bonding.txt 88 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993994995996997998999100010011002100310041005100610071008100910101011101210131014101510161017101810191020102110221023102410251026102710281029103010311032103310341035103610371038103910401041104210431044104510461047104810491050105110521053105410551056105710581059106010611062106310641065106610671068106910701071107210731074107510761077107810791080108110821083108410851086108710881089109010911092109310941095109610971098109911001101110211031104110511061107110811091110111111121113111411151116111711181119112011211122112311241125112611271128112911301131113211331134113511361137113811391140114111421143114411451146114711481149115011511152115311541155115611571158115911601161116211631164116511661167116811691170117111721173117411751176117711781179118011811182118311841185118611871188118911901191119211931194119511961197119811991200120112021203120412051206120712081209121012111212121312141215121612171218121912201221122212231224122512261227122812291230123112321233123412351236123712381239124012411242124312441245124612471248124912501251125212531254125512561257125812591260126112621263126412651266126712681269127012711272127312741275127612771278127912801281128212831284128512861287128812891290129112921293129412951296129712981299130013011302130313041305130613071308130913101311131213131314131513161317131813191320132113221323132413251326132713281329133013311332133313341335133613371338133913401341134213431344134513461347134813491350135113521353135413551356135713581359136013611362136313641365136613671368136913701371137213731374137513761377137813791380138113821383138413851386138713881389139013911392139313941395139613971398139914001401140214031404140514061407140814091410141114121413141414151416141714181419142014211422142314241425142614271428142914301431143214331434143514361437143814391440144114421443144414451446144714481449145014511452145314541455145614571458145914601461146214631464146514661467146814691470147114721473147414751476147714781479148014811482148314841485148614871488148914901491149214931494149514961497149814991500150115021503150415051506150715081509151015111512151315141515151615171518151915201521152215231524152515261527152815291530153115321533153415351536153715381539154015411542154315441545154615471548154915501551155215531554155515561557155815591560156115621563156415651566156715681569157015711572157315741575157615771578157915801581158215831584158515861587158815891590159115921593159415951596159715981599160016011602160316041605160616071608160916101611161216131614161516161617161816191620162116221623162416251626162716281629163016311632163316341635163616371638163916401641164216431644164516461647164816491650165116521653165416551656165716581659166016611662166316641665166616671668166916701671167216731674167516761677167816791680168116821683168416851686168716881689169016911692169316941695169616971698169917001701170217031704170517061707170817091710171117121713171417151716171717181719172017211722172317241725172617271728172917301731173217331734173517361737173817391740174117421743174417451746174717481749175017511752175317541755175617571758175917601761176217631764176517661767176817691770177117721773177417751776177717781779178017811782178317841785178617871788178917901791179217931794179517961797179817991800180118021803180418051806180718081809181018111812181318141815181618171818181918201821182218231824182518261827182818291830183118321833183418351836183718381839184018411842184318441845184618471848184918501851185218531854185518561857185818591860186118621863186418651866186718681869187018711872187318741875187618771878187918801881188218831884188518861887188818891890189118921893189418951896189718981899190019011902190319041905190619071908190919101911191219131914191519161917191819191920192119221923192419251926192719281929193019311932193319341935193619371938193919401941194219431944194519461947194819491950195119521953195419551956195719581959196019611962196319641965196619671968196919701971197219731974197519761977197819791980198119821983198419851986198719881989199019911992199319941995199619971998199920002001200220032004200520062007200820092010201120122013201420152016201720182019202020212022202320242025202620272028202920302031203220332034203520362037203820392040204120422043204420452046204720482049205020512052205320542055205620572058205920602061206220632064206520662067206820692070207120722073207420752076207720782079208020812082208320842085208620872088208920902091209220932094209520962097209820992100210121022103210421052106210721082109211021112112211321142115211621172118211921202121212221232124212521262127212821292130213121322133213421352136213721382139214021412142214321442145214621472148214921502151215221532154215521562157215821592160216121622163216421652166216721682169217021712172217321742175217621772178217921802181218221832184218521862187
  1. Linux Ethernet Bonding Driver HOWTO
  2. Latest update: 24 April 2006
  3. Initial release : Thomas Davis <tadavis at lbl.gov>
  4. Corrections, HA extensions : 2000/10/03-15 :
  5. - Willy Tarreau <willy at meta-x.org>
  6. - Constantine Gavrilov <const-g at xpert.com>
  7. - Chad N. Tindel <ctindel at ieee dot org>
  8. - Janice Girouard <girouard at us dot ibm dot com>
  9. - Jay Vosburgh <fubar at us dot ibm dot com>
  10. Reorganized and updated Feb 2005 by Jay Vosburgh
  11. Added Sysfs information: 2006/04/24
  12. - Mitch Williams <mitch.a.williams at intel.com>
  13. Introduction
  14. ============
  15. The Linux bonding driver provides a method for aggregating
  16. multiple network interfaces into a single logical "bonded" interface.
  17. The behavior of the bonded interfaces depends upon the mode; generally
  18. speaking, modes provide either hot standby or load balancing services.
  19. Additionally, link integrity monitoring may be performed.
  20. The bonding driver originally came from Donald Becker's
  21. beowulf patches for kernel 2.0. It has changed quite a bit since, and
  22. the original tools from extreme-linux and beowulf sites will not work
  23. with this version of the driver.
  24. For new versions of the driver, updated userspace tools, and
  25. who to ask for help, please follow the links at the end of this file.
  26. Table of Contents
  27. =================
  28. 1. Bonding Driver Installation
  29. 2. Bonding Driver Options
  30. 3. Configuring Bonding Devices
  31. 3.1 Configuration with Sysconfig Support
  32. 3.1.1 Using DHCP with Sysconfig
  33. 3.1.2 Configuring Multiple Bonds with Sysconfig
  34. 3.2 Configuration with Initscripts Support
  35. 3.2.1 Using DHCP with Initscripts
  36. 3.2.2 Configuring Multiple Bonds with Initscripts
  37. 3.3 Configuring Bonding Manually with Ifenslave
  38. 3.3.1 Configuring Multiple Bonds Manually
  39. 3.4 Configuring Bonding Manually via Sysfs
  40. 4. Querying Bonding Configuration
  41. 4.1 Bonding Configuration
  42. 4.2 Network Configuration
  43. 5. Switch Configuration
  44. 6. 802.1q VLAN Support
  45. 7. Link Monitoring
  46. 7.1 ARP Monitor Operation
  47. 7.2 Configuring Multiple ARP Targets
  48. 7.3 MII Monitor Operation
  49. 8. Potential Trouble Sources
  50. 8.1 Adventures in Routing
  51. 8.2 Ethernet Device Renaming
  52. 8.3 Painfully Slow Or No Failed Link Detection By Miimon
  53. 9. SNMP agents
  54. 10. Promiscuous mode
  55. 11. Configuring Bonding for High Availability
  56. 11.1 High Availability in a Single Switch Topology
  57. 11.2 High Availability in a Multiple Switch Topology
  58. 11.2.1 HA Bonding Mode Selection for Multiple Switch Topology
  59. 11.2.2 HA Link Monitoring for Multiple Switch Topology
  60. 12. Configuring Bonding for Maximum Throughput
  61. 12.1 Maximum Throughput in a Single Switch Topology
  62. 12.1.1 MT Bonding Mode Selection for Single Switch Topology
  63. 12.1.2 MT Link Monitoring for Single Switch Topology
  64. 12.2 Maximum Throughput in a Multiple Switch Topology
  65. 12.2.1 MT Bonding Mode Selection for Multiple Switch Topology
  66. 12.2.2 MT Link Monitoring for Multiple Switch Topology
  67. 13. Switch Behavior Issues
  68. 13.1 Link Establishment and Failover Delays
  69. 13.2 Duplicated Incoming Packets
  70. 14. Hardware Specific Considerations
  71. 14.1 IBM BladeCenter
  72. 15. Frequently Asked Questions
  73. 16. Resources and Links
  74. 1. Bonding Driver Installation
  75. ==============================
  76. Most popular distro kernels ship with the bonding driver
  77. already available as a module and the ifenslave user level control
  78. program installed and ready for use. If your distro does not, or you
  79. have need to compile bonding from source (e.g., configuring and
  80. installing a mainline kernel from kernel.org), you'll need to perform
  81. the following steps:
  82. 1.1 Configure and build the kernel with bonding
  83. -----------------------------------------------
  84. The current version of the bonding driver is available in the
  85. drivers/net/bonding subdirectory of the most recent kernel source
  86. (which is available on http://kernel.org). Most users "rolling their
  87. own" will want to use the most recent kernel from kernel.org.
  88. Configure kernel with "make menuconfig" (or "make xconfig" or
  89. "make config"), then select "Bonding driver support" in the "Network
  90. device support" section. It is recommended that you configure the
  91. driver as module since it is currently the only way to pass parameters
  92. to the driver or configure more than one bonding device.
  93. Build and install the new kernel and modules, then continue
  94. below to install ifenslave.
  95. 1.2 Install ifenslave Control Utility
  96. -------------------------------------
  97. The ifenslave user level control program is included in the
  98. kernel source tree, in the file Documentation/networking/ifenslave.c.
  99. It is generally recommended that you use the ifenslave that
  100. corresponds to the kernel that you are using (either from the same
  101. source tree or supplied with the distro), however, ifenslave
  102. executables from older kernels should function (but features newer
  103. than the ifenslave release are not supported). Running an ifenslave
  104. that is newer than the kernel is not supported, and may or may not
  105. work.
  106. To install ifenslave, do the following:
  107. # gcc -Wall -O -I/usr/src/linux/include ifenslave.c -o ifenslave
  108. # cp ifenslave /sbin/ifenslave
  109. If your kernel source is not in "/usr/src/linux," then replace
  110. "/usr/src/linux/include" in the above with the location of your kernel
  111. source include directory.
  112. You may wish to back up any existing /sbin/ifenslave, or, for
  113. testing or informal use, tag the ifenslave to the kernel version
  114. (e.g., name the ifenslave executable /sbin/ifenslave-2.6.10).
  115. IMPORTANT NOTE:
  116. If you omit the "-I" or specify an incorrect directory, you
  117. may end up with an ifenslave that is incompatible with the kernel
  118. you're trying to build it for. Some distros (e.g., Red Hat from 7.1
  119. onwards) do not have /usr/include/linux symbolically linked to the
  120. default kernel source include directory.
  121. SECOND IMPORTANT NOTE:
  122. If you plan to configure bonding using sysfs, you do not need
  123. to use ifenslave.
  124. 2. Bonding Driver Options
  125. =========================
  126. Options for the bonding driver are supplied as parameters to
  127. the bonding module at load time. They may be given as command line
  128. arguments to the insmod or modprobe command, but are usually specified
  129. in either the /etc/modules.conf or /etc/modprobe.conf configuration
  130. file, or in a distro-specific configuration file (some of which are
  131. detailed in the next section).
  132. The available bonding driver parameters are listed below. If a
  133. parameter is not specified the default value is used. When initially
  134. configuring a bond, it is recommended "tail -f /var/log/messages" be
  135. run in a separate window to watch for bonding driver error messages.
  136. It is critical that either the miimon or arp_interval and
  137. arp_ip_target parameters be specified, otherwise serious network
  138. degradation will occur during link failures. Very few devices do not
  139. support at least miimon, so there is really no reason not to use it.
  140. Options with textual values will accept either the text name
  141. or, for backwards compatibility, the option value. E.g.,
  142. "mode=802.3ad" and "mode=4" set the same mode.
  143. The parameters are as follows:
  144. arp_interval
  145. Specifies the ARP link monitoring frequency in milliseconds.
  146. The ARP monitor works by periodically checking the slave
  147. devices to determine whether they have sent or received
  148. traffic recently (the precise criteria depends upon the
  149. bonding mode, and the state of the slave). Regular traffic is
  150. generated via ARP probes issued for the addresses specified by
  151. the arp_ip_target option.
  152. This behavior can be modified by the arp_validate option,
  153. below.
  154. If ARP monitoring is used in an etherchannel compatible mode
  155. (modes 0 and 2), the switch should be configured in a mode
  156. that evenly distributes packets across all links. If the
  157. switch is configured to distribute the packets in an XOR
  158. fashion, all replies from the ARP targets will be received on
  159. the same link which could cause the other team members to
  160. fail. ARP monitoring should not be used in conjunction with
  161. miimon. A value of 0 disables ARP monitoring. The default
  162. value is 0.
  163. arp_ip_target
  164. Specifies the IP addresses to use as ARP monitoring peers when
  165. arp_interval is > 0. These are the targets of the ARP request
  166. sent to determine the health of the link to the targets.
  167. Specify these values in ddd.ddd.ddd.ddd format. Multiple IP
  168. addresses must be separated by a comma. At least one IP
  169. address must be given for ARP monitoring to function. The
  170. maximum number of targets that can be specified is 16. The
  171. default value is no IP addresses.
  172. arp_validate
  173. Specifies whether or not ARP probes and replies should be
  174. validated in the active-backup mode. This causes the ARP
  175. monitor to examine the incoming ARP requests and replies, and
  176. only consider a slave to be up if it is receiving the
  177. appropriate ARP traffic.
  178. Possible values are:
  179. none or 0
  180. No validation is performed. This is the default.
  181. active or 1
  182. Validation is performed only for the active slave.
  183. backup or 2
  184. Validation is performed only for backup slaves.
  185. all or 3
  186. Validation is performed for all slaves.
  187. For the active slave, the validation checks ARP replies to
  188. confirm that they were generated by an arp_ip_target. Since
  189. backup slaves do not typically receive these replies, the
  190. validation performed for backup slaves is on the ARP request
  191. sent out via the active slave. It is possible that some
  192. switch or network configurations may result in situations
  193. wherein the backup slaves do not receive the ARP requests; in
  194. such a situation, validation of backup slaves must be
  195. disabled.
  196. This option is useful in network configurations in which
  197. multiple bonding hosts are concurrently issuing ARPs to one or
  198. more targets beyond a common switch. Should the link between
  199. the switch and target fail (but not the switch itself), the
  200. probe traffic generated by the multiple bonding instances will
  201. fool the standard ARP monitor into considering the links as
  202. still up. Use of the arp_validate option can resolve this, as
  203. the ARP monitor will only consider ARP requests and replies
  204. associated with its own instance of bonding.
  205. This option was added in bonding version 3.1.0.
  206. downdelay
  207. Specifies the time, in milliseconds, to wait before disabling
  208. a slave after a link failure has been detected. This option
  209. is only valid for the miimon link monitor. The downdelay
  210. value should be a multiple of the miimon value; if not, it
  211. will be rounded down to the nearest multiple. The default
  212. value is 0.
  213. lacp_rate
  214. Option specifying the rate in which we'll ask our link partner
  215. to transmit LACPDU packets in 802.3ad mode. Possible values
  216. are:
  217. slow or 0
  218. Request partner to transmit LACPDUs every 30 seconds
  219. fast or 1
  220. Request partner to transmit LACPDUs every 1 second
  221. The default is slow.
  222. max_bonds
  223. Specifies the number of bonding devices to create for this
  224. instance of the bonding driver. E.g., if max_bonds is 3, and
  225. the bonding driver is not already loaded, then bond0, bond1
  226. and bond2 will be created. The default value is 1.
  227. miimon
  228. Specifies the MII link monitoring frequency in milliseconds.
  229. This determines how often the link state of each slave is
  230. inspected for link failures. A value of zero disables MII
  231. link monitoring. A value of 100 is a good starting point.
  232. The use_carrier option, below, affects how the link state is
  233. determined. See the High Availability section for additional
  234. information. The default value is 0.
  235. mode
  236. Specifies one of the bonding policies. The default is
  237. balance-rr (round robin). Possible values are:
  238. balance-rr or 0
  239. Round-robin policy: Transmit packets in sequential
  240. order from the first available slave through the
  241. last. This mode provides load balancing and fault
  242. tolerance.
  243. active-backup or 1
  244. Active-backup policy: Only one slave in the bond is
  245. active. A different slave becomes active if, and only
  246. if, the active slave fails. The bond's MAC address is
  247. externally visible on only one port (network adapter)
  248. to avoid confusing the switch.
  249. In bonding version 2.6.2 or later, when a failover
  250. occurs in active-backup mode, bonding will issue one
  251. or more gratuitous ARPs on the newly active slave.
  252. One gratuitous ARP is issued for the bonding master
  253. interface and each VLAN interfaces configured above
  254. it, provided that the interface has at least one IP
  255. address configured. Gratuitous ARPs issued for VLAN
  256. interfaces are tagged with the appropriate VLAN id.
  257. This mode provides fault tolerance. The primary
  258. option, documented below, affects the behavior of this
  259. mode.
  260. balance-xor or 2
  261. XOR policy: Transmit based on the selected transmit
  262. hash policy. The default policy is a simple [(source
  263. MAC address XOR'd with destination MAC address) modulo
  264. slave count]. Alternate transmit policies may be
  265. selected via the xmit_hash_policy option, described
  266. below.
  267. This mode provides load balancing and fault tolerance.
  268. broadcast or 3
  269. Broadcast policy: transmits everything on all slave
  270. interfaces. This mode provides fault tolerance.
  271. 802.3ad or 4
  272. IEEE 802.3ad Dynamic link aggregation. Creates
  273. aggregation groups that share the same speed and
  274. duplex settings. Utilizes all slaves in the active
  275. aggregator according to the 802.3ad specification.
  276. Slave selection for outgoing traffic is done according
  277. to the transmit hash policy, which may be changed from
  278. the default simple XOR policy via the xmit_hash_policy
  279. option, documented below. Note that not all transmit
  280. policies may be 802.3ad compliant, particularly in
  281. regards to the packet mis-ordering requirements of
  282. section 43.2.4 of the 802.3ad standard. Differing
  283. peer implementations will have varying tolerances for
  284. noncompliance.
  285. Prerequisites:
  286. 1. Ethtool support in the base drivers for retrieving
  287. the speed and duplex of each slave.
  288. 2. A switch that supports IEEE 802.3ad Dynamic link
  289. aggregation.
  290. Most switches will require some type of configuration
  291. to enable 802.3ad mode.
  292. balance-tlb or 5
  293. Adaptive transmit load balancing: channel bonding that
  294. does not require any special switch support. The
  295. outgoing traffic is distributed according to the
  296. current load (computed relative to the speed) on each
  297. slave. Incoming traffic is received by the current
  298. slave. If the receiving slave fails, another slave
  299. takes over the MAC address of the failed receiving
  300. slave.
  301. Prerequisite:
  302. Ethtool support in the base drivers for retrieving the
  303. speed of each slave.
  304. balance-alb or 6
  305. Adaptive load balancing: includes balance-tlb plus
  306. receive load balancing (rlb) for IPV4 traffic, and
  307. does not require any special switch support. The
  308. receive load balancing is achieved by ARP negotiation.
  309. The bonding driver intercepts the ARP Replies sent by
  310. the local system on their way out and overwrites the
  311. source hardware address with the unique hardware
  312. address of one of the slaves in the bond such that
  313. different peers use different hardware addresses for
  314. the server.
  315. Receive traffic from connections created by the server
  316. is also balanced. When the local system sends an ARP
  317. Request the bonding driver copies and saves the peer's
  318. IP information from the ARP packet. When the ARP
  319. Reply arrives from the peer, its hardware address is
  320. retrieved and the bonding driver initiates an ARP
  321. reply to this peer assigning it to one of the slaves
  322. in the bond. A problematic outcome of using ARP
  323. negotiation for balancing is that each time that an
  324. ARP request is broadcast it uses the hardware address
  325. of the bond. Hence, peers learn the hardware address
  326. of the bond and the balancing of receive traffic
  327. collapses to the current slave. This is handled by
  328. sending updates (ARP Replies) to all the peers with
  329. their individually assigned hardware address such that
  330. the traffic is redistributed. Receive traffic is also
  331. redistributed when a new slave is added to the bond
  332. and when an inactive slave is re-activated. The
  333. receive load is distributed sequentially (round robin)
  334. among the group of highest speed slaves in the bond.
  335. When a link is reconnected or a new slave joins the
  336. bond the receive traffic is redistributed among all
  337. active slaves in the bond by initiating ARP Replies
  338. with the selected MAC address to each of the
  339. clients. The updelay parameter (detailed below) must
  340. be set to a value equal or greater than the switch's
  341. forwarding delay so that the ARP Replies sent to the
  342. peers will not be blocked by the switch.
  343. Prerequisites:
  344. 1. Ethtool support in the base drivers for retrieving
  345. the speed of each slave.
  346. 2. Base driver support for setting the hardware
  347. address of a device while it is open. This is
  348. required so that there will always be one slave in the
  349. team using the bond hardware address (the
  350. curr_active_slave) while having a unique hardware
  351. address for each slave in the bond. If the
  352. curr_active_slave fails its hardware address is
  353. swapped with the new curr_active_slave that was
  354. chosen.
  355. primary
  356. A string (eth0, eth2, etc) specifying which slave is the
  357. primary device. The specified device will always be the
  358. active slave while it is available. Only when the primary is
  359. off-line will alternate devices be used. This is useful when
  360. one slave is preferred over another, e.g., when one slave has
  361. higher throughput than another.
  362. The primary option is only valid for active-backup mode.
  363. updelay
  364. Specifies the time, in milliseconds, to wait before enabling a
  365. slave after a link recovery has been detected. This option is
  366. only valid for the miimon link monitor. The updelay value
  367. should be a multiple of the miimon value; if not, it will be
  368. rounded down to the nearest multiple. The default value is 0.
  369. use_carrier
  370. Specifies whether or not miimon should use MII or ETHTOOL
  371. ioctls vs. netif_carrier_ok() to determine the link
  372. status. The MII or ETHTOOL ioctls are less efficient and
  373. utilize a deprecated calling sequence within the kernel. The
  374. netif_carrier_ok() relies on the device driver to maintain its
  375. state with netif_carrier_on/off; at this writing, most, but
  376. not all, device drivers support this facility.
  377. If bonding insists that the link is up when it should not be,
  378. it may be that your network device driver does not support
  379. netif_carrier_on/off. The default state for netif_carrier is
  380. "carrier on," so if a driver does not support netif_carrier,
  381. it will appear as if the link is always up. In this case,
  382. setting use_carrier to 0 will cause bonding to revert to the
  383. MII / ETHTOOL ioctl method to determine the link state.
  384. A value of 1 enables the use of netif_carrier_ok(), a value of
  385. 0 will use the deprecated MII / ETHTOOL ioctls. The default
  386. value is 1.
  387. xmit_hash_policy
  388. Selects the transmit hash policy to use for slave selection in
  389. balance-xor and 802.3ad modes. Possible values are:
  390. layer2
  391. Uses XOR of hardware MAC addresses to generate the
  392. hash. The formula is
  393. (source MAC XOR destination MAC) modulo slave count
  394. This algorithm will place all traffic to a particular
  395. network peer on the same slave.
  396. This algorithm is 802.3ad compliant.
  397. layer3+4
  398. This policy uses upper layer protocol information,
  399. when available, to generate the hash. This allows for
  400. traffic to a particular network peer to span multiple
  401. slaves, although a single connection will not span
  402. multiple slaves.
  403. The formula for unfragmented TCP and UDP packets is
  404. ((source port XOR dest port) XOR
  405. ((source IP XOR dest IP) AND 0xffff)
  406. modulo slave count
  407. For fragmented TCP or UDP packets and all other IP
  408. protocol traffic, the source and destination port
  409. information is omitted. For non-IP traffic, the
  410. formula is the same as for the layer2 transmit hash
  411. policy.
  412. This policy is intended to mimic the behavior of
  413. certain switches, notably Cisco switches with PFC2 as
  414. well as some Foundry and IBM products.
  415. This algorithm is not fully 802.3ad compliant. A
  416. single TCP or UDP conversation containing both
  417. fragmented and unfragmented packets will see packets
  418. striped across two interfaces. This may result in out
  419. of order delivery. Most traffic types will not meet
  420. this criteria, as TCP rarely fragments traffic, and
  421. most UDP traffic is not involved in extended
  422. conversations. Other implementations of 802.3ad may
  423. or may not tolerate this noncompliance.
  424. The default value is layer2. This option was added in bonding
  425. version 2.6.3. In earlier versions of bonding, this parameter does
  426. not exist, and the layer2 policy is the only policy.
  427. 3. Configuring Bonding Devices
  428. ==============================
  429. You can configure bonding using either your distro's network
  430. initialization scripts, or manually using either ifenslave or the
  431. sysfs interface. Distros generally use one of two packages for the
  432. network initialization scripts: initscripts or sysconfig. Recent
  433. versions of these packages have support for bonding, while older
  434. versions do not.
  435. We will first describe the options for configuring bonding for
  436. distros using versions of initscripts and sysconfig with full or
  437. partial support for bonding, then provide information on enabling
  438. bonding without support from the network initialization scripts (i.e.,
  439. older versions of initscripts or sysconfig).
  440. If you're unsure whether your distro uses sysconfig or
  441. initscripts, or don't know if it's new enough, have no fear.
  442. Determining this is fairly straightforward.
  443. First, issue the command:
  444. $ rpm -qf /sbin/ifup
  445. It will respond with a line of text starting with either
  446. "initscripts" or "sysconfig," followed by some numbers. This is the
  447. package that provides your network initialization scripts.
  448. Next, to determine if your installation supports bonding,
  449. issue the command:
  450. $ grep ifenslave /sbin/ifup
  451. If this returns any matches, then your initscripts or
  452. sysconfig has support for bonding.
  453. 3.1 Configuration with Sysconfig Support
  454. ----------------------------------------
  455. This section applies to distros using a version of sysconfig
  456. with bonding support, for example, SuSE Linux Enterprise Server 9.
  457. SuSE SLES 9's networking configuration system does support
  458. bonding, however, at this writing, the YaST system configuration
  459. front end does not provide any means to work with bonding devices.
  460. Bonding devices can be managed by hand, however, as follows.
  461. First, if they have not already been configured, configure the
  462. slave devices. On SLES 9, this is most easily done by running the
  463. yast2 sysconfig configuration utility. The goal is for to create an
  464. ifcfg-id file for each slave device. The simplest way to accomplish
  465. this is to configure the devices for DHCP (this is only to get the
  466. file ifcfg-id file created; see below for some issues with DHCP). The
  467. name of the configuration file for each device will be of the form:
  468. ifcfg-id-xx:xx:xx:xx:xx:xx
  469. Where the "xx" portion will be replaced with the digits from
  470. the device's permanent MAC address.
  471. Once the set of ifcfg-id-xx:xx:xx:xx:xx:xx files has been
  472. created, it is necessary to edit the configuration files for the slave
  473. devices (the MAC addresses correspond to those of the slave devices).
  474. Before editing, the file will contain multiple lines, and will look
  475. something like this:
  476. BOOTPROTO='dhcp'
  477. STARTMODE='on'
  478. USERCTL='no'
  479. UNIQUE='XNzu.WeZGOGF+4wE'
  480. _nm_name='bus-pci-0001:61:01.0'
  481. Change the BOOTPROTO and STARTMODE lines to the following:
  482. BOOTPROTO='none'
  483. STARTMODE='off'
  484. Do not alter the UNIQUE or _nm_name lines. Remove any other
  485. lines (USERCTL, etc).
  486. Once the ifcfg-id-xx:xx:xx:xx:xx:xx files have been modified,
  487. it's time to create the configuration file for the bonding device
  488. itself. This file is named ifcfg-bondX, where X is the number of the
  489. bonding device to create, starting at 0. The first such file is
  490. ifcfg-bond0, the second is ifcfg-bond1, and so on. The sysconfig
  491. network configuration system will correctly start multiple instances
  492. of bonding.
  493. The contents of the ifcfg-bondX file is as follows:
  494. BOOTPROTO="static"
  495. BROADCAST="10.0.2.255"
  496. IPADDR="10.0.2.10"
  497. NETMASK="255.255.0.0"
  498. NETWORK="10.0.2.0"
  499. REMOTE_IPADDR=""
  500. STARTMODE="onboot"
  501. BONDING_MASTER="yes"
  502. BONDING_MODULE_OPTS="mode=active-backup miimon=100"
  503. BONDING_SLAVE0="eth0"
  504. BONDING_SLAVE1="bus-pci-0000:06:08.1"
  505. Replace the sample BROADCAST, IPADDR, NETMASK and NETWORK
  506. values with the appropriate values for your network.
  507. The STARTMODE specifies when the device is brought online.
  508. The possible values are:
  509. onboot: The device is started at boot time. If you're not
  510. sure, this is probably what you want.
  511. manual: The device is started only when ifup is called
  512. manually. Bonding devices may be configured this
  513. way if you do not wish them to start automatically
  514. at boot for some reason.
  515. hotplug: The device is started by a hotplug event. This is not
  516. a valid choice for a bonding device.
  517. off or ignore: The device configuration is ignored.
  518. The line BONDING_MASTER='yes' indicates that the device is a
  519. bonding master device. The only useful value is "yes."
  520. The contents of BONDING_MODULE_OPTS are supplied to the
  521. instance of the bonding module for this device. Specify the options
  522. for the bonding mode, link monitoring, and so on here. Do not include
  523. the max_bonds bonding parameter; this will confuse the configuration
  524. system if you have multiple bonding devices.
  525. Finally, supply one BONDING_SLAVEn="slave device" for each
  526. slave. where "n" is an increasing value, one for each slave. The
  527. "slave device" is either an interface name, e.g., "eth0", or a device
  528. specifier for the network device. The interface name is easier to
  529. find, but the ethN names are subject to change at boot time if, e.g.,
  530. a device early in the sequence has failed. The device specifiers
  531. (bus-pci-0000:06:08.1 in the example above) specify the physical
  532. network device, and will not change unless the device's bus location
  533. changes (for example, it is moved from one PCI slot to another). The
  534. example above uses one of each type for demonstration purposes; most
  535. configurations will choose one or the other for all slave devices.
  536. When all configuration files have been modified or created,
  537. networking must be restarted for the configuration changes to take
  538. effect. This can be accomplished via the following:
  539. # /etc/init.d/network restart
  540. Note that the network control script (/sbin/ifdown) will
  541. remove the bonding module as part of the network shutdown processing,
  542. so it is not necessary to remove the module by hand if, e.g., the
  543. module parameters have changed.
  544. Also, at this writing, YaST/YaST2 will not manage bonding
  545. devices (they do not show bonding interfaces on its list of network
  546. devices). It is necessary to edit the configuration file by hand to
  547. change the bonding configuration.
  548. Additional general options and details of the ifcfg file
  549. format can be found in an example ifcfg template file:
  550. /etc/sysconfig/network/ifcfg.template
  551. Note that the template does not document the various BONDING_
  552. settings described above, but does describe many of the other options.
  553. 3.1.1 Using DHCP with Sysconfig
  554. -------------------------------
  555. Under sysconfig, configuring a device with BOOTPROTO='dhcp'
  556. will cause it to query DHCP for its IP address information. At this
  557. writing, this does not function for bonding devices; the scripts
  558. attempt to obtain the device address from DHCP prior to adding any of
  559. the slave devices. Without active slaves, the DHCP requests are not
  560. sent to the network.
  561. 3.1.2 Configuring Multiple Bonds with Sysconfig
  562. -----------------------------------------------
  563. The sysconfig network initialization system is capable of
  564. handling multiple bonding devices. All that is necessary is for each
  565. bonding instance to have an appropriately configured ifcfg-bondX file
  566. (as described above). Do not specify the "max_bonds" parameter to any
  567. instance of bonding, as this will confuse sysconfig. If you require
  568. multiple bonding devices with identical parameters, create multiple
  569. ifcfg-bondX files.
  570. Because the sysconfig scripts supply the bonding module
  571. options in the ifcfg-bondX file, it is not necessary to add them to
  572. the system /etc/modules.conf or /etc/modprobe.conf configuration file.
  573. 3.2 Configuration with Initscripts Support
  574. ------------------------------------------
  575. This section applies to distros using a version of initscripts
  576. with bonding support, for example, Red Hat Linux 9 or Red Hat
  577. Enterprise Linux version 3 or 4. On these systems, the network
  578. initialization scripts have some knowledge of bonding, and can be
  579. configured to control bonding devices.
  580. These distros will not automatically load the network adapter
  581. driver unless the ethX device is configured with an IP address.
  582. Because of this constraint, users must manually configure a
  583. network-script file for all physical adapters that will be members of
  584. a bondX link. Network script files are located in the directory:
  585. /etc/sysconfig/network-scripts
  586. The file name must be prefixed with "ifcfg-eth" and suffixed
  587. with the adapter's physical adapter number. For example, the script
  588. for eth0 would be named /etc/sysconfig/network-scripts/ifcfg-eth0.
  589. Place the following text in the file:
  590. DEVICE=eth0
  591. USERCTL=no
  592. ONBOOT=yes
  593. MASTER=bond0
  594. SLAVE=yes
  595. BOOTPROTO=none
  596. The DEVICE= line will be different for every ethX device and
  597. must correspond with the name of the file, i.e., ifcfg-eth1 must have
  598. a device line of DEVICE=eth1. The setting of the MASTER= line will
  599. also depend on the final bonding interface name chosen for your bond.
  600. As with other network devices, these typically start at 0, and go up
  601. one for each device, i.e., the first bonding instance is bond0, the
  602. second is bond1, and so on.
  603. Next, create a bond network script. The file name for this
  604. script will be /etc/sysconfig/network-scripts/ifcfg-bondX where X is
  605. the number of the bond. For bond0 the file is named "ifcfg-bond0",
  606. for bond1 it is named "ifcfg-bond1", and so on. Within that file,
  607. place the following text:
  608. DEVICE=bond0
  609. IPADDR=192.168.1.1
  610. NETMASK=255.255.255.0
  611. NETWORK=192.168.1.0
  612. BROADCAST=192.168.1.255
  613. ONBOOT=yes
  614. BOOTPROTO=none
  615. USERCTL=no
  616. Be sure to change the networking specific lines (IPADDR,
  617. NETMASK, NETWORK and BROADCAST) to match your network configuration.
  618. Finally, it is necessary to edit /etc/modules.conf (or
  619. /etc/modprobe.conf, depending upon your distro) to load the bonding
  620. module with your desired options when the bond0 interface is brought
  621. up. The following lines in /etc/modules.conf (or modprobe.conf) will
  622. load the bonding module, and select its options:
  623. alias bond0 bonding
  624. options bond0 mode=balance-alb miimon=100
  625. Replace the sample parameters with the appropriate set of
  626. options for your configuration.
  627. Finally run "/etc/rc.d/init.d/network restart" as root. This
  628. will restart the networking subsystem and your bond link should be now
  629. up and running.
  630. 3.2.1 Using DHCP with Initscripts
  631. ---------------------------------
  632. Recent versions of initscripts (the version supplied with
  633. Fedora Core 3 and Red Hat Enterprise Linux 4 is reported to work) do
  634. have support for assigning IP information to bonding devices via DHCP.
  635. To configure bonding for DHCP, configure it as described
  636. above, except replace the line "BOOTPROTO=none" with "BOOTPROTO=dhcp"
  637. and add a line consisting of "TYPE=Bonding". Note that the TYPE value
  638. is case sensitive.
  639. 3.2.2 Configuring Multiple Bonds with Initscripts
  640. -------------------------------------------------
  641. At this writing, the initscripts package does not directly
  642. support loading the bonding driver multiple times, so the process for
  643. doing so is the same as described in the "Configuring Multiple Bonds
  644. Manually" section, below.
  645. NOTE: It has been observed that some Red Hat supplied kernels
  646. are apparently unable to rename modules at load time (the "-o bond1"
  647. part). Attempts to pass that option to modprobe will produce an
  648. "Operation not permitted" error. This has been reported on some
  649. Fedora Core kernels, and has been seen on RHEL 4 as well. On kernels
  650. exhibiting this problem, it will be impossible to configure multiple
  651. bonds with differing parameters.
  652. 3.3 Configuring Bonding Manually with Ifenslave
  653. -----------------------------------------------
  654. This section applies to distros whose network initialization
  655. scripts (the sysconfig or initscripts package) do not have specific
  656. knowledge of bonding. One such distro is SuSE Linux Enterprise Server
  657. version 8.
  658. The general method for these systems is to place the bonding
  659. module parameters into /etc/modules.conf or /etc/modprobe.conf (as
  660. appropriate for the installed distro), then add modprobe and/or
  661. ifenslave commands to the system's global init script. The name of
  662. the global init script differs; for sysconfig, it is
  663. /etc/init.d/boot.local and for initscripts it is /etc/rc.d/rc.local.
  664. For example, if you wanted to make a simple bond of two e100
  665. devices (presumed to be eth0 and eth1), and have it persist across
  666. reboots, edit the appropriate file (/etc/init.d/boot.local or
  667. /etc/rc.d/rc.local), and add the following:
  668. modprobe bonding mode=balance-alb miimon=100
  669. modprobe e100
  670. ifconfig bond0 192.168.1.1 netmask 255.255.255.0 up
  671. ifenslave bond0 eth0
  672. ifenslave bond0 eth1
  673. Replace the example bonding module parameters and bond0
  674. network configuration (IP address, netmask, etc) with the appropriate
  675. values for your configuration.
  676. Unfortunately, this method will not provide support for the
  677. ifup and ifdown scripts on the bond devices. To reload the bonding
  678. configuration, it is necessary to run the initialization script, e.g.,
  679. # /etc/init.d/boot.local
  680. or
  681. # /etc/rc.d/rc.local
  682. It may be desirable in such a case to create a separate script
  683. which only initializes the bonding configuration, then call that
  684. separate script from within boot.local. This allows for bonding to be
  685. enabled without re-running the entire global init script.
  686. To shut down the bonding devices, it is necessary to first
  687. mark the bonding device itself as being down, then remove the
  688. appropriate device driver modules. For our example above, you can do
  689. the following:
  690. # ifconfig bond0 down
  691. # rmmod bonding
  692. # rmmod e100
  693. Again, for convenience, it may be desirable to create a script
  694. with these commands.
  695. 3.3.1 Configuring Multiple Bonds Manually
  696. -----------------------------------------
  697. This section contains information on configuring multiple
  698. bonding devices with differing options for those systems whose network
  699. initialization scripts lack support for configuring multiple bonds.
  700. If you require multiple bonding devices, but all with the same
  701. options, you may wish to use the "max_bonds" module parameter,
  702. documented above.
  703. To create multiple bonding devices with differing options, it
  704. is necessary to load the bonding driver multiple times. Note that
  705. current versions of the sysconfig network initialization scripts
  706. handle this automatically; if your distro uses these scripts, no
  707. special action is needed. See the section Configuring Bonding
  708. Devices, above, if you're not sure about your network initialization
  709. scripts.
  710. To load multiple instances of the module, it is necessary to
  711. specify a different name for each instance (the module loading system
  712. requires that every loaded module, even multiple instances of the same
  713. module, have a unique name). This is accomplished by supplying
  714. multiple sets of bonding options in /etc/modprobe.conf, for example:
  715. alias bond0 bonding
  716. options bond0 -o bond0 mode=balance-rr miimon=100
  717. alias bond1 bonding
  718. options bond1 -o bond1 mode=balance-alb miimon=50
  719. will load the bonding module two times. The first instance is
  720. named "bond0" and creates the bond0 device in balance-rr mode with an
  721. miimon of 100. The second instance is named "bond1" and creates the
  722. bond1 device in balance-alb mode with an miimon of 50.
  723. In some circumstances (typically with older distributions),
  724. the above does not work, and the second bonding instance never sees
  725. its options. In that case, the second options line can be substituted
  726. as follows:
  727. install bond1 /sbin/modprobe --ignore-install bonding -o bond1 \
  728. mode=balance-alb miimon=50
  729. This may be repeated any number of times, specifying a new and
  730. unique name in place of bond1 for each subsequent instance.
  731. 3.4 Configuring Bonding Manually via Sysfs
  732. ------------------------------------------
  733. Starting with version 3.0, Channel Bonding may be configured
  734. via the sysfs interface. This interface allows dynamic configuration
  735. of all bonds in the system without unloading the module. It also
  736. allows for adding and removing bonds at runtime. Ifenslave is no
  737. longer required, though it is still supported.
  738. Use of the sysfs interface allows you to use multiple bonds
  739. with different configurations without having to reload the module.
  740. It also allows you to use multiple, differently configured bonds when
  741. bonding is compiled into the kernel.
  742. You must have the sysfs filesystem mounted to configure
  743. bonding this way. The examples in this document assume that you
  744. are using the standard mount point for sysfs, e.g. /sys. If your
  745. sysfs filesystem is mounted elsewhere, you will need to adjust the
  746. example paths accordingly.
  747. Creating and Destroying Bonds
  748. -----------------------------
  749. To add a new bond foo:
  750. # echo +foo > /sys/class/net/bonding_masters
  751. To remove an existing bond bar:
  752. # echo -bar > /sys/class/net/bonding_masters
  753. To show all existing bonds:
  754. # cat /sys/class/net/bonding_masters
  755. NOTE: due to 4K size limitation of sysfs files, this list may be
  756. truncated if you have more than a few hundred bonds. This is unlikely
  757. to occur under normal operating conditions.
  758. Adding and Removing Slaves
  759. --------------------------
  760. Interfaces may be enslaved to a bond using the file
  761. /sys/class/net/<bond>/bonding/slaves. The semantics for this file
  762. are the same as for the bonding_masters file.
  763. To enslave interface eth0 to bond bond0:
  764. # ifconfig bond0 up
  765. # echo +eth0 > /sys/class/net/bond0/bonding/slaves
  766. To free slave eth0 from bond bond0:
  767. # echo -eth0 > /sys/class/net/bond0/bonding/slaves
  768. NOTE: The bond must be up before slaves can be added. All
  769. slaves are freed when the interface is brought down.
  770. When an interface is enslaved to a bond, symlinks between the
  771. two are created in the sysfs filesystem. In this case, you would get
  772. /sys/class/net/bond0/slave_eth0 pointing to /sys/class/net/eth0, and
  773. /sys/class/net/eth0/master pointing to /sys/class/net/bond0.
  774. This means that you can tell quickly whether or not an
  775. interface is enslaved by looking for the master symlink. Thus:
  776. # echo -eth0 > /sys/class/net/eth0/master/bonding/slaves
  777. will free eth0 from whatever bond it is enslaved to, regardless of
  778. the name of the bond interface.
  779. Changing a Bond's Configuration
  780. -------------------------------
  781. Each bond may be configured individually by manipulating the
  782. files located in /sys/class/net/<bond name>/bonding
  783. The names of these files correspond directly with the command-
  784. line parameters described elsewhere in this file, and, with the
  785. exception of arp_ip_target, they accept the same values. To see the
  786. current setting, simply cat the appropriate file.
  787. A few examples will be given here; for specific usage
  788. guidelines for each parameter, see the appropriate section in this
  789. document.
  790. To configure bond0 for balance-alb mode:
  791. # ifconfig bond0 down
  792. # echo 6 > /sys/class/net/bond0/bonding/mode
  793. - or -
  794. # echo balance-alb > /sys/class/net/bond0/bonding/mode
  795. NOTE: The bond interface must be down before the mode can be
  796. changed.
  797. To enable MII monitoring on bond0 with a 1 second interval:
  798. # echo 1000 > /sys/class/net/bond0/bonding/miimon
  799. NOTE: If ARP monitoring is enabled, it will disabled when MII
  800. monitoring is enabled, and vice-versa.
  801. To add ARP targets:
  802. # echo +192.168.0.100 > /sys/class/net/bond0/bonding/arp_ip_target
  803. # echo +192.168.0.101 > /sys/class/net/bond0/bonding/arp_ip_target
  804. NOTE: up to 10 target addresses may be specified.
  805. To remove an ARP target:
  806. # echo -192.168.0.100 > /sys/class/net/bond0/bonding/arp_ip_target
  807. Example Configuration
  808. ---------------------
  809. We begin with the same example that is shown in section 3.3,
  810. executed with sysfs, and without using ifenslave.
  811. To make a simple bond of two e100 devices (presumed to be eth0
  812. and eth1), and have it persist across reboots, edit the appropriate
  813. file (/etc/init.d/boot.local or /etc/rc.d/rc.local), and add the
  814. following:
  815. modprobe bonding
  816. modprobe e100
  817. echo balance-alb > /sys/class/net/bond0/bonding/mode
  818. ifconfig bond0 192.168.1.1 netmask 255.255.255.0 up
  819. echo 100 > /sys/class/net/bond0/bonding/miimon
  820. echo +eth0 > /sys/class/net/bond0/bonding/slaves
  821. echo +eth1 > /sys/class/net/bond0/bonding/slaves
  822. To add a second bond, with two e1000 interfaces in
  823. active-backup mode, using ARP monitoring, add the following lines to
  824. your init script:
  825. modprobe e1000
  826. echo +bond1 > /sys/class/net/bonding_masters
  827. echo active-backup > /sys/class/net/bond1/bonding/mode
  828. ifconfig bond1 192.168.2.1 netmask 255.255.255.0 up
  829. echo +192.168.2.100 /sys/class/net/bond1/bonding/arp_ip_target
  830. echo 2000 > /sys/class/net/bond1/bonding/arp_interval
  831. echo +eth2 > /sys/class/net/bond1/bonding/slaves
  832. echo +eth3 > /sys/class/net/bond1/bonding/slaves
  833. 4. Querying Bonding Configuration
  834. =================================
  835. 4.1 Bonding Configuration
  836. -------------------------
  837. Each bonding device has a read-only file residing in the
  838. /proc/net/bonding directory. The file contents include information
  839. about the bonding configuration, options and state of each slave.
  840. For example, the contents of /proc/net/bonding/bond0 after the
  841. driver is loaded with parameters of mode=0 and miimon=1000 is
  842. generally as follows:
  843. Ethernet Channel Bonding Driver: 2.6.1 (October 29, 2004)
  844. Bonding Mode: load balancing (round-robin)
  845. Currently Active Slave: eth0
  846. MII Status: up
  847. MII Polling Interval (ms): 1000
  848. Up Delay (ms): 0
  849. Down Delay (ms): 0
  850. Slave Interface: eth1
  851. MII Status: up
  852. Link Failure Count: 1
  853. Slave Interface: eth0
  854. MII Status: up
  855. Link Failure Count: 1
  856. The precise format and contents will change depending upon the
  857. bonding configuration, state, and version of the bonding driver.
  858. 4.2 Network configuration
  859. -------------------------
  860. The network configuration can be inspected using the ifconfig
  861. command. Bonding devices will have the MASTER flag set; Bonding slave
  862. devices will have the SLAVE flag set. The ifconfig output does not
  863. contain information on which slaves are associated with which masters.
  864. In the example below, the bond0 interface is the master
  865. (MASTER) while eth0 and eth1 are slaves (SLAVE). Notice all slaves of
  866. bond0 have the same MAC address (HWaddr) as bond0 for all modes except
  867. TLB and ALB that require a unique MAC address for each slave.
  868. # /sbin/ifconfig
  869. bond0 Link encap:Ethernet HWaddr 00:C0:F0:1F:37:B4
  870. inet addr:XXX.XXX.XXX.YYY Bcast:XXX.XXX.XXX.255 Mask:255.255.252.0
  871. UP BROADCAST RUNNING MASTER MULTICAST MTU:1500 Metric:1
  872. RX packets:7224794 errors:0 dropped:0 overruns:0 frame:0
  873. TX packets:3286647 errors:1 dropped:0 overruns:1 carrier:0
  874. collisions:0 txqueuelen:0
  875. eth0 Link encap:Ethernet HWaddr 00:C0:F0:1F:37:B4
  876. UP BROADCAST RUNNING SLAVE MULTICAST MTU:1500 Metric:1
  877. RX packets:3573025 errors:0 dropped:0 overruns:0 frame:0
  878. TX packets:1643167 errors:1 dropped:0 overruns:1 carrier:0
  879. collisions:0 txqueuelen:100
  880. Interrupt:10 Base address:0x1080
  881. eth1 Link encap:Ethernet HWaddr 00:C0:F0:1F:37:B4
  882. UP BROADCAST RUNNING SLAVE MULTICAST MTU:1500 Metric:1
  883. RX packets:3651769 errors:0 dropped:0 overruns:0 frame:0
  884. TX packets:1643480 errors:0 dropped:0 overruns:0 carrier:0
  885. collisions:0 txqueuelen:100
  886. Interrupt:9 Base address:0x1400
  887. 5. Switch Configuration
  888. =======================
  889. For this section, "switch" refers to whatever system the
  890. bonded devices are directly connected to (i.e., where the other end of
  891. the cable plugs into). This may be an actual dedicated switch device,
  892. or it may be another regular system (e.g., another computer running
  893. Linux),
  894. The active-backup, balance-tlb and balance-alb modes do not
  895. require any specific configuration of the switch.
  896. The 802.3ad mode requires that the switch have the appropriate
  897. ports configured as an 802.3ad aggregation. The precise method used
  898. to configure this varies from switch to switch, but, for example, a
  899. Cisco 3550 series switch requires that the appropriate ports first be
  900. grouped together in a single etherchannel instance, then that
  901. etherchannel is set to mode "lacp" to enable 802.3ad (instead of
  902. standard EtherChannel).
  903. The balance-rr, balance-xor and broadcast modes generally
  904. require that the switch have the appropriate ports grouped together.
  905. The nomenclature for such a group differs between switches, it may be
  906. called an "etherchannel" (as in the Cisco example, above), a "trunk
  907. group" or some other similar variation. For these modes, each switch
  908. will also have its own configuration options for the switch's transmit
  909. policy to the bond. Typical choices include XOR of either the MAC or
  910. IP addresses. The transmit policy of the two peers does not need to
  911. match. For these three modes, the bonding mode really selects a
  912. transmit policy for an EtherChannel group; all three will interoperate
  913. with another EtherChannel group.
  914. 6. 802.1q VLAN Support
  915. ======================
  916. It is possible to configure VLAN devices over a bond interface
  917. using the 8021q driver. However, only packets coming from the 8021q
  918. driver and passing through bonding will be tagged by default. Self
  919. generated packets, for example, bonding's learning packets or ARP
  920. packets generated by either ALB mode or the ARP monitor mechanism, are
  921. tagged internally by bonding itself. As a result, bonding must
  922. "learn" the VLAN IDs configured above it, and use those IDs to tag
  923. self generated packets.
  924. For reasons of simplicity, and to support the use of adapters
  925. that can do VLAN hardware acceleration offloading, the bonding
  926. interface declares itself as fully hardware offloading capable, it gets
  927. the add_vid/kill_vid notifications to gather the necessary
  928. information, and it propagates those actions to the slaves. In case
  929. of mixed adapter types, hardware accelerated tagged packets that
  930. should go through an adapter that is not offloading capable are
  931. "un-accelerated" by the bonding driver so the VLAN tag sits in the
  932. regular location.
  933. VLAN interfaces *must* be added on top of a bonding interface
  934. only after enslaving at least one slave. The bonding interface has a
  935. hardware address of 00:00:00:00:00:00 until the first slave is added.
  936. If the VLAN interface is created prior to the first enslavement, it
  937. would pick up the all-zeroes hardware address. Once the first slave
  938. is attached to the bond, the bond device itself will pick up the
  939. slave's hardware address, which is then available for the VLAN device.
  940. Also, be aware that a similar problem can occur if all slaves
  941. are released from a bond that still has one or more VLAN interfaces on
  942. top of it. When a new slave is added, the bonding interface will
  943. obtain its hardware address from the first slave, which might not
  944. match the hardware address of the VLAN interfaces (which was
  945. ultimately copied from an earlier slave).
  946. There are two methods to insure that the VLAN device operates
  947. with the correct hardware address if all slaves are removed from a
  948. bond interface:
  949. 1. Remove all VLAN interfaces then recreate them
  950. 2. Set the bonding interface's hardware address so that it
  951. matches the hardware address of the VLAN interfaces.
  952. Note that changing a VLAN interface's HW address would set the
  953. underlying device -- i.e. the bonding interface -- to promiscuous
  954. mode, which might not be what you want.
  955. 7. Link Monitoring
  956. ==================
  957. The bonding driver at present supports two schemes for
  958. monitoring a slave device's link state: the ARP monitor and the MII
  959. monitor.
  960. At the present time, due to implementation restrictions in the
  961. bonding driver itself, it is not possible to enable both ARP and MII
  962. monitoring simultaneously.
  963. 7.1 ARP Monitor Operation
  964. -------------------------
  965. The ARP monitor operates as its name suggests: it sends ARP
  966. queries to one or more designated peer systems on the network, and
  967. uses the response as an indication that the link is operating. This
  968. gives some assurance that traffic is actually flowing to and from one
  969. or more peers on the local network.
  970. The ARP monitor relies on the device driver itself to verify
  971. that traffic is flowing. In particular, the driver must keep up to
  972. date the last receive time, dev->last_rx, and transmit start time,
  973. dev->trans_start. If these are not updated by the driver, then the
  974. ARP monitor will immediately fail any slaves using that driver, and
  975. those slaves will stay down. If networking monitoring (tcpdump, etc)
  976. shows the ARP requests and replies on the network, then it may be that
  977. your device driver is not updating last_rx and trans_start.
  978. 7.2 Configuring Multiple ARP Targets
  979. ------------------------------------
  980. While ARP monitoring can be done with just one target, it can
  981. be useful in a High Availability setup to have several targets to
  982. monitor. In the case of just one target, the target itself may go
  983. down or have a problem making it unresponsive to ARP requests. Having
  984. an additional target (or several) increases the reliability of the ARP
  985. monitoring.
  986. Multiple ARP targets must be separated by commas as follows:
  987. # example options for ARP monitoring with three targets
  988. alias bond0 bonding
  989. options bond0 arp_interval=60 arp_ip_target=192.168.0.1,192.168.0.3,192.168.0.9
  990. For just a single target the options would resemble:
  991. # example options for ARP monitoring with one target
  992. alias bond0 bonding
  993. options bond0 arp_interval=60 arp_ip_target=192.168.0.100
  994. 7.3 MII Monitor Operation
  995. -------------------------
  996. The MII monitor monitors only the carrier state of the local
  997. network interface. It accomplishes this in one of three ways: by
  998. depending upon the device driver to maintain its carrier state, by
  999. querying the device's MII registers, or by making an ethtool query to
  1000. the device.
  1001. If the use_carrier module parameter is 1 (the default value),
  1002. then the MII monitor will rely on the driver for carrier state
  1003. information (via the netif_carrier subsystem). As explained in the
  1004. use_carrier parameter information, above, if the MII monitor fails to
  1005. detect carrier loss on the device (e.g., when the cable is physically
  1006. disconnected), it may be that the driver does not support
  1007. netif_carrier.
  1008. If use_carrier is 0, then the MII monitor will first query the
  1009. device's (via ioctl) MII registers and check the link state. If that
  1010. request fails (not just that it returns carrier down), then the MII
  1011. monitor will make an ethtool ETHOOL_GLINK request to attempt to obtain
  1012. the same information. If both methods fail (i.e., the driver either
  1013. does not support or had some error in processing both the MII register
  1014. and ethtool requests), then the MII monitor will assume the link is
  1015. up.
  1016. 8. Potential Sources of Trouble
  1017. ===============================
  1018. 8.1 Adventures in Routing
  1019. -------------------------
  1020. When bonding is configured, it is important that the slave
  1021. devices not have routes that supersede routes of the master (or,
  1022. generally, not have routes at all). For example, suppose the bonding
  1023. device bond0 has two slaves, eth0 and eth1, and the routing table is
  1024. as follows:
  1025. Kernel IP routing table
  1026. Destination Gateway Genmask Flags MSS Window irtt Iface
  1027. 10.0.0.0 0.0.0.0 255.255.0.0 U 40 0 0 eth0
  1028. 10.0.0.0 0.0.0.0 255.255.0.0 U 40 0 0 eth1
  1029. 10.0.0.0 0.0.0.0 255.255.0.0 U 40 0 0 bond0
  1030. 127.0.0.0 0.0.0.0 255.0.0.0 U 40 0 0 lo
  1031. This routing configuration will likely still update the
  1032. receive/transmit times in the driver (needed by the ARP monitor), but
  1033. may bypass the bonding driver (because outgoing traffic to, in this
  1034. case, another host on network 10 would use eth0 or eth1 before bond0).
  1035. The ARP monitor (and ARP itself) may become confused by this
  1036. configuration, because ARP requests (generated by the ARP monitor)
  1037. will be sent on one interface (bond0), but the corresponding reply
  1038. will arrive on a different interface (eth0). This reply looks to ARP
  1039. as an unsolicited ARP reply (because ARP matches replies on an
  1040. interface basis), and is discarded. The MII monitor is not affected
  1041. by the state of the routing table.
  1042. The solution here is simply to insure that slaves do not have
  1043. routes of their own, and if for some reason they must, those routes do
  1044. not supersede routes of their master. This should generally be the
  1045. case, but unusual configurations or errant manual or automatic static
  1046. route additions may cause trouble.
  1047. 8.2 Ethernet Device Renaming
  1048. ----------------------------
  1049. On systems with network configuration scripts that do not
  1050. associate physical devices directly with network interface names (so
  1051. that the same physical device always has the same "ethX" name), it may
  1052. be necessary to add some special logic to either /etc/modules.conf or
  1053. /etc/modprobe.conf (depending upon which is installed on the system).
  1054. For example, given a modules.conf containing the following:
  1055. alias bond0 bonding
  1056. options bond0 mode=some-mode miimon=50
  1057. alias eth0 tg3
  1058. alias eth1 tg3
  1059. alias eth2 e1000
  1060. alias eth3 e1000
  1061. If neither eth0 and eth1 are slaves to bond0, then when the
  1062. bond0 interface comes up, the devices may end up reordered. This
  1063. happens because bonding is loaded first, then its slave device's
  1064. drivers are loaded next. Since no other drivers have been loaded,
  1065. when the e1000 driver loads, it will receive eth0 and eth1 for its
  1066. devices, but the bonding configuration tries to enslave eth2 and eth3
  1067. (which may later be assigned to the tg3 devices).
  1068. Adding the following:
  1069. add above bonding e1000 tg3
  1070. causes modprobe to load e1000 then tg3, in that order, when
  1071. bonding is loaded. This command is fully documented in the
  1072. modules.conf manual page.
  1073. On systems utilizing modprobe.conf (or modprobe.conf.local),
  1074. an equivalent problem can occur. In this case, the following can be
  1075. added to modprobe.conf (or modprobe.conf.local, as appropriate), as
  1076. follows (all on one line; it has been split here for clarity):
  1077. install bonding /sbin/modprobe tg3; /sbin/modprobe e1000;
  1078. /sbin/modprobe --ignore-install bonding
  1079. This will, when loading the bonding module, rather than
  1080. performing the normal action, instead execute the provided command.
  1081. This command loads the device drivers in the order needed, then calls
  1082. modprobe with --ignore-install to cause the normal action to then take
  1083. place. Full documentation on this can be found in the modprobe.conf
  1084. and modprobe manual pages.
  1085. 8.3. Painfully Slow Or No Failed Link Detection By Miimon
  1086. ---------------------------------------------------------
  1087. By default, bonding enables the use_carrier option, which
  1088. instructs bonding to trust the driver to maintain carrier state.
  1089. As discussed in the options section, above, some drivers do
  1090. not support the netif_carrier_on/_off link state tracking system.
  1091. With use_carrier enabled, bonding will always see these links as up,
  1092. regardless of their actual state.
  1093. Additionally, other drivers do support netif_carrier, but do
  1094. not maintain it in real time, e.g., only polling the link state at
  1095. some fixed interval. In this case, miimon will detect failures, but
  1096. only after some long period of time has expired. If it appears that
  1097. miimon is very slow in detecting link failures, try specifying
  1098. use_carrier=0 to see if that improves the failure detection time. If
  1099. it does, then it may be that the driver checks the carrier state at a
  1100. fixed interval, but does not cache the MII register values (so the
  1101. use_carrier=0 method of querying the registers directly works). If
  1102. use_carrier=0 does not improve the failover, then the driver may cache
  1103. the registers, or the problem may be elsewhere.
  1104. Also, remember that miimon only checks for the device's
  1105. carrier state. It has no way to determine the state of devices on or
  1106. beyond other ports of a switch, or if a switch is refusing to pass
  1107. traffic while still maintaining carrier on.
  1108. 9. SNMP agents
  1109. ===============
  1110. If running SNMP agents, the bonding driver should be loaded
  1111. before any network drivers participating in a bond. This requirement
  1112. is due to the interface index (ipAdEntIfIndex) being associated to
  1113. the first interface found with a given IP address. That is, there is
  1114. only one ipAdEntIfIndex for each IP address. For example, if eth0 and
  1115. eth1 are slaves of bond0 and the driver for eth0 is loaded before the
  1116. bonding driver, the interface for the IP address will be associated
  1117. with the eth0 interface. This configuration is shown below, the IP
  1118. address 192.168.1.1 has an interface index of 2 which indexes to eth0
  1119. in the ifDescr table (ifDescr.2).
  1120. interfaces.ifTable.ifEntry.ifDescr.1 = lo
  1121. interfaces.ifTable.ifEntry.ifDescr.2 = eth0
  1122. interfaces.ifTable.ifEntry.ifDescr.3 = eth1
  1123. interfaces.ifTable.ifEntry.ifDescr.4 = eth2
  1124. interfaces.ifTable.ifEntry.ifDescr.5 = eth3
  1125. interfaces.ifTable.ifEntry.ifDescr.6 = bond0
  1126. ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.10.10.10 = 5
  1127. ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.192.168.1.1 = 2
  1128. ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.74.20.94 = 4
  1129. ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.127.0.0.1 = 1
  1130. This problem is avoided by loading the bonding driver before
  1131. any network drivers participating in a bond. Below is an example of
  1132. loading the bonding driver first, the IP address 192.168.1.1 is
  1133. correctly associated with ifDescr.2.
  1134. interfaces.ifTable.ifEntry.ifDescr.1 = lo
  1135. interfaces.ifTable.ifEntry.ifDescr.2 = bond0
  1136. interfaces.ifTable.ifEntry.ifDescr.3 = eth0
  1137. interfaces.ifTable.ifEntry.ifDescr.4 = eth1
  1138. interfaces.ifTable.ifEntry.ifDescr.5 = eth2
  1139. interfaces.ifTable.ifEntry.ifDescr.6 = eth3
  1140. ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.10.10.10 = 6
  1141. ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.192.168.1.1 = 2
  1142. ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.74.20.94 = 5
  1143. ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.127.0.0.1 = 1
  1144. While some distributions may not report the interface name in
  1145. ifDescr, the association between the IP address and IfIndex remains
  1146. and SNMP functions such as Interface_Scan_Next will report that
  1147. association.
  1148. 10. Promiscuous mode
  1149. ====================
  1150. When running network monitoring tools, e.g., tcpdump, it is
  1151. common to enable promiscuous mode on the device, so that all traffic
  1152. is seen (instead of seeing only traffic destined for the local host).
  1153. The bonding driver handles promiscuous mode changes to the bonding
  1154. master device (e.g., bond0), and propagates the setting to the slave
  1155. devices.
  1156. For the balance-rr, balance-xor, broadcast, and 802.3ad modes,
  1157. the promiscuous mode setting is propagated to all slaves.
  1158. For the active-backup, balance-tlb and balance-alb modes, the
  1159. promiscuous mode setting is propagated only to the active slave.
  1160. For balance-tlb mode, the active slave is the slave currently
  1161. receiving inbound traffic.
  1162. For balance-alb mode, the active slave is the slave used as a
  1163. "primary." This slave is used for mode-specific control traffic, for
  1164. sending to peers that are unassigned or if the load is unbalanced.
  1165. For the active-backup, balance-tlb and balance-alb modes, when
  1166. the active slave changes (e.g., due to a link failure), the
  1167. promiscuous setting will be propagated to the new active slave.
  1168. 11. Configuring Bonding for High Availability
  1169. =============================================
  1170. High Availability refers to configurations that provide
  1171. maximum network availability by having redundant or backup devices,
  1172. links or switches between the host and the rest of the world. The
  1173. goal is to provide the maximum availability of network connectivity
  1174. (i.e., the network always works), even though other configurations
  1175. could provide higher throughput.
  1176. 11.1 High Availability in a Single Switch Topology
  1177. --------------------------------------------------
  1178. If two hosts (or a host and a single switch) are directly
  1179. connected via multiple physical links, then there is no availability
  1180. penalty to optimizing for maximum bandwidth. In this case, there is
  1181. only one switch (or peer), so if it fails, there is no alternative
  1182. access to fail over to. Additionally, the bonding load balance modes
  1183. support link monitoring of their members, so if individual links fail,
  1184. the load will be rebalanced across the remaining devices.
  1185. See Section 13, "Configuring Bonding for Maximum Throughput"
  1186. for information on configuring bonding with one peer device.
  1187. 11.2 High Availability in a Multiple Switch Topology
  1188. ----------------------------------------------------
  1189. With multiple switches, the configuration of bonding and the
  1190. network changes dramatically. In multiple switch topologies, there is
  1191. a trade off between network availability and usable bandwidth.
  1192. Below is a sample network, configured to maximize the
  1193. availability of the network:
  1194. | |
  1195. |port3 port3|
  1196. +-----+----+ +-----+----+
  1197. | |port2 ISL port2| |
  1198. | switch A +--------------------------+ switch B |
  1199. | | | |
  1200. +-----+----+ +-----++---+
  1201. |port1 port1|
  1202. | +-------+ |
  1203. +-------------+ host1 +---------------+
  1204. eth0 +-------+ eth1
  1205. In this configuration, there is a link between the two
  1206. switches (ISL, or inter switch link), and multiple ports connecting to
  1207. the outside world ("port3" on each switch). There is no technical
  1208. reason that this could not be extended to a third switch.
  1209. 11.2.1 HA Bonding Mode Selection for Multiple Switch Topology
  1210. -------------------------------------------------------------
  1211. In a topology such as the example above, the active-backup and
  1212. broadcast modes are the only useful bonding modes when optimizing for
  1213. availability; the other modes require all links to terminate on the
  1214. same peer for them to behave rationally.
  1215. active-backup: This is generally the preferred mode, particularly if
  1216. the switches have an ISL and play together well. If the
  1217. network configuration is such that one switch is specifically
  1218. a backup switch (e.g., has lower capacity, higher cost, etc),
  1219. then the primary option can be used to insure that the
  1220. preferred link is always used when it is available.
  1221. broadcast: This mode is really a special purpose mode, and is suitable
  1222. only for very specific needs. For example, if the two
  1223. switches are not connected (no ISL), and the networks beyond
  1224. them are totally independent. In this case, if it is
  1225. necessary for some specific one-way traffic to reach both
  1226. independent networks, then the broadcast mode may be suitable.
  1227. 11.2.2 HA Link Monitoring Selection for Multiple Switch Topology
  1228. ----------------------------------------------------------------
  1229. The choice of link monitoring ultimately depends upon your
  1230. switch. If the switch can reliably fail ports in response to other
  1231. failures, then either the MII or ARP monitors should work. For
  1232. example, in the above example, if the "port3" link fails at the remote
  1233. end, the MII monitor has no direct means to detect this. The ARP
  1234. monitor could be configured with a target at the remote end of port3,
  1235. thus detecting that failure without switch support.
  1236. In general, however, in a multiple switch topology, the ARP
  1237. monitor can provide a higher level of reliability in detecting end to
  1238. end connectivity failures (which may be caused by the failure of any
  1239. individual component to pass traffic for any reason). Additionally,
  1240. the ARP monitor should be configured with multiple targets (at least
  1241. one for each switch in the network). This will insure that,
  1242. regardless of which switch is active, the ARP monitor has a suitable
  1243. target to query.
  1244. 12. Configuring Bonding for Maximum Throughput
  1245. ==============================================
  1246. 12.1 Maximizing Throughput in a Single Switch Topology
  1247. ------------------------------------------------------
  1248. In a single switch configuration, the best method to maximize
  1249. throughput depends upon the application and network environment. The
  1250. various load balancing modes each have strengths and weaknesses in
  1251. different environments, as detailed below.
  1252. For this discussion, we will break down the topologies into
  1253. two categories. Depending upon the destination of most traffic, we
  1254. categorize them into either "gatewayed" or "local" configurations.
  1255. In a gatewayed configuration, the "switch" is acting primarily
  1256. as a router, and the majority of traffic passes through this router to
  1257. other networks. An example would be the following:
  1258. +----------+ +----------+
  1259. | |eth0 port1| | to other networks
  1260. | Host A +---------------------+ router +------------------->
  1261. | +---------------------+ | Hosts B and C are out
  1262. | |eth1 port2| | here somewhere
  1263. +----------+ +----------+
  1264. The router may be a dedicated router device, or another host
  1265. acting as a gateway. For our discussion, the important point is that
  1266. the majority of traffic from Host A will pass through the router to
  1267. some other network before reaching its final destination.
  1268. In a gatewayed network configuration, although Host A may
  1269. communicate with many other systems, all of its traffic will be sent
  1270. and received via one other peer on the local network, the router.
  1271. Note that the case of two systems connected directly via
  1272. multiple physical links is, for purposes of configuring bonding, the
  1273. same as a gatewayed configuration. In that case, it happens that all
  1274. traffic is destined for the "gateway" itself, not some other network
  1275. beyond the gateway.
  1276. In a local configuration, the "switch" is acting primarily as
  1277. a switch, and the majority of traffic passes through this switch to
  1278. reach other stations on the same network. An example would be the
  1279. following:
  1280. +----------+ +----------+ +--------+
  1281. | |eth0 port1| +-------+ Host B |
  1282. | Host A +------------+ switch |port3 +--------+
  1283. | +------------+ | +--------+
  1284. | |eth1 port2| +------------------+ Host C |
  1285. +----------+ +----------+port4 +--------+
  1286. Again, the switch may be a dedicated switch device, or another
  1287. host acting as a gateway. For our discussion, the important point is
  1288. that the majority of traffic from Host A is destined for other hosts
  1289. on the same local network (Hosts B and C in the above example).
  1290. In summary, in a gatewayed configuration, traffic to and from
  1291. the bonded device will be to the same MAC level peer on the network
  1292. (the gateway itself, i.e., the router), regardless of its final
  1293. destination. In a local configuration, traffic flows directly to and
  1294. from the final destinations, thus, each destination (Host B, Host C)
  1295. will be addressed directly by their individual MAC addresses.
  1296. This distinction between a gatewayed and a local network
  1297. configuration is important because many of the load balancing modes
  1298. available use the MAC addresses of the local network source and
  1299. destination to make load balancing decisions. The behavior of each
  1300. mode is described below.
  1301. 12.1.1 MT Bonding Mode Selection for Single Switch Topology
  1302. -----------------------------------------------------------
  1303. This configuration is the easiest to set up and to understand,
  1304. although you will have to decide which bonding mode best suits your
  1305. needs. The trade offs for each mode are detailed below:
  1306. balance-rr: This mode is the only mode that will permit a single
  1307. TCP/IP connection to stripe traffic across multiple
  1308. interfaces. It is therefore the only mode that will allow a
  1309. single TCP/IP stream to utilize more than one interface's
  1310. worth of throughput. This comes at a cost, however: the
  1311. striping often results in peer systems receiving packets out
  1312. of order, causing TCP/IP's congestion control system to kick
  1313. in, often by retransmitting segments.
  1314. It is possible to adjust TCP/IP's congestion limits by
  1315. altering the net.ipv4.tcp_reordering sysctl parameter. The
  1316. usual default value is 3, and the maximum useful value is 127.
  1317. For a four interface balance-rr bond, expect that a single
  1318. TCP/IP stream will utilize no more than approximately 2.3
  1319. interface's worth of throughput, even after adjusting
  1320. tcp_reordering.
  1321. Note that this out of order delivery occurs when both the
  1322. sending and receiving systems are utilizing a multiple
  1323. interface bond. Consider a configuration in which a
  1324. balance-rr bond feeds into a single higher capacity network
  1325. channel (e.g., multiple 100Mb/sec ethernets feeding a single
  1326. gigabit ethernet via an etherchannel capable switch). In this
  1327. configuration, traffic sent from the multiple 100Mb devices to
  1328. a destination connected to the gigabit device will not see
  1329. packets out of order. However, traffic sent from the gigabit
  1330. device to the multiple 100Mb devices may or may not see
  1331. traffic out of order, depending upon the balance policy of the
  1332. switch. Many switches do not support any modes that stripe
  1333. traffic (instead choosing a port based upon IP or MAC level
  1334. addresses); for those devices, traffic flowing from the
  1335. gigabit device to the many 100Mb devices will only utilize one
  1336. interface.
  1337. If you are utilizing protocols other than TCP/IP, UDP for
  1338. example, and your application can tolerate out of order
  1339. delivery, then this mode can allow for single stream datagram
  1340. performance that scales near linearly as interfaces are added
  1341. to the bond.
  1342. This mode requires the switch to have the appropriate ports
  1343. configured for "etherchannel" or "trunking."
  1344. active-backup: There is not much advantage in this network topology to
  1345. the active-backup mode, as the inactive backup devices are all
  1346. connected to the same peer as the primary. In this case, a
  1347. load balancing mode (with link monitoring) will provide the
  1348. same level of network availability, but with increased
  1349. available bandwidth. On the plus side, active-backup mode
  1350. does not require any configuration of the switch, so it may
  1351. have value if the hardware available does not support any of
  1352. the load balance modes.
  1353. balance-xor: This mode will limit traffic such that packets destined
  1354. for specific peers will always be sent over the same
  1355. interface. Since the destination is determined by the MAC
  1356. addresses involved, this mode works best in a "local" network
  1357. configuration (as described above), with destinations all on
  1358. the same local network. This mode is likely to be suboptimal
  1359. if all your traffic is passed through a single router (i.e., a
  1360. "gatewayed" network configuration, as described above).
  1361. As with balance-rr, the switch ports need to be configured for
  1362. "etherchannel" or "trunking."
  1363. broadcast: Like active-backup, there is not much advantage to this
  1364. mode in this type of network topology.
  1365. 802.3ad: This mode can be a good choice for this type of network
  1366. topology. The 802.3ad mode is an IEEE standard, so all peers
  1367. that implement 802.3ad should interoperate well. The 802.3ad
  1368. protocol includes automatic configuration of the aggregates,
  1369. so minimal manual configuration of the switch is needed
  1370. (typically only to designate that some set of devices is
  1371. available for 802.3ad). The 802.3ad standard also mandates
  1372. that frames be delivered in order (within certain limits), so
  1373. in general single connections will not see misordering of
  1374. packets. The 802.3ad mode does have some drawbacks: the
  1375. standard mandates that all devices in the aggregate operate at
  1376. the same speed and duplex. Also, as with all bonding load
  1377. balance modes other than balance-rr, no single connection will
  1378. be able to utilize more than a single interface's worth of
  1379. bandwidth.
  1380. Additionally, the linux bonding 802.3ad implementation
  1381. distributes traffic by peer (using an XOR of MAC addresses),
  1382. so in a "gatewayed" configuration, all outgoing traffic will
  1383. generally use the same device. Incoming traffic may also end
  1384. up on a single device, but that is dependent upon the
  1385. balancing policy of the peer's 8023.ad implementation. In a
  1386. "local" configuration, traffic will be distributed across the
  1387. devices in the bond.
  1388. Finally, the 802.3ad mode mandates the use of the MII monitor,
  1389. therefore, the ARP monitor is not available in this mode.
  1390. balance-tlb: The balance-tlb mode balances outgoing traffic by peer.
  1391. Since the balancing is done according to MAC address, in a
  1392. "gatewayed" configuration (as described above), this mode will
  1393. send all traffic across a single device. However, in a
  1394. "local" network configuration, this mode balances multiple
  1395. local network peers across devices in a vaguely intelligent
  1396. manner (not a simple XOR as in balance-xor or 802.3ad mode),
  1397. so that mathematically unlucky MAC addresses (i.e., ones that
  1398. XOR to the same value) will not all "bunch up" on a single
  1399. interface.
  1400. Unlike 802.3ad, interfaces may be of differing speeds, and no
  1401. special switch configuration is required. On the down side,
  1402. in this mode all incoming traffic arrives over a single
  1403. interface, this mode requires certain ethtool support in the
  1404. network device driver of the slave interfaces, and the ARP
  1405. monitor is not available.
  1406. balance-alb: This mode is everything that balance-tlb is, and more.
  1407. It has all of the features (and restrictions) of balance-tlb,
  1408. and will also balance incoming traffic from local network
  1409. peers (as described in the Bonding Module Options section,
  1410. above).
  1411. The only additional down side to this mode is that the network
  1412. device driver must support changing the hardware address while
  1413. the device is open.
  1414. 12.1.2 MT Link Monitoring for Single Switch Topology
  1415. ----------------------------------------------------
  1416. The choice of link monitoring may largely depend upon which
  1417. mode you choose to use. The more advanced load balancing modes do not
  1418. support the use of the ARP monitor, and are thus restricted to using
  1419. the MII monitor (which does not provide as high a level of end to end
  1420. assurance as the ARP monitor).
  1421. 12.2 Maximum Throughput in a Multiple Switch Topology
  1422. -----------------------------------------------------
  1423. Multiple switches may be utilized to optimize for throughput
  1424. when they are configured in parallel as part of an isolated network
  1425. between two or more systems, for example:
  1426. +-----------+
  1427. | Host A |
  1428. +-+---+---+-+
  1429. | | |
  1430. +--------+ | +---------+
  1431. | | |
  1432. +------+---+ +-----+----+ +-----+----+
  1433. | Switch A | | Switch B | | Switch C |
  1434. +------+---+ +-----+----+ +-----+----+
  1435. | | |
  1436. +--------+ | +---------+
  1437. | | |
  1438. +-+---+---+-+
  1439. | Host B |
  1440. +-----------+
  1441. In this configuration, the switches are isolated from one
  1442. another. One reason to employ a topology such as this is for an
  1443. isolated network with many hosts (a cluster configured for high
  1444. performance, for example), using multiple smaller switches can be more
  1445. cost effective than a single larger switch, e.g., on a network with 24
  1446. hosts, three 24 port switches can be significantly less expensive than
  1447. a single 72 port switch.
  1448. If access beyond the network is required, an individual host
  1449. can be equipped with an additional network device connected to an
  1450. external network; this host then additionally acts as a gateway.
  1451. 12.2.1 MT Bonding Mode Selection for Multiple Switch Topology
  1452. -------------------------------------------------------------
  1453. In actual practice, the bonding mode typically employed in
  1454. configurations of this type is balance-rr. Historically, in this
  1455. network configuration, the usual caveats about out of order packet
  1456. delivery are mitigated by the use of network adapters that do not do
  1457. any kind of packet coalescing (via the use of NAPI, or because the
  1458. device itself does not generate interrupts until some number of
  1459. packets has arrived). When employed in this fashion, the balance-rr
  1460. mode allows individual connections between two hosts to effectively
  1461. utilize greater than one interface's bandwidth.
  1462. 12.2.2 MT Link Monitoring for Multiple Switch Topology
  1463. ------------------------------------------------------
  1464. Again, in actual practice, the MII monitor is most often used
  1465. in this configuration, as performance is given preference over
  1466. availability. The ARP monitor will function in this topology, but its
  1467. advantages over the MII monitor are mitigated by the volume of probes
  1468. needed as the number of systems involved grows (remember that each
  1469. host in the network is configured with bonding).
  1470. 13. Switch Behavior Issues
  1471. ==========================
  1472. 13.1 Link Establishment and Failover Delays
  1473. -------------------------------------------
  1474. Some switches exhibit undesirable behavior with regard to the
  1475. timing of link up and down reporting by the switch.
  1476. First, when a link comes up, some switches may indicate that
  1477. the link is up (carrier available), but not pass traffic over the
  1478. interface for some period of time. This delay is typically due to
  1479. some type of autonegotiation or routing protocol, but may also occur
  1480. during switch initialization (e.g., during recovery after a switch
  1481. failure). If you find this to be a problem, specify an appropriate
  1482. value to the updelay bonding module option to delay the use of the
  1483. relevant interface(s).
  1484. Second, some switches may "bounce" the link state one or more
  1485. times while a link is changing state. This occurs most commonly while
  1486. the switch is initializing. Again, an appropriate updelay value may
  1487. help.
  1488. Note that when a bonding interface has no active links, the
  1489. driver will immediately reuse the first link that goes up, even if the
  1490. updelay parameter has been specified (the updelay is ignored in this
  1491. case). If there are slave interfaces waiting for the updelay timeout
  1492. to expire, the interface that first went into that state will be
  1493. immediately reused. This reduces down time of the network if the
  1494. value of updelay has been overestimated, and since this occurs only in
  1495. cases with no connectivity, there is no additional penalty for
  1496. ignoring the updelay.
  1497. In addition to the concerns about switch timings, if your
  1498. switches take a long time to go into backup mode, it may be desirable
  1499. to not activate a backup interface immediately after a link goes down.
  1500. Failover may be delayed via the downdelay bonding module option.
  1501. 13.2 Duplicated Incoming Packets
  1502. --------------------------------
  1503. It is not uncommon to observe a short burst of duplicated
  1504. traffic when the bonding device is first used, or after it has been
  1505. idle for some period of time. This is most easily observed by issuing
  1506. a "ping" to some other host on the network, and noticing that the
  1507. output from ping flags duplicates (typically one per slave).
  1508. For example, on a bond in active-backup mode with five slaves
  1509. all connected to one switch, the output may appear as follows:
  1510. # ping -n 10.0.4.2
  1511. PING 10.0.4.2 (10.0.4.2) from 10.0.3.10 : 56(84) bytes of data.
  1512. 64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.7 ms
  1513. 64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
  1514. 64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
  1515. 64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
  1516. 64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)
  1517. 64 bytes from 10.0.4.2: icmp_seq=2 ttl=64 time=0.216 ms
  1518. 64 bytes from 10.0.4.2: icmp_seq=3 ttl=64 time=0.267 ms
  1519. 64 bytes from 10.0.4.2: icmp_seq=4 ttl=64 time=0.222 ms
  1520. This is not due to an error in the bonding driver, rather, it
  1521. is a side effect of how many switches update their MAC forwarding
  1522. tables. Initially, the switch does not associate the MAC address in
  1523. the packet with a particular switch port, and so it may send the
  1524. traffic to all ports until its MAC forwarding table is updated. Since
  1525. the interfaces attached to the bond may occupy multiple ports on a
  1526. single switch, when the switch (temporarily) floods the traffic to all
  1527. ports, the bond device receives multiple copies of the same packet
  1528. (one per slave device).
  1529. The duplicated packet behavior is switch dependent, some
  1530. switches exhibit this, and some do not. On switches that display this
  1531. behavior, it can be induced by clearing the MAC forwarding table (on
  1532. most Cisco switches, the privileged command "clear mac address-table
  1533. dynamic" will accomplish this).
  1534. 14. Hardware Specific Considerations
  1535. ====================================
  1536. This section contains additional information for configuring
  1537. bonding on specific hardware platforms, or for interfacing bonding
  1538. with particular switches or other devices.
  1539. 14.1 IBM BladeCenter
  1540. --------------------
  1541. This applies to the JS20 and similar systems.
  1542. On the JS20 blades, the bonding driver supports only
  1543. balance-rr, active-backup, balance-tlb and balance-alb modes. This is
  1544. largely due to the network topology inside the BladeCenter, detailed
  1545. below.
  1546. JS20 network adapter information
  1547. --------------------------------
  1548. All JS20s come with two Broadcom Gigabit Ethernet ports
  1549. integrated on the planar (that's "motherboard" in IBM-speak). In the
  1550. BladeCenter chassis, the eth0 port of all JS20 blades is hard wired to
  1551. I/O Module #1; similarly, all eth1 ports are wired to I/O Module #2.
  1552. An add-on Broadcom daughter card can be installed on a JS20 to provide
  1553. two more Gigabit Ethernet ports. These ports, eth2 and eth3, are
  1554. wired to I/O Modules 3 and 4, respectively.
  1555. Each I/O Module may contain either a switch or a passthrough
  1556. module (which allows ports to be directly connected to an external
  1557. switch). Some bonding modes require a specific BladeCenter internal
  1558. network topology in order to function; these are detailed below.
  1559. Additional BladeCenter-specific networking information can be
  1560. found in two IBM Redbooks (www.ibm.com/redbooks):
  1561. "IBM eServer BladeCenter Networking Options"
  1562. "IBM eServer BladeCenter Layer 2-7 Network Switching"
  1563. BladeCenter networking configuration
  1564. ------------------------------------
  1565. Because a BladeCenter can be configured in a very large number
  1566. of ways, this discussion will be confined to describing basic
  1567. configurations.
  1568. Normally, Ethernet Switch Modules (ESMs) are used in I/O
  1569. modules 1 and 2. In this configuration, the eth0 and eth1 ports of a
  1570. JS20 will be connected to different internal switches (in the
  1571. respective I/O modules).
  1572. A passthrough module (OPM or CPM, optical or copper,
  1573. passthrough module) connects the I/O module directly to an external
  1574. switch. By using PMs in I/O module #1 and #2, the eth0 and eth1
  1575. interfaces of a JS20 can be redirected to the outside world and
  1576. connected to a common external switch.
  1577. Depending upon the mix of ESMs and PMs, the network will
  1578. appear to bonding as either a single switch topology (all PMs) or as a
  1579. multiple switch topology (one or more ESMs, zero or more PMs). It is
  1580. also possible to connect ESMs together, resulting in a configuration
  1581. much like the example in "High Availability in a Multiple Switch
  1582. Topology," above.
  1583. Requirements for specific modes
  1584. -------------------------------
  1585. The balance-rr mode requires the use of passthrough modules
  1586. for devices in the bond, all connected to an common external switch.
  1587. That switch must be configured for "etherchannel" or "trunking" on the
  1588. appropriate ports, as is usual for balance-rr.
  1589. The balance-alb and balance-tlb modes will function with
  1590. either switch modules or passthrough modules (or a mix). The only
  1591. specific requirement for these modes is that all network interfaces
  1592. must be able to reach all destinations for traffic sent over the
  1593. bonding device (i.e., the network must converge at some point outside
  1594. the BladeCenter).
  1595. The active-backup mode has no additional requirements.
  1596. Link monitoring issues
  1597. ----------------------
  1598. When an Ethernet Switch Module is in place, only the ARP
  1599. monitor will reliably detect link loss to an external switch. This is
  1600. nothing unusual, but examination of the BladeCenter cabinet would
  1601. suggest that the "external" network ports are the ethernet ports for
  1602. the system, when it fact there is a switch between these "external"
  1603. ports and the devices on the JS20 system itself. The MII monitor is
  1604. only able to detect link failures between the ESM and the JS20 system.
  1605. When a passthrough module is in place, the MII monitor does
  1606. detect failures to the "external" port, which is then directly
  1607. connected to the JS20 system.
  1608. Other concerns
  1609. --------------
  1610. The Serial Over LAN (SoL) link is established over the primary
  1611. ethernet (eth0) only, therefore, any loss of link to eth0 will result
  1612. in losing your SoL connection. It will not fail over with other
  1613. network traffic, as the SoL system is beyond the control of the
  1614. bonding driver.
  1615. It may be desirable to disable spanning tree on the switch
  1616. (either the internal Ethernet Switch Module, or an external switch) to
  1617. avoid fail-over delay issues when using bonding.
  1618. 15. Frequently Asked Questions
  1619. ==============================
  1620. 1. Is it SMP safe?
  1621. Yes. The old 2.0.xx channel bonding patch was not SMP safe.
  1622. The new driver was designed to be SMP safe from the start.
  1623. 2. What type of cards will work with it?
  1624. Any Ethernet type cards (you can even mix cards - a Intel
  1625. EtherExpress PRO/100 and a 3com 3c905b, for example). For most modes,
  1626. devices need not be of the same speed.
  1627. 3. How many bonding devices can I have?
  1628. There is no limit.
  1629. 4. How many slaves can a bonding device have?
  1630. This is limited only by the number of network interfaces Linux
  1631. supports and/or the number of network cards you can place in your
  1632. system.
  1633. 5. What happens when a slave link dies?
  1634. If link monitoring is enabled, then the failing device will be
  1635. disabled. The active-backup mode will fail over to a backup link, and
  1636. other modes will ignore the failed link. The link will continue to be
  1637. monitored, and should it recover, it will rejoin the bond (in whatever
  1638. manner is appropriate for the mode). See the sections on High
  1639. Availability and the documentation for each mode for additional
  1640. information.
  1641. Link monitoring can be enabled via either the miimon or
  1642. arp_interval parameters (described in the module parameters section,
  1643. above). In general, miimon monitors the carrier state as sensed by
  1644. the underlying network device, and the arp monitor (arp_interval)
  1645. monitors connectivity to another host on the local network.
  1646. If no link monitoring is configured, the bonding driver will
  1647. be unable to detect link failures, and will assume that all links are
  1648. always available. This will likely result in lost packets, and a
  1649. resulting degradation of performance. The precise performance loss
  1650. depends upon the bonding mode and network configuration.
  1651. 6. Can bonding be used for High Availability?
  1652. Yes. See the section on High Availability for details.
  1653. 7. Which switches/systems does it work with?
  1654. The full answer to this depends upon the desired mode.
  1655. In the basic balance modes (balance-rr and balance-xor), it
  1656. works with any system that supports etherchannel (also called
  1657. trunking). Most managed switches currently available have such
  1658. support, and many unmanaged switches as well.
  1659. The advanced balance modes (balance-tlb and balance-alb) do
  1660. not have special switch requirements, but do need device drivers that
  1661. support specific features (described in the appropriate section under
  1662. module parameters, above).
  1663. In 802.3ad mode, it works with systems that support IEEE
  1664. 802.3ad Dynamic Link Aggregation. Most managed and many unmanaged
  1665. switches currently available support 802.3ad.
  1666. The active-backup mode should work with any Layer-II switch.
  1667. 8. Where does a bonding device get its MAC address from?
  1668. If not explicitly configured (with ifconfig or ip link), the
  1669. MAC address of the bonding device is taken from its first slave
  1670. device. This MAC address is then passed to all following slaves and
  1671. remains persistent (even if the first slave is removed) until the
  1672. bonding device is brought down or reconfigured.
  1673. If you wish to change the MAC address, you can set it with
  1674. ifconfig or ip link:
  1675. # ifconfig bond0 hw ether 00:11:22:33:44:55
  1676. # ip link set bond0 address 66:77:88:99:aa:bb
  1677. The MAC address can be also changed by bringing down/up the
  1678. device and then changing its slaves (or their order):
  1679. # ifconfig bond0 down ; modprobe -r bonding
  1680. # ifconfig bond0 .... up
  1681. # ifenslave bond0 eth...
  1682. This method will automatically take the address from the next
  1683. slave that is added.
  1684. To restore your slaves' MAC addresses, you need to detach them
  1685. from the bond (`ifenslave -d bond0 eth0'). The bonding driver will
  1686. then restore the MAC addresses that the slaves had before they were
  1687. enslaved.
  1688. 16. Resources and Links
  1689. =======================
  1690. The latest version of the bonding driver can be found in the latest
  1691. version of the linux kernel, found on http://kernel.org
  1692. The latest version of this document can be found in either the latest
  1693. kernel source (named Documentation/networking/bonding.txt), or on the
  1694. bonding sourceforge site:
  1695. http://www.sourceforge.net/projects/bonding
  1696. Discussions regarding the bonding driver take place primarily on the
  1697. bonding-devel mailing list, hosted at sourceforge.net. If you have
  1698. questions or problems, post them to the list. The list address is:
  1699. bonding-devel@lists.sourceforge.net
  1700. The administrative interface (to subscribe or unsubscribe) can
  1701. be found at:
  1702. https://lists.sourceforge.net/lists/listinfo/bonding-devel
  1703. Donald Becker's Ethernet Drivers and diag programs may be found at :
  1704. - http://www.scyld.com/network/
  1705. You will also find a lot of information regarding Ethernet, NWay, MII,
  1706. etc. at www.scyld.com.
  1707. -- END --