaces; for
1277those instances, see the "Configuring Multiple Bonds Manually" section,
1278below.
1279
12803.3 Configuring Bonding Manually with iproute2
1281-----------------------------------------------
1282
1283        This section applies to distros whose network initialization
1284scripts (the sysconfig or initscripts package) do not have specific
1285knowledge of bonding.  One such distro is SuSE Linux Enterprise Server
1286version 8.
1287
1288        The general method for these systems is to place the bonding
1289module parameters into a config file in /etc/modprobe.d/ (as
1290appropriate for the installed distro), then add modprobe and/or
1291`ip link` commands to the system's global init script.  The name of
1292the global init script differs; for sysconfig, it is
1293/etc/init.d/boot.local and for initscripts it is /etc/rc.d/rc.local.
1294
1295        For example, if you wanted to make a simple bond of two e100
1296devices (presumed to be eth0 and eth1), and have it persist across
1297reboots, edit the appropriate file (/etc/init.d/boot.local or
1298/etc/rc.d/rc.local), and add the following:
1299
1300modprobe bonding mode=balance-alb miimon=100
1301modprobe e100
1302ifconfig bond0 192.168.1.1 netmask 255.255.255.0 up
1303ip link set eth0 master bond0
1304ip link set eth1 master bond0
1305
1306        Replace the example bonding module parameters and bond0
1307network configuration (IP address, netmask, etc) with the appropriate
1308values for your configuration.
1309
1310        Unfortunately, this method will not provide support for the
1311ifup and ifdown scripts on the bond devices.  To reload the bonding
1312configuration, it is necessary to run the initialization script, e.g.,
1313
1314# /etc/init.d/boot.local
1315
1316        or
1317
1318# /etc/rc.d/rc.local
1319
1320        It may be desirable in such a case to create a separate script
1321which only initializes the bonding configuration, then call that
1322separate script from within boot.local.  This allows for bonding to be
1323enabled without re-running the entire global init script.
1324
1325        To shut down the bonding devices, it is necessary to first
1326mark the bonding device itself as being down, then remove the
1327appropriate device driver modules.  For our example above, you can do
1328the following:
1329
1330# ifconfig bond0 down
1331# rmmod bonding
1332# rmmod e100
1333
1334        Again, for convenience, it may be desirable to create a script
1335with these commands.
1336
1337
13383.3.1 Configuring Multiple Bonds Manually
1339-----------------------------------------
1340
1341        This section contains information on configuring multiple
1342bonding devices with differing options for those systems whose network
1343initialization scripts lack support for configuring multiple bonds.
1344
1345        If you require multiple bonding devices, but all with the same
1346options, you may wish to use the "max_bonds" module parameter,
1347documented above.
1348
1349        To create multiple bonding devices with differing options, it is
1350preferable to use bonding parameters exported by sysfs, documented in the
1351section below.
1352
1353        For versions of bonding without sysfs support, the only means to
1354provide multiple instances of bonding with differing options is to load
1355the bonding driver multiple times.  Note that current versions of the
1356sysconfig network initialization scripts handle this automatically; if
1357your distro uses these scripts, no special action is needed.  See the
1358section Configuring Bonding Devices, above, if you're not sure about your
1359network initialization scripts.
1360
1361        To load multiple instances of the module, it is necessary to
1362specify a different name for each instance (the module loading system
1363requires that every loaded module, even multiple instances of the same
1364module, have a unique name).  This is accomplished by supplying multiple
1365sets of bonding options in /etc/modprobe.d/*.conf, for example:
1366
1367alias bond0 bonding
1368options bond0 -o bond0 mode=balance-rr miimon=100
1369
1370alias bond1 bonding
1371options bond1 -o bond1 mode=balance-alb miimon=50
1372
1373        will load the bonding module two times.  The first instance is
1374named "bond0" and creates the bond0 device in balance-rr mode with an
1375miimon of 100.  The second instance is named "bond1" and creates the
1376bond1 device in balance-alb mode with an miimon of 50.
1377
1378        In some circumstances (typically with older distributions),
1379the above does not work, and the second bonding instance never sees
1380its options.  In that case, the second options line can be substituted
1381as follows:
1382
1383install bond1 /sbin/modprobe --ignore-install bonding -o bond1 \
1384        mode=balance-alb miimon=50
1385
1386        This may be repeated any number of times, specifying a new and
1387unique name in place of bond1 for each subsequent instance.
1388
1389        It has been observed that some Red Hat supplied kernels are unable
1390to rename modules at load time (the "-o bond1" part).  Attempts to pass
1391that option to modprobe will produce an "Operation not permitted" error.
1392This has been reported on some Fedora Core kernels, and has been seen on
1393RHEL 4 as well.  On kernels exhibiting this problem, it will be impossible
1394to configure multiple bonds with differing parameters (as they are older
1395kernels, and also lack sysfs support).
1396
13973.4 Configuring Bonding Manually via Sysfs
1398------------------------------------------
1399
1400        Starting with version 3.0.0, Channel Bonding may be configured
1401via the sysfs interface.  This interface allows dynamic configuration
1402of all bonds in the system without unloading the module.  It also
1403allows for adding and removing bonds at runtime.  Ifenslave is no
1404longer required, though it is still supported.
1405
1406        Use of the sysfs interface allows you to use multiple bonds
1407with different configurations without having to reload the module.
1408It also allows you to use multiple, differently configured bonds when
1409bonding is compiled into the kernel.
1410
1411        You must have the sysfs filesystem mounted to configure
1412bonding this way.  The examples in this document assume that you
1413are using the standard mount point for sysfs, e.g. /sys.  If your
1414sysfs filesystem is mounted elsewhere, you will need to adjust the
1415example paths accordingly.
1416
1417Creating and Destroying Bonds
1418-----------------------------
1419To add a new bond foo:
1420# echo +foo > /sys/class/net/bonding_masters
1421
1422To remove an existing bond bar:
1423# echo -bar > /sys/class/net/bonding_masters
1424
1425To show all existing bonds:
1426# cat /sys/class/net/bonding_masters
1427
1428NOTE: due to 4K size limitation of sysfs files, this list may be
1429truncated if you have more than a few hundred bonds.  This is unlikely
1430to occur under normal operating conditions.
1431
1432Adding and Removing Slaves
1433--------------------------
1434        Interfaces may be enslaved to a bond using the file
1435/sys/class/net/<bond>/bonding/slaves.  The semantics for this file
1436are the same as for the bonding_masters file.
1437
1438To enslave interface eth0 to bond bond0:
1439# ifconfig bond0 up
1440# echo +eth0 > /sys/class/net/bond0/bonding/slaves
1441
1442To free slave eth0 from bond bond0:
1443# echo -eth0 > /sys/class/net/bond0/bonding/slaves
1444
1445        When an interface is enslaved to a bond, symlinks between the
1446two are created in the sysfs filesystem.  In this case, you would get
1447/sys/class/net/bond0/slave_eth0 pointing to /sys/class/net/eth0, and
1448/sys/class/net/eth0/master pointing to /sys/class/net/bond0.
1449
1450        This means that you can tell quickly whether or not an
1451interface is enslaved by looking for the master symlink.  Thus:
1452# echo -eth0 > /sys/class/net/eth0/master/bonding/slaves
1453will free eth0 from whatever bond it is enslaved to, regardless of
1454the name of the bond interface.
1455
1456Changing a Bond's Configuration
1457-------------------------------
1458        Each bond may be configured individually by manipulating the
1459files located in /sys/class/net/<bond name>/bonding
1460
1461        The names of these files correspond directly with the command-
1462line parameters described elsewhere in this file, and, with the
1463exception of arp_ip_target, they accept the same values.  To see the
1464current setting, simply cat the appropriate file.
1465
1466        A few examples will be given here; for specific usage
1467guidelines for each parameter, see the appropriate section in this
1468document.
1469
1470To configure bond0 for balance-alb mode:
1471# ifconfig bond0 down
1472# echo 6 > /sys/class/net/bond0/bonding/mode
1473 - or -
1474# echo balance-alb > /sys/class/net/bond0/bonding/mode
1475        NOTE: The bond interface must be down before the mode can be
1476changed.
1477
1478To enable MII monitoring on bond0 with a 1 second interval:
1479# echo 1000 > /sys/class/net/bond0/bonding/miimon
1480        NOTE: If ARP monitoring is enabled, it will disabled when MII
1481monitoring is enabled, and vice-versa.
1482
1483To add ARP targets:
1484# echo +192.168.0.100 > /sys/class/net/bond0/bonding/arp_ip_target
1485# echo +192.168.0.101 > /sys/class/net/bond0/bonding/arp_ip_target
1486        NOTE:  up to 16 target addresses may be specified.
1487
1488To remove an ARP target:
1489# echo -192.168.0.100 > /sys/class/net/bond0/bonding/arp_ip_target
1490
1491To configure the interval between learning packet transmits:
1492# echo 12 > /sys/class/net/bond0/bonding/lp_interval
1493        NOTE: the lp_inteval is the number of seconds between instances where
1494the bonding driver sends learning packets to each slaves peer switch.  The
1495default interval is 1 second.
1496
1497Example Configuration
1498---------------------
1499        We begin with the same example that is shown in section 3.3,
1500executed with sysfs, and without using ifenslave.
1501
1502        To make a simple bond of two e100 devices (presumed to be eth0
1503and eth1), and have it persist across reboots, edit the appropriate
1504file (/etc/init.d/boot.local or /etc/rc.d/rc.local), and add the
1505following:
1506
1507modprobe bonding
1508modprobe e100
1509echo balance-alb > /sys/class/net/bond0/bonding/mode
1510ifconfig bond0 192.168.1.1 netmask 255.255.255.0 up
1511echo 100 > /sys/class/net/bond0/bonding/miimon
1512echo +eth0 > /sys/class/net/bond0/bonding/slaves
1513echo +eth1 > /sys/class/net/bond0/bonding/slaves
1514
1515        To add a second bond, with two e1000 interfaces in
1516active-backup mode, using ARP monitoring, add the following lines to
1517your init script:
1518
1519modprobe e1000
1520echo +bond1 > /sys/class/net/bonding_masters
1521echo active-backup > /sys/class/net/bond1/bonding/mode
1522ifconfig bond1 192.168.2.1 netmask 255.255.255.0 up
1523echo +192.168.2.100 /sys/class/net/bond1/bonding/arp_ip_target
1524echo 2000 > /sys/class/net/bond1/bonding/arp_interval
1525echo +eth2 > /sys/class/net/bond1/bonding/slaves
1526echo +eth3 > /sys/class/net/bond1/bonding/slaves
1527
15283.5 Configuration with Interfaces Support
1529-----------------------------------------
1530
1531        This section applies to distros which use /etc/network/interfaces file
1532to describe network interface configuration, most notably Debian and it's
1533derivatives.
1534
1535        The ifup and ifdown commands on Debian don't support bonding out of
1536the box. The ifenslave-2.6 package should be installed to provide bonding
1537support.  Once installed, this package will provide bond-* options to be used
1538into /etc/network/interfaces.
1539
1540        Note that ifenslave-2.6 package will load the bonding module and use
1541the ifenslave command when appropriate.
1542
1543Example Configurations
1544----------------------
1545
1546In /etc/network/interfaces, the following stanza will configure bond0, in
1547active-backup mode, with eth0 and eth1 as slaves.
1548
1549auto bond0
1550iface bond0 inet dhcp
1551        bond-slaves eth0 eth1
1552        bond-mode active-backup
1553        bond-miimon 100
1554        bond-primary eth0 eth1
1555
1556If the above configuration doesn't work, you might have a system using
1557upstart for system startup. This is most notably true for recent
1558Ubuntu versions. The following stanza in /etc/network/interfaces will
1559produce the same result on those systems.
1560
1561auto bond0
1562iface bond0 inet dhcp
1563        bond-slaves none
1564        bond-mode active-backup
1565        bond-miimon 100
1566
1567auto eth0
1568iface eth0 inet manual
1569        bond-master bond0
1570        bond-primary eth0 eth1
1571
1572auto eth1
1573iface eth1 inet manual
1574        bond-master bond0
1575        bond-primary eth0 eth1
1576
1577For a full list of bond-* supported options in /etc/network/interfaces and some
1578more advanced examples tailored to you particular distros, see the files in
1579/usr/share/doc/ifenslave-2.6.
1580
15813.6 Overriding Configuration for Special Cases
1582----------------------------------------------
1583
1584When using the bonding driver, the physical port which transmits a frame is
1585typically selected by the bonding driver, and is not relevant to the user or
1586system administrator.  The output port is simply selected using the policies of
1587the selected bonding mode.  On occasion however, it is helpful to direct certain
1588classes of traffic to certain physical interfaces on output to implement
1589slightly more complex policies.  For example, to reach a web server over a
1590bonded interface in which eth0 connects to a private network, while eth1
1591connects via a public network, it may be desirous to bias the bond to send said
1592traffic over eth0 first, using eth1 only as a fall back, while all other traffic
1593can safely be sent over either interface.  Such configurations may be achieved
1594using the traffic control utilities inherent in linux.
1595
1596By default the bonding driver is multiqueue aware and 16 queues are created
1597when the driver initializes (see Documentation/networking/multiqueue.txt
1598for details).  If more or less queues are desired the module parameter
1599tx_queues can be used to change this value.  There is no sysfs parameter
1600available as the allocation is done at module init time.
1601
1602The output of the file /proc/net/bonding/bondX has changed so the output Queue
1603ID is now printed for each slave:
1604
1605Bonding Mode: fault-tolerance (active-backup)
1606Primary Slave: None
1607Currently Active Slave: eth0
1608MII Status: up
1609MII Polling Interval (ms): 0
1610Up Delay (ms): 0
1611Down Delay (ms): 0
1612
1613Slave Interface: eth0
1614MII Status: up
1615Link Failure Count: 0
1616Permanent HW addr: 00:1a:a0:12:8f:cb
1617Slave queue ID: 0
1618
1619Slave Interface: eth1
1620MII Status: up
1621Link Failure Count: 0
1622Permanent HW addr: 00:1a:a0:12:8f:cc
1623Slave queue ID: 2
1624
1625The queue_id for a slave can be set using the command:
1626
1627# echo "eth1:2" > /sys/class/net/bond0/bonding/queue_id
1628
1629Any interface that needs a queue_id set should set it with multiple calls
1630like the one above until proper priorities are set for all interfaces.  On
1631distributions that allow configuration via initscripts, multiple 'queue_id'
1632arguments can be added to BONDING_OPTS to set all needed slave queues.
1633
1634These queue id's can be used in conjunction with the tc utility to configure
1635a multiqueue qdisc and filters to bias certain traffic to transmit on certain
1636slave devices.  For instance, say we wanted, in the above configuration to
1637force all traffic bound to 192.168.1.100 to use eth1 in the bond as its output
1638device. The following commands would accomplish this:
1639
1640# tc qdisc add dev bond0 handle 1 root multiq
1641
1642# tc filter add dev bond0 protocol ip parent 1: prio 1 u32 match ip dst \
1643        192.168.1.100 action skbedit queue_mapping 2
1644
1645These commands tell the kernel to attach a multiqueue queue discipline to the
1646bond0 interface and filter traffic enqueued to it, such that packets with a dst
1647ip of 192.168.1.100 have their output queue mapping value overwritten to 2.
1648This value is then passed into the driver, causing the normal output path
1649selection policy to be overridden, selecting instead qid 2, which maps to eth1.
1650
1651Note that qid values begin at 1.  Qid 0 is reserved to initiate to the driver
1652that normal output policy selection should take place.  One benefit to simply
1653leaving the qid for a slave to 0 is the multiqueue awareness in the bonding
1654driver that is now present.  This awareness allows tc filters to be placed on
1655slave devices as well as bond devices and the bonding driver will simply act as
1656a pass-through for selecting output queues on the slave device rather than 
1657output port selection.
1658
1659This feature first appeared in bonding driver version 3.7.0 and support for
1660output slave selection was limited to round-robin and active-backup modes.
1661
16623.7 Configuring LACP for 802.3ad mode in a more secure way
1663----------------------------------------------------------
1664
1665When using 802.3ad bonding mode, the Actor (host) and Partner (switch)
1666exchange LACPDUs.  These LACPDUs cannot be sniffed, because they are
1667destined to link local mac addresses (which switches/bridges are not
1668supposed to forward).  However, most of the values are easily predictable
1669or are simply the machine's MAC address (which is trivially known to all
1670other hosts in the same L2).  This implies that other machines in the L2
1671domain can spoof LACPDU packets from other hosts to the switch and potentially
1672cause mayhem by joining (from the point of view of the switch) another
1673machine's aggregate, thus receiving a portion of that hosts incoming
1674traffic and / or spoofing traffic from that machine themselves (potentially
1675even successfully terminating some portion of flows). Though this is not
1676a likely scenario, one could avoid this possibility by simply configuring
1677few bonding parameters:
1678
1679   (a) ad_actor_system : You can set a random mac-address that can be used for
1680       these LACPDU exchanges. The value can not be either NULL or Multicast.
1681       Also it's preferable to set the local-admin bit. Following shell code
1682       generates a random mac-address as described above.
1683
1684       # sys_mac_addr=$(printf '%02x:%02x:%02x:%02x:%02x:%02x' \
1685                                $(( (RANDOM & 0xFE) | 0x02 )) \
1686                                $(( RANDOM & 0xFF )) \
1687                                $(( RANDOM & 0xFF )) \
1688                                $(( RANDOM & 0xFF )) \
1689                                $(( RANDOM & 0xFF )) \
1690                                $(( RANDOM & 0xFF )))
1691       # echo $sys_mac_addr > /sys/class/net/bond0/bonding/ad_actor_system
1692
1693   (b) ad_actor_sys_prio : Randomize the system priority. The default value
1694       is 65535, but system can take the value from 1 - 65535. Following shell
1695       code generates random priority and sets it.
1696
1697       # sys_prio=$(( 1 + RANDOM + RANDOM ))
1698       # echo $sys_prio > /sys/class/net/bond0/bonding/ad_actor_sys_prio
1699
1700   (c) ad_user_port_key : Use the user portion of the port-key. The default
1701       keeps this empty. These are the upper 10 bits of the port-key and value
1702       ranges from 0 - 1023. Following shell code generates these 10 bits and
1703       sets it.
1704
1705       # usr_port_key=$(( RANDOM & 0x3FF ))
1706       # echo $usr_port_key > /sys/class/net/bond0/bonding/ad_user_port_key
1707
1708
17094 Querying Bonding Configuration
1710=================================
1711
17124.1 Bonding Configuration
1713-------------------------
1714
1715        Each bonding device has a read-only file residing in the
1716/proc/net/bonding directory.  The file contents include information
1717about the bonding configuration, options and state of each slave.
1718
1719        For example, the contents of /proc/net/bonding/bond0 after the
1720driver is loaded with parameters of mode=0 and miimon=1000 is
1721generally as follows:
1722
1723        Ethernet Channel Bonding Driver: 2.6.1 (October 29, 2004)
1724        Bonding Mode: load balancing (round-robin)
1725        Currently Active Slave: eth0
1726        MII Status: up
1727        MII Polling Interval (ms): 1000
1728        Up Delay (ms): 0
1729        Down Delay (ms): 0
1730
1731        Slave Interface: eth1
1732        MII Status: up
1733        Link Failure Count: 1
1734
1735        Slave Interface: eth0
1736        MII Status: up
1737        Link Failure Count: 1
1738
1739        The precise format and contents will change depending upon the
1740bonding configuration, state, and version of the bonding driver.
1741
17424.2 Network configuration
1743-------------------------
1744
1745        The network configuration can be inspected using the ifconfig
1746command.  Bonding devices will have the MASTER flag set; Bonding slave
1747devices will have the SLAVE flag set.  The ifconfig output does not
1748contain information on which slaves are associated with which masters.
1749
1750        In the example below, the bond0 interface is the master
1751(MASTER) while eth0 and eth1 are slaves (SLAVE). Notice all slaves of
1752bond0 have the same MAC address (HWaddr) as bond0 for all modes except
1753TLB and ALB that require a unique MAC address for each slave.
1754
1755# /sbin/ifconfig
1756bond0     Link encap:Ethernet  HWaddr 00:C0:F0:1F:37:B4
1757          inet addr:XXX.XXX.XXX.YYY  Bcast:XXX.XXX.XXX.255  Mask:255.255.252.0
1758          UP BROADCAST RUNNING MASTER MULTICAST  MTU:1500  Metric:1
1759          RX packets:7224794 errors:0 dropped:0 overruns:0 frame:0
1760          TX packets:3286647 errors:1 dropped:0 overruns:1 carrier:0
1761          collisions:0 txqueuelen:0
1762
1763eth0      Link encap:Ethernet  HWaddr 00:C0:F0:1F:37:B4
1764          UP BROADCAST RUNNING SLAVE MULTICAST  MTU:1500  Metric:1
1765          RX packets:3573025 errors:0 dropped:0 overruns:0 frame:0
1766          TX packets:1643167 errors:1 dropped:0 overruns:1 carrier:0
1767          collisions:0 txqueuelen:100
1768          Interrupt:10 Base address:0x1080
1769
1770eth1      Link encap:Ethernet  HWaddr 00:C0:F0:1F:37:B4
1771          UP BROADCAST RUNNING SLAVE MULTICAST  MTU:1500  Metric:1
1772          RX packets:3651769 errors:0 dropped:0 overruns:0 frame:0
1773          TX packets:1643480 errors:0 dropped:0 overruns:0 carrier:0
1774          collisions:0 txqueuelen:100
1775          Interrupt:9 Base address:0x1400
1776
17775. Switch Configuration
1778=======================
1779
1780        For this section, "switch" refers to whatever system the
1781bonded devices are directly connected to (i.e., where the other end of
1782the cable plugs into).  This may be an actual dedicated switch device,
1783or it may be another regular system (e.g., another computer running
1784Linux),
1785
1786        The active-backup, balance-tlb and balance-alb modes do not
1787require any specific configuration of the switch.
1788
1789        The 802.3ad mode requires that the switch have the appropriate
1790ports configured as an 802.3ad aggregation.  The precise method used
1791to configure this varies from switch to switch, but, for example, a
1792Cisco 3550 series switch requires that the appropriate ports first be
1793grouped together in a single etherchannel instance, then that
1794etherchannel is set to mode "lacp" to enable 802.3ad (instead of
1795standard EtherChannel).
1796
1797        The balance-rr, balance-xor and broadcast modes generally
1798require that the switch have the appropriate ports grouped together.
1799The nomenclature for such a group differs between switches, it may be
1800called an "etherchannel" (as in the Cisco example, above), a "trunk
1801group" or some other similar variation.  For these modes, each switch
1802will also have its own configuration options for the switch's transmit
1803policy to the bond.  Typical choices include XOR of either the MAC or
1804IP addresses.  The transmit policy of the two peers does not need to
1805match.  For these three modes, the bonding mode really selects a
1806transmit policy for an EtherChannel group; all three will interoperate
1807with another EtherChannel group.
1808
1809
18106. 802.1q VLAN Support
1811======================
1812
1813        It is possible to configure VLAN devices over a bond interface
1814using the 8021q driver.  However, only packets coming from the 8021q
1815driver and passing through bonding will be tagged by default.  Self
1816generated packets, for example, bonding's learning packets or ARP
1817packets generated by either ALB mode or the ARP monitor mechanism, are
1818tagged internally by bonding itself.  As a result, bonding must
1819"learn" the VLAN IDs configured above it, and use those IDs to tag
1820self generated packets.
1821
1822        For reasons of simplicity, and to support the use of adapters
1823that can do VLAN hardware acceleration offloading, the bonding
1824interface declares itself as fully hardware offloading capable, it gets
1825the add_vid/kill_vid notifications to gather the necessary
1826information, and it propagates those actions to the slaves.  In case
1827of mixed adapter types, hardware accelerated tagged packets that
1828should go through an adapter that is not offloading capable are
1829"un-accelerated" by the bonding driver so the VLAN tag sits in the
1830regular location.
1831
1832        VLAN interfaces *must* be added on top of a bonding interface
1833only after enslaving at least one slave.  The bonding interface has a
1834hardware address of 00:00:00:00:00:00 until the first slave is added.
1835If the VLAN interface is created prior to the first enslavement, it
1836would pick up the all-zeroes hardware address.  Once the first slave
1837is attached to the bond, the bond device itself will pick up the
1838slave's hardware address, which is then available for the VLAN device.
1839
1840        Also, be aware that a similar problem can occur if all slaves
1841are released from a bond that still has one or more VLAN interfaces on
1842top of it.  When a new slave is added, the bonding interface will
1843obtain its hardware address from the first slave, which might not
1844match the hardware address of the VLAN interfaces (which was
1845ultimately copied from an earlier slave).
1846
1847        There are two methods to insure that the VLAN device operates
1848with the correct hardware address if all slaves are removed from a
1849bond interface:
1850
1851        1. Remove all VLAN interfaces then recreate them
1852
1853        2. Set the bonding interface's hardware address so that it
1854matches the hardware address of the VLAN interfaces.
1855
1856        Note that changing a VLAN interface's HW address would set the
1857underlying device -- i.e. the bonding interface -- to promiscuous
1858mode, which might not be what you want.
1859
1860
18617. Link Monitoring
1862==================
1863
1864        The bonding driver at present supports two schemes for
1865monitoring a slave device's link state: the ARP monitor and the MII
1866monitor.
1867
1868        At the present time, due to implementation restrictions in the
1869bonding driver itself, it is not possible to enable both ARP and MII
1870monitoring simultaneously.
1871
18727.1 ARP Monitor Operation
1873-------------------------
1874
1875        The ARP monitor operates as its name suggests: it sends ARP
1876queries to one or more designated peer systems on the network, and
1877uses the response as an indication that the link is operating.  This
1878gives some assurance that traffic is actually flowing to and from one
1879or more peers on the local network.
1880
1881        The ARP monitor relies on the device driver itself to verify
1882that traffic is flowing.  In particular, the driver must keep up to
1883date the last receive time, dev->last_rx.  Drivers that use NETIF_F_LLTX
1884flag must also update netdev_queue->trans_start.  If they do not, then the
1885ARP monitor will immediately fail any slaves using that driver, and
1886those slaves will stay down.  If networking monitoring (tcpdump, etc)
1887shows the ARP requests and replies on the network, then it may be that
1888your device driver is not updating last_rx and trans_start.
1889
18907.2 Configuring Multiple ARP Targets
1891------------------------------------
1892
1893        While ARP monitoring can be done with just one target, it can
1894be useful in a High Availability setup to have several targets to
1895monitor.  In the case of just one target, the target itself may go
1896down or have a problem making it unresponsive to ARP requests.  Having
1897an additional target (or several) increases the reliability of the ARP
1898monitoring.
1899
1900        Multiple ARP targets must be separated by commas as follows:
1901
1902# example options for ARP monitoring with three targets
1903alias bond0 bonding
1904options bond0 arp_interval=60 arp_ip_target=192.168.0.1,192.168.0.3,192.168.0.9
1905
1906        For just a single target the options would resemble:
1907
1908# example options for ARP monitoring with one target
1909alias bond0 bonding
1910options bond0 arp_interval=60 arp_ip_target=192.168.0.100
1911
1912
19137.3 MII Monitor Operation
1914-------------------------
1915
1916        The MII monitor monitors only the carrier state of the local
1917network interface.  It accomplishes this in one of three ways: by
1918depending upon the device driver to maintain its carrier state, by
1919querying the device's MII registers, or by making an ethtool query to
1920the device.
1921
1922        If the use_carrier module parameter is 1 (the default value),
1923then the MII monitor will rely on the driver for carrier state
1924information (via the netif_carrier subsystem).  As explained in the
1925use_carrier parameter information, above, if the MII monitor fails to
1926detect carrier loss on the device (e.g., when the cable is physically
1927disconnected), it may be that the driver does not support
1928netif_carrier.
1929
1930        If use_carrier is 0, then the MII monitor will first query the
1931device's (via ioctl) MII registers and check the link state.  If that
1932request fails (not just that it returns carrier down), then the MII
1933monitor will make an ethtool ETHOOL_GLINK request to attempt to obtain
1934the same information.  If both methods fail (i.e., the driver either
1935does not support or had some error in processing both the MII register
1936and ethtool requests), then the MII monitor will assume the link is
1937up.
1938
19398. Potential Sources of Trouble
1940===============================
1941
19428.1 Adventures in Routing
1943-------------------------
1944
1945        When bonding is configured, it is important that the slave
1946devices not have routes that supersede routes of the master (or,
1947generally, not have routes at all).  For example, suppose the bonding
1948device bond0 has two slaves, eth0 and eth1, and the routing table is
1949as follows:
1950
1951Kernel IP routing table
1952Destination     Gateway         Genmask         Flags   MSS Window  irtt Iface
195310.0.0.0        0.0.0.0         255.255.0.0     U        40 0          0 eth0
195410.0.0.0        0.0.0.0         255.255.0.0     U        40 0          0 eth1
195510.0.0.0        0.0.0.0         255.255.0.0     U        40 0          0 bond0
1956127.0.0.0       0.0.0.0         255.0.0.0       U        40 0          0 lo
1957
1958        This routing configuration will likely still update the
1959receive/transmit times in the driver (needed by the ARP monitor), but
1960may bypass the bonding driver (because outgoing traffic to, in this
1961case, another host on network 10 would use eth0 or eth1 before bond0).
1962
1963        The ARP monitor (and ARP itself) may become confused by this
1964configuration, because ARP requests (generated by the ARP monitor)
1965will be sent on one interface (bond0), but the corresponding reply
1966will arrive on a different interface (eth0).  This reply looks to ARP
1967as an unsolicited ARP reply (because ARP matches replies on an
1968interface basis), and is discarded.  The MII monitor is not affected
1969by the state of the routing table.
1970
1971        The solution here is simply to insure that slaves do not have
1972routes of their own, and if for some reason they must, those routes do
1973not supersede routes of their master.  This should generally be the
1974case, but unusual configurations or errant manual or automatic static
1975route additions may cause trouble.
1976
19778.2 Ethernet Device Renaming
1978----------------------------
1979
1980        On systems with network configuration scripts that do not
1981associate physical devices directly with network interface names (so
1982that the same physical device always has the same "ethX" name), it may
1983be necessary to add some special logic to config files in
1984/etc/modprobe.d/.
1985
1986        For example, given a modules.conf containing the following:
1987
1988alias bond0 bonding
1989options bond0 mode=some-mode miimon=50
1990alias eth0 tg3
1991alias eth1 tg3
1992alias eth2 e1000
1993alias eth3 e1000
1994
1995        If neither eth0 and eth1 are slaves to bond0, then when the
1996bond0 interface comes up, the devices may end up reordered.  This
1997happens because bonding is loaded first, then its slave device's
1998drivers are loaded next.  Since no other drivers have been loaded,
1999when the e1000 driver loads, it will receive eth0 and eth1 for its
2000devices, but the bonding configuration tries to enslave eth2 and eth3
2001(which may later be assigned to the tg3 devices).
2002
2003        Adding the following:
2004
2005add above bonding e1000 tg3
2006
2007        causes modprobe to load e1000 then tg3, in that order, when
2008bonding is loaded.  This command is fully documented in the
2009modules.conf manual page.
2010
2011        On systems utilizing modprobe an equivalent problem can occur.
2012In this case, the following can be added to config files in
2013/etc/modprobe.d/ as:
2014
2015softdep bonding pre: tg3 e1000
2016
2017        This will load tg3 and e1000 modules before loading the bonding one.
2018Full documentation on this can be found in the modprobe.d and modprobe
2019manual pages.
2020
20218.3. Painfully Slow Or No Failed Link Detection By Miimon
2022---------------------------------------------------------
2023
2024        By default, bonding enables the use_carrier option, which
2025instructs bonding to trust the driver to maintain carrier state.
2026
2027        As discussed in the options section, above, some drivers do
2028not support the netif_carrier_on/_off link state tracking system.
2029With use_carrier enabled, bonding will always see these links as up,
2030regardless of their actual state.
2031
2032        Additionally, other drivers do support netif_carrier, but do
2033not maintain it in real time, e.g., only polling the link state at
2034some fixed interval.  In this case, miimon will detect failures, but
2035only after some long period of time has expired.  If it appears that
2036miimon is very slow in detecting link failures, try specifying
2037use_carrier=0 to see if that improves the failure detection time.  If
2038it does, then it may be that the driver checks the carrier state at a
2039fixed interval, but does not cache the MII register values (so the
2040use_carrier=0 method of querying the registers directly works).  If
2041use_carrier=0 does not improve the failover, then the driver may cache
2042the registers, or the problem may be elsewhere.
2043
2044        Also, remember that miimon only checks for the device's
2045carrier state.  It has no way to determine the state of devices on or
2046beyond other ports of a switch, or if a switch is refusing to pass
2047traffic while still maintaining carrier on.
2048
20499. SNMP agents
2050===============
2051
2052        If running SNMP agents, the bonding driver should be loaded
2053before any network drivers participating in a bond.  This requirement
2054is due to the interface index (ipAdEntIfIndex) being associated to
2055the first interface found with a given IP address.  That is, there is
2056only one ipAdEntIfIndex for each IP address.  For example, if eth0 and
2057eth1 are slaves of bond0 and the driver for eth0 is loaded before the
2058bonding driver, the interface for the IP address will be associated
2059with the eth0 interface.  This configuration is shown below, the IP
2060address 192.168.1.1 has an interface index of 2 which indexes to eth0
2061in the ifDescr table (ifDescr.2).
2062
2063     interfaces.ifTable.ifEntry.ifDescr.1 = lo
2064     interfaces.ifTable.ifEntry.ifDescr.2 = eth0
2065     interfaces.ifTable.ifEntry.ifDescr.3 = eth1
2066     interfaces.ifTable.ifEntry.ifDescr.4 = eth2
2067     interfaces.ifTable.ifEntry.ifDescr.5 = eth3
2068     interfaces.ifTable.ifEntry.ifDescr.6 = bond0
2069     ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.10.10.10 = 5
2070     ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.192.168.1.1 = 2
2071     ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.74.20.94 = 4
2072     ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.127.0.0.1 = 1
2073
2074        This problem is avoided by loading the bonding driver before
2075any network drivers participating in a bond.  Below is an example of
2076loading the bonding driver first, the IP address 192.168.1.1 is
2077correctly associated with ifDescr.2.
2078
2079     interfaces.ifTable.ifEntry.ifDescr.1 = lo
2080     interfaces.ifTable.ifEntry.ifDescr.2 = bond0
2081     interfaces.ifTable.ifEntry.ifDescr.3 = eth0
2082     interfaces.ifTable.ifEntry.ifDescr.4 = eth1
2083     interfaces.ifTable.ifEntry.ifDescr.5 = eth2
2084     interfaces.ifTable.ifEntry.ifDescr.6 = eth3
2085     ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.10.10.10 = 6
2086     ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.192.168.1.1 = 2
2087     ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.74.20.94 = 5
2088     ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.127.0.0.1 = 1
2089
2090        While some distributions may not report the interface name in
2091ifDescr, the association between the IP address and IfIndex remains
2092and SNMP functions such as Interface_Scan_Next will report that
2093association.
2094
209510. Promiscuous mode
2096====================
2097
2098        When running network monitoring tools, e.g., tcpdump, it is
2099common to enable promiscuous mode on the device, so that all traffic
2100is seen (instead of seeing only traffic destined for the local host).
2101The bonding driver handles promiscuous mode changes to the bonding
2102master device (e.g., bond0), and propagates the setting to the slave
2103devices.
2104
2105        For the balance-rr, balance-xor, broadcast, and 802.3ad modes,
2106the promiscuous mode setting is propagated to all slaves.
2107
2108        For the active-backup, balance-tlb and balance-alb modes, the
2109promiscuous mode setting is propagated only to the active slave.
2110
2111        For balance-tlb mode, the active slave is the slave currently
2112receiving inbound traffic.
2113
2114        For balance-alb mode, the active slave is the slave used as a
2115"primary."  This slave is used for mode-specific control traffic, for
2116sending to peers that are unassigned or if the load is unbalanced.
2117
2118        For the active-backup, balance-tlb and balance-alb modes, when
2119the active slave changes (e.g., due to a link failure), the
2120promiscuous setting will be propagated to the new active slave.
2121
212211. Configuring Bonding for High Availability
2123=============================================
2124
2125        High Availability refers to configurations that provide
2126maximum network availability by having redundant or backup devices,
2127links or switches between the host and the rest of the world.  The
2128goal is to provide the maximum availability of network connectivity
2129(i.e., the network always works), even though other configurations
2130could provide higher throughput.
2131
213211.1 High Availability in a Single Switch Topology
2133--------------------------------------------------
2134
2135        If two hosts (or a host and a single switch) are directly
2136connected via multiple physical links, then there is no availability
2137penalty to optimizing for maximum bandwidth.  In this case, there is
2138only one switch (or peer), so if it fails, there is no alternative
2139access to fail over to.  Additionally, the bonding load balance modes
2140support link monitoring of their members, so if individual links fail,
2141the load will be rebalanced across the remaining devices.
2142
2143        See Section 12, "Configuring Bonding for Maximum Throughput"
2144for information on configuring bonding with one peer device.
2145
214611.2 High Availability in a Multiple Switch Topology
2147----------------------------------------------------
2148
2149        With multiple switches, the configuration of bonding and the
2150network changes dramatically.  In multiple switch topologies, there is
2151a trade off between network availability and usable bandwidth.
2152
2153        Below is a sample network, configured to maximize the
2154availability of the network:
2155
2156                |                                     |
2157                |port3                           port3|
2158          +-----+----+                          +-----+----+
2159          |          |port2       ISL      port2|          |
2160          | switch A +--------------------------+ switch B |
2161          |          |                          |          |
2162          +-----+----+                          +-----++---+
2163                |port1                           port1|
2164                |             +-------+               |
2165                +-------------+ host1 +---------------+
2166                         eth0 +-------+ eth1
2167
2168        In this configuration, there is a link between the two
2169switches (ISL, or inter switch link), and multiple ports connecting to
2170the outside world ("port3" on each switch).  There is no technical
2171reason that this could not be extended to a third switch.
2172
217311.2.1 HA Bonding Mode Selection for Multiple Switch Topology
2174-------------------------------------------------------------
2175
2176        In a topology such as the example above, the active-backup and
2177broadcast modes are the only useful bonding modes when optimizing for
2178availability; the other modes require all links to terminate on the
2179same peer for them to behave rationally.
2180
2181active-backup: This is generally the preferred mode, particularly if
2182        the switches have an ISL and play together well.  If the
2183        network configuration is such that one switch is specifically
2184        a backup switch (e.g., has lower capacity, higher cost, etc),
2185        then the primary option can be used to insure that the
2186        preferred link is always used when it is available.
2187
2188broadcast: This mode is really a special purpose mode, and is suitable
2189        only for very specific needs.  For example, if the two
2190        switches are not connected (no ISL), and the networks beyond
2191        them are totally independent.  In this case, if it is
2192        necessary for some specific one-way traffic to reach both
2193        independent networks, then the broadcast mode may be suitable.
2194
219511.2.2 HA Link Monitoring Selection for Multiple Switch Topology
2196----------------------------------------------------------------
2197
2198        The choice of link monitoring ultimately depends upon your
2199switch.  If the switch can reliably fail ports in response to other
2200failures, then either the MII or ARP monitors should work.  For
2201example, in the above example, if the "port3" link fails at the remote
2202end, the MII monitor has no direct means to detect this.  The ARP
2203monitor could be configured with a target at the remote end of port3,
2204thus detecting that failure without switch support.
2205
2206        In general, however, in a multiple switch topology, the ARP
2207monitor can provide a higher level of reliability in detecting end to
2208end connectivity failures (which may be caused by the failure of any
2209individual component to pass traffic for any reason).  Additionally,
2210the ARP monitor should be configured with multiple targets (at least
2211one for each switch in the network).  This will insure that,
2212regardless of which switch is active, the ARP monitor has a suitable
2213target to query.
2214
2215        Note, also, that of late many switches now support a functionality
2216generally referred to as "trunk failover."  This is a feature of the
2217switch that causes the link state of a particular switch port to be set
2218down (or up) when the state of another switch port goes down (or up).
2219Its purpose is to propagate link failures from logically "exterior" ports
2220to the logically "interior" ports that bonding is able to monitor via
2221miimon.  Availability and configuration for trunk failover varies by
2222switch, but this can be a viable alternative to the ARP monitor when using
2223suitable switches.
2224
222512. Configuring Bonding for Maximum Throughput
2226==============================================
2227
222812.1 Maximizing Throughput in a Single Switch Topology
2229------------------------------------------------------
2230
2231        In a single switch configuration, the best method to maximize
2232throughput depends upon the application and network environment.  The
2233various load balancing modes each have strengths and weaknesses in
2234different environments, as detailed below.
2235
2236        For this discussion, we will break down the topologies into
2237two categories.  Depending upon the destination of most traffic, we
2238categorize them into either "gatewayed" or "local" configurations.
2239
2240        In a gatewayed configuration, the "switch" is acting primarily
2241as a router, and the majority of traffic passes through this router to
2242other networks.  An example would be the following:
2243
2244
2245     +----------+                     +----------+
2246     |          |eth0            port1|          | to other networks
2247     | Host A   +---------------------+ router   +------------------->
2248     |          +---------------------+          | Hosts B and C are out
2249     |          |eth1            port2|          | here somewhere
2250     +----------+                     +----------+
2251
2252        The router may be a dedicated router device, or another host
2253acting as a gateway.  For our discussion, the important point is that
2254the majority of traffic from Host A will pass through the router to
2255some other network before reaching its final destination.
2256
2257        In a gatewayed network configuration, although Host A may
2258communicate with many other systems, all of its traffic will be sent
2259and received via one other peer on the local network, the router.
2260
2261        Note that the case of two systems connected directly via
2262multiple physical links is, for purposes of configuring bonding, the
2263same as a gatewayed configuration.  In that case, it happens that all
2264traffic is destined for the "gateway" itself, not some other network
2265beyond the gateway.
2266
2267        In a local configuration, the "switch" is acting primarily as
2268a switch, and the majority of traffic passes through this switch to
2269reach other stations on the same network.  An example would be the
2270following:
2271
2272    +----------+            +----------+       +--------+
2273    |          |eth0   port1|          +-------+ Host B |
2274    |  Host A  +------------+  switch  |port3  +--------+
2275    |          +------------+          |                  +--------+
2276    |          |eth1   port2|          +------------------+ Host C |
2277    +----------+            +----------+port4             +--------+
2278
2279
2280        Again, the switch may be a dedicated switch device, or another
2281host acting as a gateway.  For our discussion, the important point is
2282that the majority of traffic from Host A is destined for other hosts
2283on the same local network (Hosts B and C in the above example).
2284
2285        In summary, in a gatewayed configuration, traffic to and from
2286the bonded device will be to the same MAC level peer on the network
2287(the gateway itself, i.e., the router), regardless of its final
2288destination.  In a local configuration, traffic flows directly to and
2289from the final destinations, thus, each destination (Host B, Host C)
2290will be addressed directly by their individual MAC addresses.
2291
2292        This distinction between a gatewayed and a local network
2293configuration is important because many of the load balancing modes
2294available use the MAC addresses of the local network source and
2295destination to make load balancing decisions.  The behavior of each
2296mode is described below.
2297
2298
229912.1.1 MT Bonding Mode Selection for Single Switch Topology
2300-----------------------------------------------------------
2301
2302        This configuration is the easiest to set up and to understand,
2303although you will have to decide which bonding mode best suits your
2304needs.  The trade offs for each mode are detailed below:
2305
2306balance-rr: This mode is the only mode that will permit a single
2307        TCP/IP connection to stripe traffic across multiple
2308        interfaces. It is therefore the only mode that will allow a
2309        single TCP/IP stream to utilize more than one interface's
2310        worth of throughput.  This comes at a cost, however: the
2311        striping generally results in peer systems receiving packets out
2312        of order, causing TCP/IP's congestion control system to kick
2313        in, often by retransmitting segments.
2314
2315        It is possible to adjust TCP/IP's congestion limits by
2316        altering the net.ipv4.tcp_reordering sysctl parameter.  The
2317        usual default value is 3. But keep in mind TCP stack is able
2318        to automatically increase this when it detects reorders.
2319
2320        Note that the fraction of packets that will be delivered out of
2321        order is highly variable, and is unlikely to be zero.  The level
2322        of reordering depends upon a variety of factors, including the
2323        networking interfaces, the switch, and the topology of the
2324        configuration.  Speaking in general terms, higher speed network
2325        cards produce more reordering (due to factors such as packet
2326        coalescing), and a "many to many" topology will reorder at a
2327        higher rate than a "many slow to one fast" configuration.
2328
2329        Many switches do not support any modes that stripe traffic
2330        (instead choosing a port based upon IP or MAC level addresses);
2331        for those devices, traffic for a particular connection flowing
2332        through the switch to a balance-rr bond will not utilize greater
2333        than one interface's worth of bandwidth.
2334
2335        If you are utilizing protocols other than TCP/IP, UDP for
2336        example, and your application can tolerate out of order
2337        delivery, then this mode can allow for single stream datagram
2338        performance that scales near linearly as interfaces are added
2339        to the bond.
2340
2341        This mode requires the switch to have the appropriate ports
2342        configured for "etherchannel" or "trunking."
2343
2344active-backup: There is not much advantage in this network topology to
2345        the active-backup mode, as the inactive backup devices are all
2346        connected to the same peer as the primary.  In this case, a
2347        load balancing mode (with link monitoring) will provide the
2348        same level of network availability, but with increased
2349        available bandwidth.  On the plus side, active-backup mode
2350        does not require any configuration of the switch, so it may
2351        have value if the hardware available does not support any of
2352        the load balance modes.
2353
2354balance-xor: This mode will limit traffic such that packets destined
2355        for specific peers will always be sent over the same
2356        interface.  Since the destination is determined by the MAC
2357        addresses involved, this mode works best in a "local" network
2358        configuration (as described above), with destinations all on
2359        the same local network.  This mode is likely to be suboptimal
2360        if all your traffic is passed through a single router (i.e., a
2361        "gatewayed" network configuration, as described above).
2362
2363        As with balance-rr, the switch ports need to be configured for
2364        "etherchannel" or "trunking."
2365
2366broadcast: Like active-backup, there is not much advantage to this
2367        mode in this type of network topology.
2368
2369802.3ad: This mode can be a good choice for this type of network
2370        topology.  The 802.3ad mode is an IEEE standard, so all peers
2371        that implement 802.3ad should interoperate well.  The 802.3ad
2372        protocol includes automatic configuration of the aggregates,
2373        so minimal manual configuration of the switch is needed
2374        (typically only to designate that some set of devices is
2375        available for 802.3ad).  The 802.3ad standard also mandates
2376        that frames be delivered in order (within certain limits), so
2377        in general single connections will not see misordering of
2378        packets.  The 802.3ad mode does have some drawbacks: the
2379        standard mandates that all devices in the aggregate operate at
2380        the same speed and duplex.  Also, as with all bonding load
2381        balance modes other than balance-rr, no single connection will
2382        be able to utilize more than a single interface's worth of
2383        bandwidth.  
2384
2385        Additionally, the linux bonding 802.3ad implementation
2386        distributes traffic by peer (using an XOR of MAC addresses
2387        and packet type ID), so in a "gatewayed" configuration, all
2388        outgoing traffic will generally use the same device.  Incoming
2389        traffic may also end up on a single device, but that is
2390        dependent upon the balancing policy of the peer's 8023.ad
2391        implementation.  In a "local" configuration, traffic will be
2392        distributed across the devices in the bond.
2393
2394        Finally, the 802.3ad mode mandates the use of the MII monitor,
2395        therefore, the ARP monitor is not available in this mode.
2396
2397balance-tlb: The balance-tlb mode balances outgoing traffic by peer.
2398        Since the balancing is done according to MAC address, in a
2399        "gatewayed" configuration (as described above), this mode will
2400        send all traffic across a single device.  However, in a
2401        "local" network configuration, this mode balances multiple
2402        local network peers across devices in a vaguely intelligent
2403        manner (not a simple XOR as in balance-xor or 802.3ad mode),
2404        so that mathematically unlucky MAC addresses (i.e., ones that
2405        XOR to the same value) will not all "bunch up" on a single
2406        interface.
2407
2408        Unlike 802.3ad, interfaces may be of differing speeds, and no
2409        special switch configuration is required.  On the down side,
2410        in this mode all incoming traffic arrives over a single
2411        interface, this mode requires certain ethtool support in the
2412        network device driver of the slave interfaces, and the ARP
2413        monitor is not available.
2414
2415balance-alb: This mode is everything that balance-tlb is, and more.
2416        It has all of the features (and restrictions) of balance-tlb,
2417        and will also balance incoming traffic from local network
2418        peers (as described in the Bonding Module Options section,
2419        above).
2420
2421        The only additional down side to this mode is that the network
2422        device driver must support changing the hardware address while
2423        the device is open.
2424
242512.1.2 MT Link Monitoring for Single Switch Topology
2426----------------------------------------------------
2427
2428        The choice of link monitoring may largely depend upon which
2429mode you choose to use.  The more advanced load balancing modes do not
2430support the use of the ARP monitor, and are thus restricted to using
2431the MII monitor (which does not provide as high a level of end to end
2432assurance as the ARP monitor).
2433
243412.2 Maximum Throughput in a Multiple Switch Topology
2435-----------------------------------------------------
2436
2437        Multiple switches may be utilized to optimize for throughput
2438when they are configured in parallel as part of an isolated network
2439between two or more systems, for example:
2440
2441                       +-----------+
2442                       |  Host A   | 
2443                       +-+---+---+-+
2444                         |   |   |
2445                +--------+   |   +---------+
2446                |            |             |
2447         +------+---+  +-----+----+  +-----+----+
2448         | Switch A |  | Switch B |  | Switch C |
2449         +------+---+  +-----+----+  +-----+----+
2450                |            |             |
2451                +--------+   |   +---------+
2452                         |   |   |
2453                       +-+---+---+-+
2454                       |  Host B   | 
2455                       +-----------+
2456
2457        In this configuration, the switches are isolated from one
2458another.  One reason to employ a topology such as this is for an
2459isolated network with many hosts (a cluster configured for high
2460performance, for example), using multiple smaller switches can be more
2461cost effective than a single larger switch, e.g., on a network with 24
2462hosts, three 24 port switches can be significantly less expensive than
2463a single 72 port switch.
2464
2465        If access beyond the network is required, an individual host
2466can be equipped with an additional network device connected to an
2467external network; this host then additionally acts as a gateway.
2468
246912.2.1 MT Bonding Mode Selection for Multiple Switch Topology
2470-------------------------------------------------------------
2471
2472        In actual practice, the bonding mode typically employed in
2473configurations of this type is balance-rr.  Historically, in this
2474network configuration, the usual caveats about out of order packet
2475delivery are mitigated by the use of network adapters that do not do
2476any kind of packet coalescing (via the use of NAPI, or because the
2477device itself does not generate interrupts until some number of
2478packets has arrived).  When employed in this fashion, the balance-rr
2479mode allows individual connections between two hosts to effectively
2480utilize greater than one interface's bandwidth.
2481
248212.2.2 MT Link Monitoring for Multiple Switch Topology
2483------------------------------------------------------
2484
2485        Again, in actual practice, the MII monitor is most often used
2486in this configuration, as performance is given preference over
2487availability.  The ARP monitor will function in this topology, but its
2488advantages over the MII monitor are mitigated by the volume of probes
2489needed as the number of systems involved grows (remember that each
2490host in the network is configured with bonding).
2491
249213. Switch Behavior Issues
2493==========================
2494
249513.1 Link Establishment and Failover Delays
2496-------------------------------------------
2497
2498        Some switches exhibit undesirable behavior with regard to the
2499timing of link up and down reporting by the switch.
2500
2501        First, when a link comes up, some switches may indicate that
2502the link is up (carrier available), but not pass traffic over the
2503interface for some period of time.  This delay is typically due to
2504some type of autonegotiation or routing protocol, but may also occur
2505during switch initialization (e.g., during recovery after a switch
2506failure).  If you find this to be a problem, specify an appropriate
2507value to the updelay bonding module option to delay the use of the
2508relevant interface(s).
2509
2510        Second, some switches may "bounce" the link state one or more
2511times while a link is changing state.  This occurs most commonly while
2512the switch is initializing.  Again, an appropriate updelay value may
2513help.
2514
2515        Note that when a bonding interface has no active links, the
2516driver will immediately reuse the first link that goes up, even if the
2517updelay parameter has been specified (the updelay is ignored in this
2518case).  If there are slave interfaces waiting for the updelay timeout
2519to expire, the interface that first went into that state will be
2520immediately reused.  This reduces down time of the network if the
2521value of updelay has been overestimated, and since this occurs only in
2522cases with no connectivity, there is no additional penalty for
2523ignoring the updelay.
2524
2525        In addition to the concerns about switch timings, if your
2526switches take a long time to go into backup mode, it may be desirable
2527to not activate a backup interface immediately after a link goes down.
2528Failover may be delayed via the downdelay bonding module option.
2529
253013.2 Duplicated Incoming Packets
2531--------------------------------
2532
2533        NOTE: Starting with version 3.0.2, the bonding driver has logic to
2534suppress duplicate packets, which should largely eliminate this problem.
2535The following description is kept for reference.
2536
2537        It is not uncommon to observe a short burst of duplicated
2538traffic when the bonding device is first used, or after it has been
2539idle for some period of time.  This is most easily observed by issuing
2540a "ping" to some other host on the network, and noticing that the
2541output from ping flags duplicates (typically one per slave).
2542
2543        For example, on a bond in active-backup mode with five slaves
2544all connected to one switch, the output may appear as follows:
2545
2546# ping -n 10.0.4.2
2547PING 10.0.4.2 (10.0.4.2) from 10.0.3.10 : 56(84) bytes of data.
254864 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.7 ms
254964 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
255064 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
255164 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
255264 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
255364 bytes from 10.0.4.2: icmp_seq=2 ttl=64 time=0.216 ms
255464 bytes from 10.0.4.2: icmp_seq=3 ttl=64 time=0.267 ms
255564 bytes from 10.0.4.2: icmp_seq=4 ttl=64 time=0.222 ms
2556
2557        This is not due to an error in the bonding driver, rather, it
2558is a side effect of how many switches update their MAC forwarding
2559tables.  Initially, the switch does not associate the MAC address in
2560the packet with a particular switch port, and so it may send the
2561traffic to all ports until its MAC forwarding table is updated.  Since
2562the interfaces attached to the bond may occupy multiple ports on a
2563single switch, when the switch (temporarily) floods the traffic to all
2564ports, the bond device receives multiple copies of the same packet
2565(one per slave device).
2566
2567        The duplicated packet behavior is switch dependent, some
2568switches exhibit this, and some do not.  On switches that display this
2569behavior, it can be induced by clearing the MAC forwarding table (on
2570most Cisco switches, the privileged command "clear mac address-table
2571dynamic" will accomplish this).
2572
257314. Hardware Specific Considerations
2574====================================
2575
2576        This section contains additional information for configuring
2577bonding on specific hardware platforms, or for interfacing bonding
2578with particular switches or other devices.
2579
258014.1 IBM BladeCenter
2581--------------------
2582
2583        This applies to the JS20 and similar systems.
2584
2585        On the JS20 blades, the bonding driver supports only
2586balance-rr, active-backup, balance-tlb and balance-alb modes.  This is
2587largely due to the network topology inside the BladeCenter, detailed
2588below.
2589
2590JS20 network adapter information
2591--------------------------------
2592
2593        All JS20s come with two Broadcom Gigabit Ethernet ports
2594integrated on the planar (that's "motherboard" in IBM-speak).  In the
2595BladeCenter chassis, the eth0 port of all JS20 blades is hard wired to
2596I/O Module #1; similarly, all eth1 ports are wired to I/O Module #2.
2597An add-on Broadcom daughter card can be installed on a JS20 to provide
2598two more Gigabit Ethernet ports.  These ports, eth2 and eth3, are
2599wired to I/O Modules 3 and 4, respectively.
2600
2601        Each I/O Module may contain either a switch or a passthrough
2602module (which allows ports to be directly connected to an external
2603switch).  Some bonding modes require a specific BladeCenter internal
2604network topology in order to function; these are detailed below.
2605
2606        Additional BladeCenter-specific networking information can be
2607found in two IBM Redbooks (www.ibm.com/redbooks):
2608
2609"IBM eServer BladeCenter Networking Options"
2610"IBM eServer BladeCenter Layer 2-7 Network Switching"
2611
2612BladeCenter networking configuration
2613------------------------------------
2614
2615        Because a BladeCenter can be configured in a very large number
2616of ways, this discussion will be confined to describing basic
2617configurations.
2618
2619        Normally, Ethernet Switch Modules (ESMs) are used in I/O
2620modules 1 and 2.  In this configuration, the eth0 and eth1 ports of a
2621JS20 will be connected to different internal switches (in the
2622respective I/O modules).
2623
2624        A passthrough module (OPM or CPM, optical or copper,
2625passthrough module) connects the I/O module directly to an external
2626switch.  By using PMs in I/O module #1 and #2, the eth0 and eth1
2627interfaces of a JS20 can be redirected to the outside world and
2628connected to a common external switch.
2629
2630        Depending upon the mix of ESMs and PMs, the network will
2631appear to bonding as either a single switch topology (all PMs) or as a
2632multiple switch topology (one or more ESMs, zero or more PMs).  It is
2633also possible to connect ESMs together, resulting in a configuration
2634much like the example in "High Availability in a Multiple Switch
2635Topology," above.
2636
2637Requirements for specific modes
2638-------------------------------
2639
2640        The balance-rr mode requires the use of passthrough modules
2641for devices in the bond, all connected to an common external switch.
2642That switch must be configured for "etherchannel" or "trunking" on the
2643appropriate ports, as is usual for balance-rr.
2644
2645        The balance-alb and balance-tlb modes will function with
2646either switch modules or passthrough modules (or a mix).  The only
2647specific requirement for these modes is that all network interfaces
2648must be able to reach all destinations for traffic sent over the
2649bonding device (i.e., the network must converge at some point outside
2650the BladeCenter).
2651
2652        The active-backup mode has no additional requirements.
2653
2654Link monitoring issues
2655----------------------
2656
2657        When an Ethernet Switch Module is in place, only the ARP
2658monitor will reliably detect link loss to an external switch.  This is
2659nothing unusual, but examination of the BladeCenter cabinet would
2660suggest that the "external" network ports are the ethernet ports for
2661the system, when it fact there is a switch between these "external"
2662ports and the devices on the JS20 system itself.  The MII monitor is
2663only able to detect link failures between the ESM and the JS20 system.
2664
2665        When a passthrough module is in place, the MII monitor does
2666detect failures to the "external" port, which is then directly
2667connected to the JS20 system.
2668
2669Other concerns
2670--------------
2671
2672        The Serial Over LAN (SoL) link is established over the primary
2673ethernet (eth0) only, therefore, any loss of link to eth0 will result
2674in losing your SoL connection.  It will not fail over with other
2675network traffic, as the SoL system is beyond the control of the
2676bonding driver.
2677
2678        It may be desirable to disable spanning tree on the switch
2679(either the internal Ethernet Switch Module, or an external switch) to
2680avoid fail-over delay issues when using bonding.
2681
2682        
268315. Frequently Asked Questions
2684==============================
2685
26861.  Is it SMP safe?
2687
2688        Yes. The old 2.0.xx channel bonding patch was not SMP safe.
2689The new driver was designed to be SMP safe from the start.
2690
26912.  What type of cards will work with it?
2692
2693        Any Ethernet type cards (you can even mix cards - a Intel
2694EtherExpress PRO/100 and a 3com 3c905b, for example).  For most modes,
2695devices need not be of the same speed.
2696
2697        Starting with version 3.2.1, bonding also supports Infiniband
2698slaves in active-backup mode.
2699
27003.  How many bonding devices can I have?
2701
2702        There is no limit.
2703
27044.  How many slaves can a bonding device have?
2705
2706        This is limited only by the number of network interfaces Linux
2707supports and/or the number of network cards you can place in your
2708system.
2709
27105.  What happens when a slave link dies?
2711
2712        If link monitoring is enabled, then the failing device will be
2713disabled.  The active-backup mode will fail over to a backup link, and
2714other modes will ignore the failed link.  The link will continue to be
2715monitored, and should it recover, it will rejoin the bond (in whatever
2716manner is appropriate for the mode). See the sections on High
2717Availability and the documentation for each mode for additional
2718information.
2719        
2720        Link monitoring can be enabled via either the miimon or
2721arp_interval parameters (described in the module parameters section,
2722above).  In general, miimon monitors the carrier state as sensed by
2723the underlying network device, and the arp monitor (arp_interval)
2724monitors connectivity to another host on the local network.
2725
2726        If no link monitoring is configured, the bonding driver will
2727be unable to detect link failures, and will assume that all links are
2728always available.  This will likely result in lost packets, and a
2729resulting degradation of performance.  The precise performance loss
2730depends upon the bonding mode and network configuration.
2731
27326.  Can bonding be used for High Availability?
2733
2734        Yes.  See the section on High Availability for details.
2735
27367.  Which switches/systems does it work with?
2737
2738        The full answer to this depends upon the desired mode.
2739
2740        In the basic balance modes (balance-rr and balance-xor), it
2741works with any system that supports etherchannel (also called
2742trunking).  Most managed switches currently available have such
2743support, and many unmanaged switches as well.
2744
2745        The advanced balance modes (balance-tlb and balance-alb) do
2746not have special switch requirements, but do need device drivers that
2747support specific features (described in the appropriate section under
2748module parameters, above).
2749
2750        In 802.3ad mode, it works with systems that support IEEE
2751802.3ad Dynamic Link Aggregation.  Most managed and many unmanaged
2752switches currently available support 802.3ad.
2753
2754        The active-backup mode should work with any Layer-II switch.
2755
27568.  Where does a bonding device get its MAC address from?
2757
2758        When using slave devices that have fixed MAC addresses, or when
2759the fail_over_mac option is enabled, the bonding device's MAC address is
2760the MAC address of the active slave.
2761
2762        For other configurations, if not explicitly configured (with
2763ifconfig or ip link), the MAC address of the bonding device is taken from
2764its first slave device.  This MAC address is then passed to all following
2765slaves and remains persistent (even if the first slave is removed) until
2766the bonding device is brought down or reconfigured.
2767
2768        If you wish to change the MAC address, you can set it with
2769ifconfig or ip link:
2770
2771# ifconfig bond0 hw ether 00:11:22:33:44:55
2772
2773# ip link set bond0 address 66:77:88:99:aa:bb
2774
2775        The MAC address can be also changed by bringing down/up the
2776device and then changing its slaves (or their order):
2777
2778# ifconfig bond0 down ; modprobe -r bonding
2779# ifconfig bond0 .... up
2780# ifenslave bond0 eth...
2781
2782        This method will automatically take the address from the next
2783slave that is added.
2784
2785        To restore your slaves' MAC addresses, you need to detach them
2786from the bond (`ifenslave -d bond0 eth0'). The bonding driver will
2787then restore the MAC addresses that the slaves had before they were
2788enslaved.
2789
279016. Resources and Links
2791=======================
2792
2793        The latest version of the bonding driver can be found in the latest
2794version of the linux kernel, found on http://kernel.org
2795
2796        The latest version of this document can be found in the latest kernel
2797source (named Documentation/networking/bonding.txt).
2798
2799        Discussions regarding the usage of the bonding driver take place on the
2800bonding-devel mailing list, hosted at sourceforge.net. If you have questions or
2801problems, post them to the list.  The list address is:
2802
2803bonding-devel@lists.sourceforge.net
2804
2805        The administrative interface (to subscribe or unsubscribe) can
2806be found at:
2807
2808https://lists.sourceforge.net/lists/listinfo/bonding-devel
2809
2810        Discussions regarding the development of the bonding driver take place
2811on the main Linux network mailing list, hosted at vger.kernel.org. The list
2812address is:
2813
2814netdev@vger.kernel.org
2815
2816        The administrative interface (to subscribe or unsubscribe) can
2817be found at:
2818
2819http://vger.kernel.org/vger-lists.html#netdev
2820
2821Donald Becker's Ethernet Drivers and diag programs may be found at :
2822 - http://web.archive.org/web/*/http://www.scyld.com/network/ 
2823
2824You will also find a lot of information regarding Ethernet, NWay, MII,
2825etc. at www.scyld.com.
2826
2827-- END --
2828
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.