idnits 2.17.1 draft-ietf-ipoib-architecture-03.txt: Checking boilerplate required by RFC 5378 and the IETF Trust (see https://trustee.ietf.org/license-info): ---------------------------------------------------------------------------- ** Looks like you're using RFC 2026 boilerplate. This must be updated to follow RFC 3978/3979, as updated by RFC 4748. Checking nits according to https://www.ietf.org/id-info/1id-guidelines.txt: ---------------------------------------------------------------------------- ** The document seems to lack a 1id_guidelines paragraph about Internet-Drafts being working documents -- however, there's a paragraph with a matching beginning. Boilerplate error? ** The document seems to lack a 1id_guidelines paragraph about 6 months document validity -- however, there's a paragraph with a matching beginning. Boilerplate error? == No 'Intended status' indicated for this document; assuming Proposed Standard == The page length should not exceed 58 lines per page, but there was 23 longer pages, the longest (page 2) being 60 lines == It seems as if not all pages are separated by form feeds - found 0 form feeds but 24 pages Checking nits according to https://www.ietf.org/id-info/checklist : ---------------------------------------------------------------------------- ** The document seems to lack an IANA Considerations section. (See Section 2.2 of https://www.ietf.org/id-info/checklist for how to handle the case when there are no actions for IANA.) ** The document seems to lack separate sections for Informative/Normative References. All references will be assumed normative when checking for downward references. ** There are 6 instances of too long lines in the document, the longest one being 4 characters in excess of 72. Miscellaneous warnings: ---------------------------------------------------------------------------- == The copyright year in the RFC 3978 Section 5.4 Copyright Line does not match the current year -- The document seems to lack a disclaimer for pre-RFC5378 work, but may have content which was first submitted before 10 November 2008. If you have contacted all the original authors and they are all willing to grant the BCP78 rights to the IETF Trust, then this is fine, and you can ignore this comment. If not, you may need to add the pre-RFC5378 disclaimer. (See the Legal Provisions document at https://trustee.ietf.org/license-info for more information.) -- Couldn't find a document date in the document -- date freshness check skipped. Checking references for intended status: Proposed Standard ---------------------------------------------------------------------------- (See RFCs 3967 and 4897 for information about using normative references to lower-maturity documents in RFCs) -- Missing reference section? 'RFC2373' on line 240 looks like a reference -- Missing reference section? 'IANA' on line 986 looks like a reference Summary: 6 errors (**), 0 flaws (~~), 4 warnings (==), 4 comments (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 1 INTERNET DRAFT 2 Vivek Kashyap 3 Expiration Date: April, 2004 IBM 4 October, 2003 6 IP over InfiniBand(IPoIB) Architecture 8 Status of this memo 10 This document is an Internet-Draft and is in full conformance 11 with all provisions of Section 10 of RFC 2026. 13 Internet-Drafts are working documents of the Internet 14 Engineering Task Force (IETF), its areas, and its working 15 groups. Note that other groups may also distribute working 16 documents as Internet- Drafts. 18 Internet-Drafts are draft documents valid for a maximum of six 19 months and may be updated, replaced, or obsoleted by other 20 documents at any time. It is inappropriate to use 21 Internet-Drafts as Reference material or to cite them other 22 than as ``work in progress''. 24 The list of current Internet-Drafts can be accessed at 25 http://www.ietf.org/ietf/1id-abstracts.txt 27 The list of Internet-Draft Shadow Directories can be accessed 28 at http://www.ietf.org/shadow.html 30 This memo provides information for the Internet community. 31 This memo does not specify an Internet standard of any kind. 32 Distribution of this memo is unlimited. 34 Copyright Notice 36 Copyright (C) The Internet Society (2001). All Rights Reserved. 38 Abstract 40 InfiniBand is a high speed, channel based interconnect between 41 systems and devices. 43 This document presents an overview of the InfiniBand 44 architecture. It further describes the requirements and 45 guidelines for the transmission of IP over InfiniBand. 46 Discussions in this document are applicable to both IPv4 and 47 IPv6 unless explicitly specified. The encapsulation of IP over 48 InfiniBand and the mechanism for IP address resolution on IB 49 fabrics are covered in [IPOIB_ENCAP] and [IPOIB_DHCP]. 51 Table of Contents 53 1.0 Introduction to InfiniBand 54 1.1 InfiniBand Architecture Specification 55 1.2 Overview of InfiniBand Architecture 56 1.2.1 InfiniBand Addresses 57 1.2.1.1 Unicast GIDs 58 1.2.1.2 Multicast GIDs 59 1.3 InfiniBand Multicast Group Management 60 1.3.1 Multicast Member Record 61 1.3.1.1 JoinState 62 1.3.2 Join and Leave operations 63 1.3.2.1 Creating a Multicast Group 64 1.3.2.3 Deleting a Multicast Group 65 1.3.2.4 Multicast Group Create/Delete Traps 66 2.0 Management of InfiniBand Subnet 67 3.0 IP over IB 68 3.1 InfiniBand as Datalink 69 3.2 Multicast Support 70 3.2.1 Mapping IP Multicast to IB Multicast 71 3.2.2 Transient Flag in IB MGIDs 72 3.3 IP Subnet Across IB Subnets ? 73 4.0 IP Subnets in InfiniBand Fabrics 74 4.1 IPoIB VLANs 75 4.2 Multicast in IPoIB Subnets 76 4.2.1 Sending IP Multicast Datagrams 77 4.2.2 Receiving Multicast Packets 78 4.2.3 Forwarding Multicast Packets 79 4.2.4 Impact of InfiniBand Architecture Limits 80 4.2.5 Leaving/Deleting a Multicast Group 81 5.0 QoS and Related Issues 82 6.0 Security Considerations 83 7.0 Acknowledgements 84 8.0 References 85 9.0 Author's Address 87 1.0 Introduction to InfiniBand 89 The InfiniBand Trade Association(IBTA) was formed to develop 90 an I/O specification to deliver a channel based, switched 91 fabric technology. The InfiniBand standard is aimed at meeting 92 the requirements of scalability, reliability, availability and 93 performance of servers in data centers. 95 1.1 InfiniBand Architecture Specification 97 The InfiniBand Trade Association specification is available 98 for download from http://www.infinibandta.org. 100 1.2 Overview of InfiniBand Architecture 102 For a more complete overview the reader is referred to 103 chapter 3 of the InfiniBand specification. 105 InfiniBand Architecture (IBA) defines a System Area 106 Network(SAN) for connecting multiple independent processor 107 platforms, I/O platforms and I/O devices. The IBA SAN is a 108 communications and management infrastructure supporting both 109 I/O and inter-processor communications for one or more 110 computer systems. 112 An IBA SAN consists of processor nodes and I/O units connected 113 through an IBA fabric made up of cascaded switches and IB 114 routers (connecting IB subnets). I/O units can range in 115 complexity from single ASIC IBA attached devices such as a LAN 116 adapter to a large memory rich RAID subsystem. 118 An IBA network may be subdivided into subnets interconnected 119 by routers. These are IB routers and IB subnets and not IP 120 routers or IP subnets. This document will refer to InfiniBand 121 routers and subnets as 'IB routers' and 'IB subnets' 122 respectively. The IP routers and IP subnets will be referred 123 to as 'routers' and 'subnets' respectively. 125 Each IB node or switch may attach to a single or multiple 126 switches or directly with each other. Each IB unit interfaces 127 with the link by way of channel adapters (CAs). The 128 architecture supports multiple CAs per unit with each CA 129 providing one or more ports that connect to the fabric. Each 130 CA appears as a node to the fabric. 132 The ports are the endpoints to which the data is sent. 133 However, each of the ports may include multiple QPs (queue 134 pairs) that may be directly addressed from a remote peer. From 135 the point of view of data transfer the QP number (QPN) is part 136 of the address. 138 IBA supports both connection oriented and datagram service 139 between the ports. The peers are identified by QPN and the 140 port identifier. There are a two exceptions. QPNs are not used 141 when packets are multicast. QPNs are also not used in the raw 142 datagram mode. 144 A port, in a data packet, is identified by a local ID (LID) 145 and optionally a Global ID (GID). The GID in the packet is 146 needed only when communicating across an IB subnet though it 147 may always be included. 149 The GID is 128 bits long and is formed by the concatenation of 150 a 64 bit IB subnet prefix and a 64 bit EUI-64 compliant 151 portion (GUID). The LID is a 16 bit value that is assigned 152 when the port becomes active. Note that the GUID is the only 153 persistent identifier of a port. However, it cannot be used as 154 an address in a packet. If the prefix is modified then the GID 155 may change. The subnet manager may attempt to keep the LID 156 values constant across reboots but that is not a requirement. 158 The assignment of the GID and the LID is done by the subnet 159 manager. Every IB subnet has at least one subnet manager 160 component that controls the fabric. It assigns the LIDs and 161 GIDs. The subnet manager also programs the switches so that 162 they route packets between destinations. The subnet manager 163 and a related component, the subnet administrator (SA) are the 164 central repository of all information that is required to 165 setup and bring up the fabric. 167 IB routers are components that route packets between IB 168 subnets based on the GIDs. Thus within an IB subnet a packet 169 may or may not include a GID but when going across an IB 170 subnet the GID must be included. A LID is always needed in a 171 packet since the destination within a subnet is determined by 172 it. 174 A CA and a switch may have multiple ports. Each CA port is 175 assigned its own LID or a range of LIDs. The ports of a switch 176 are not addressable by LIDs/GIDs or in other words, are 177 transparent to other end nodes. Each port has its own set of 178 buffers. The buffering is channeled through virtual lanes(VL) 179 where each VL has its own flow control. There may be up to 16 180 VLs. 182 VLs provide a mechanism for creating multiple virtual links 183 within a single physical link. All ports must support VL15 184 which is reserved exclusively for subnet management datagrams 185 and hence doesn't concern the IPoIB discussions. The actual VL 186 that a packet uses is configured by the SM in the 187 switch/channel adapter tables and is determined based on the 188 Service Level (SL) specified in every packet. There are 16 189 possible SLs. 191 In addition to the features described above viz. Queue 192 Pairs(QPs), Service Levels(SLs) and addressing(GID/LID), IBA 193 also defines the following: 195 Partitioning: 197 Every packet, but for the raw datagrams, carries the 198 partition key (P_key). These values are used for 199 isolation in the fabric. A switch (this is an optional 200 feature) may be programmed by the SM to drop packets 201 not having a certain key. The CA ports always check 202 for the P_Keys. A CA port may belong to multiple 203 partitions. P_Key checking is optional at IB routers. 205 A P_Key may be described as having 'limited 206 membership' or 'full membership'. For a packet to be 207 accepted at least one of the P_Keys i.e. the P_Key in 208 the packet or the P_Key in the port, must be 'full 209 membership' P_Keys. 211 Q_Keys: 213 Q_Keys are used to enforce access rights for reliable 214 and unreliable IB datagram services. Raw datagram 215 services don't use Q_Keys. At communication 216 establishment the endpoints exchange the Q_Keys and 217 must always use the relevant Q_Keys when communicating 218 with one another. Multicast packets use the Q_Key 219 associated with the multicast group. 221 Q_Keys with the most significant bit set are 222 considered controlled Q_Keys (such as the GSI Q_Key) 223 and a HCA does not allow a consumer to arbitrarily 224 specify a controlled Q_Key. An attempt to send a 225 controlled Q_Key results in using the Q_Key in the QP 226 context. Thus the OS maintains control since it can 227 configure the QP context for the controlled Q_Key for 228 privileged consumers. It must be noted that though the 229 notion of a 'controlled Q_Key' is suggested by IB 230 specification it does not require its use or 231 implementation. 233 Multicast support: 235 A switch may support multicasting i.e. replication of 236 packets across multiple output ports. This is an 237 optional feature. Similarly, support for 238 sending/receiving multicast packets is optional in 239 CAs. A multicast group is identified by a GID. The GID 240 format is as defined in [RFC2373] on IPv6 addressing. 241 Thus from an IPv6 over InfiniBand's point of view the 242 data link multicast address looks like the network 243 address. An IB port must explicitly join a multicast 244 group by sending a request to the SM to receive 245 multicast packets. A port may send packets to any 246 multicast group. In both cases the multicast LID to be 247 used in the packets is received from the SM. 249 There are 6 methods for data transfer in IB architecture. 250 These are : 252 1. Unreliable Datagram (unacknowledged - connectionless) 254 The UD service is connectionless and unacknowledged. 255 It allows the QP to communicate with any unreliable 256 datagram QP on any node. 258 The switches and hence each link can support only a 259 certain MTU. The MTU ranges are 256 bytes, 512 bytes, 260 1024 bytes, 2048 bytes, 4096 bytes. A UD packet cannot 261 be larger than the smallest link MTU between the two 262 peers. 264 2. Reliable Datagram (acknowledged - multiplexed) 266 The RD service is multiplexed over connections between 267 nodes called End to end contexts (EEC) which allows 268 each RD QP to communicate with any RD QP on any node 269 with an established EEC. Multiple QPs can use the same 270 EEC and a single QP can use multiple EECs (one for 271 each remote node per reliable datagram domain). 273 3. Reliable Connected (acknowledged - connection oriented) 275 The RC service associates a local QP with one and only 276 one remote QP. The message sizes maybe as large as 277 2^31 bytes in length. The CA implementation takes care 278 of segmentation and assembly. 280 4. Unreliable Connected (unacknowledged - connection oriented) 282 The UC service associates one local QP with one and 283 only one remote QP. There is no acknowledgment and 284 hence no resend of lost or corrupted packets. Such 285 packets are therefore simply dropped. It is similar to 286 RC otherwise. 288 5. Raw Ethertype (unacknowledged - connectionless) 290 The Ethertype raw datagram packet contains a generic 291 transport header that is not interpreted by the CA but 292 it specifies the protocol type. The values for 293 ethertype are the same as defined in RFC1700 for 294 ethertype. 296 6. Raw IPv6 ( unacknowledged - connectionless) 298 Using IPv6 raw datagram service, the IBA CA can 299 support standard protocol layers atop IPv6 (such as 300 TCP/UDP). Thus native IPv6 packets can be bridged into 301 the IBA SAN and delivered directly to a port and to 302 its IPv6 raw datagram QP. 304 The first 4 types are referred to as IB transports. The latter 305 two are classified as Raw datagrams. There is no indication of 306 the QP number in the raw datagram packets. The raw datagram 307 packets are limited by the link MTU in size. 309 The two connected modes and the reliable datagram mode may 310 also support 'Automatic Path Migration(APM)'. This is an 311 optional facility that provides for a hardware based path 312 failover. An alternate path is associated with the QP when the 313 connection/EE context is first created. If unrecoverable 314 errors are encountered the connection switches to using the 315 alternate path. 317 1.2.1 InfiniBand Addresses 319 The InfiniBand architecture borrows heavily from the IPv6 320 architecture in terms of the InfiniBand subnet structure and 321 global identifiers (GIDs). 323 The InfiniBand architecture defines the global identifier 324 associated with a port as follows: 326 GID (Global Identifier): A 128-bit unicast or 327 multicast identifier used to identify a port on a 328 channel adapter, a port on a router, a switch, or a 329 multicast group. A GID is a valid 128-bit IPv6 330 address(per RFC 2373) with additional 331 properties/restrictions defined within IBA to 332 facilitate efficient discovery, communication, and 333 routing. 335 Note: These rules apply only to IBA operation and do 336 not apply to raw IPv6 operation unless specifically 337 called out. 339 The raw IPv6 operation referred to in the note 340 above is the IPv6 mode of InfiniBand's raw datagram 341 service. It does not mean IPv6 itself. The routers and 342 switches referred to in the above definition are the 343 InfiniBand routers and switches. 345 The InfiniBand(IB) specification defines two types of GIDs: 346 unicast and multicast. 348 1.2.1.1 Unicast GIDs 350 The unicast GIDs are defined, as in IPv6, with three scopes. 351 The IB specification states: 353 a. link local: This is defined to be FE80/10. 355 The IB routers will not forward packets with a 356 link local address in source or destination 357 beyond the IB subnet. 359 b. site local: FEC0/10 361 A unicast GID used within a collection of 362 subnets which is unique within that collection 363 (e.g. a data center or campus) but is not 364 necessarily globally unique. IB routers must 365 not forward any packets with either a 366 site-local Source GID or a site-local 367 Destination GID outside of the site. 369 c. global: 370 A unicast GID with a global prefix, i.e. an IB 371 router may use this GID to route packets 372 throughout an enterprise or internet. 374 1.2.1.2 Multicast GIDs 376 The multicast GIDs also parallel the IPv6 multicast addresses. 377 The IB specification defines the multicast GIDs as follows: 379 FFxy:<112 bits> 381 Flag bits: 383 The nibble, denoted by x above, are the 4 flag bits: 000T. 385 The first three bits are reserved and are set to zero. The 386 last bit is defined as follows: 388 T=0: denotes a permanently assigned i.e. well known GID 389 T=1: denotes a transient group 391 Scope bits: 393 The 4 bits, denoted by y in the GID above, are the scope 394 bits. These scope values are described in Table 1. 396 scope value Address value 398 0 Reserved 399 1 Unassigned 400 2 Link-local 401 3 Unassigned 402 4 Unassigned 403 5 Site-local 404 6 Unassigned 405 7 Unassigned 406 8 Organization-local 407 9 Unassigned 408 0xA Unassigned 409 0xB Unassigned 410 0xC Unassigned 411 0xD Unassigned 412 0xE Global 413 0xF Reserved 415 Table 1 417 The IB specification further refers to [RFC_2373] and 418 [RFC_2375] while defining the well known multicast addresses. 419 However, it then states that the well known addresses apply to 420 IB raw IPv6 datagrams only. It must be noted though that a 421 multicast group can be associated with only a single MGID. 422 Thus the same MGID cannot be associated with the UD mode and 423 the raw datagram mode. 425 1.3 InfiniBand Multicast Group Management 427 IB multicast groups, identified by Multicast Global 428 Identifiers (MGIDs), are managed by the subnet manager(SM). 429 The SM explicitly programs the IB switches in the fabric to 430 ensure that the packets are received by all the members of the 431 multicast group that request the reception of packets. SM also 432 needs to program the switches such that packets transmitted to 433 the group by any group member reach all receivers in the 434 multicast group. 436 IBA distinguishes between multicast senders and receivers. 437 Though all members of a multicast group can transmit to the 438 group (and expect their packets to be correctly forwarded) not 439 all members of the group are receivers. A port needs to 440 explicitly request that multicast packets addressed to the 441 group be forwarded to it. 443 A multicast group is created by sending a join request to the 444 SM. As will be explained later, IBA defines multiple modes for 445 joining a multicast group. The subnet manager records the 446 group's multicast GID and the associated characteristics. The 447 group characteristics are defined by the group path MTU, 448 whether the group will be used for raw datagrams or unreliable 449 datagrams, the service level, the partition key associated 450 with the group, the Local Identifier(LID) associated with the 451 group etc. These characteristics are defined at the time of 452 the group creation. The interested reader may lookup the 453 'MCMemberRecord' attribute in the IB architecture 454 specification[IB_ARCH] for the complete list of 455 characteristics that define a group. 457 A LID is associated with the multicast group by the subnet 458 manager(SM) at the time of the multicast group creation. The 459 SM determines the multicast tree based on all the group 460 members and programs the relevant switches. The Multicast 461 LID(MLID) is used by the switches to route the packets. 463 Any member IB port wanting to participate in the multicast 464 group must join the group. As part of the join operation the 465 port receives the group characteristics from the SM. At the 466 same time the subnet manager ensures that the requester can 467 indeed participate in the group by verifying that it can 468 support the group MTU, and accessibility to the rest of the 469 group members. Other group characteristics may need 470 verification too. 472 The SM, for groups that span IB subnet boundaries, must 473 interact with IB routers to determine the presence of this 474 group in other IB subnets. If present the MTU must match 475 across the IB subnets. 477 P_Key is another characteristic that must match across IB 478 subnets since the P_Key inserted into a packet is not modified 479 by the IB switches or IB routers. Thus if the P_Keys didn't 480 match the IB router(s) itself might drop the packets or 481 destinations on other subnets might drop the packets. 483 A join operation may cause the SM to reprogram the fabric so 484 that the new member can participate in the multicast group. By 485 the same token a leave may cause the SM to reprogram the 486 fabric to stop forwarding the packets to the requester. 488 1.3.1 Multicast Member Record 490 The multicast group is maintained by the SM with each of the 491 group members represented by an MCMemberRecord[IB_ARCH]. Some 492 of its components are: 494 MGID - Multicast GID for this multicast group 495 PortGID - Valid GID of the port joining this multicast group 496 Q_Key - Q_Key to be used by this multicast group 497 MLID - Multicast LID for this multicast group 498 MTU - MTU for this multicast group 499 P_Key - Partition key for this multicast group 500 SL - Service Level for this multicast group 501 Scope - Same as MGID address scope 502 JoinState - Join/Leave status requested by the port: 503 bit 0: FullMemeber 504 bit 1: NonMember 505 bit 2: SendOnlyNonMember 507 1.3.1.1 JoinState 509 The JoinState indicates the membership qualities a port wishes 510 to add while joining/creating a group or delete when leaving a 511 group. The meaning of the JoinState bits are: 513 FullMember: 514 Messages destined for the group are routed to and from 515 the port. A group may be deleted by the SM if there 516 are no FullMembers in the group. 518 NonMember: 519 Messages destined for the group are routed to and from 520 the port. The port is not considered a member for 521 purposes of group creation/deletion. 523 SendOnlyNonMember: 524 Group messages are only routed from the port but not 525 to the port. The port is not considered a member for 526 purposes of group creation/deletion. 528 A port may have multiple bits set in its record. In such case 529 the membership qualities are a union of the JoinStates. A port 530 may leave the multicast group for each of the JoinStates 531 individually or in any combination of JoinState 532 bits[IB_ARCH]. 534 1.3.2 Join and Leave Operations 536 An IB port joins a multicast group by sending a join 537 request(SubnAdmSet() method) and leaves a multicast group by 538 sending a leave message (SubnAdmDelete() method) to the SM. 539 The IBA specification[IB_ARCH] describes the methods and 540 attributes to be used when sending these messages. 542 1.3.2.1 Creating a Multicast Group 544 There is no 'create' command to form a new multicast group. 545 The FullMember bit in the JoinState must be set to create a 546 multicast group. In other words, the first FullMember join 547 request will cause the group to be created as a side effect of 548 the join request. Subsequent join or leave requests may 549 contain any combination of the JoinState bits. 551 The creator of the group specifies the Q_Key, MTU, P_Key, SL, 552 FlowLabel, TClass and the Scope value. A creator may request 553 that a suitable MGID be created for it. Alternatively, the 554 request can specify the desired MGID. In both cases the MLID 555 is assigned by the SM. 557 Thus a group will be created with the specified values when 558 the requester sets the FullMember bit and no such group 559 already exists in the subnet. 561 1.3.2.3 Deleting a Multicast Group 563 When the last FullMember leaves the multicast group the SM may 564 delete the multicast group releasing all resources, including 565 those that might exist in the fabric itself, associated with 566 the group. 568 Note that a special 'delete' message does not exist. It is a 569 side effect of the last FullMember 'leave' operation. 571 1.3.2.4 Multicast Group Create/Delete Traps 573 The SA may be requested by the ports to generate a report 574 whenever a multicast group is created or deleted. The port can 575 specify the multicast group it is interested in i.e. use a 576 specific MGID or use a wildcard request. The SA will report 577 these events using traps 66 (for creates) and 67 (for 578 deletes)[IB_ARCH]. 580 Therefore, a port wishing to join a group but not create it by 581 itself may request a create notification or a port might even 582 request a notification for all groups that are created(a 583 wildcarded request). The SA will diligently inform them of the 584 creation utilising the aforementioned traps. The requestor can 585 then join the multicast group indicated. Similarly, a 586 SendOnlyNonMember or a NonMember might request the SA to 587 inform it of group deletions. The endnode, on receiving a 588 delete report, can safely release the resources associated 589 with the group. The associated MLID is no longer valid for the 590 group and may be reassigned to a new multicast group by the 591 SM. 593 2.0 Management of InfiniBand Subnet 595 To aid in the monitoring and configuration of InfiniBand 596 subnet components a set of MIBs need to be defined. MIBs are 597 needed for the channel adapters, InfiniBand interfaces, 598 InfiniBand subnet manager, InfiniBand subnet management agents 599 and to allow the management of specific device properties. It 600 must be noted that the management objects addressed in the 601 IPoIB documents are for all of the IB subnet components and 602 are not limited to IP(over IB). The relevant MIBs are 603 described in separate documents and are not covered here. 605 3.0 IP over IB 607 As described in section 1.0, the InfiniBand architecture 608 provides a broad set of capabilities to choose from when 609 implementing IP over InfiniBand networks. 611 The IPoIB specification must not, and does not, require 612 changes in IP and higher layer protocols. Nor does it mandate 613 requirements on IP stacks to implement special user level 614 programs. It is an aim of IPoIB specification that the IPoIB 615 changes be amenable to modularisation and incorporation into 616 existing implementations at the same level as other media 617 types. 619 3.1 InfiniBand as Datalink 621 InfiniBand architecture provides multiple methods of data 622 exchange between two endpoints as was noted above. These are: 624 Reliable Connected (RC) 625 Reliable Datagram (RD) 626 Unreliable Connected (UC) 627 Unreliable Datagram (UD) 628 Raw Datagram : Raw IPv6 (R6) 629 : Raw Ethertype (RE) 631 IPoIB can be implemented over any, multiple or all of these 632 services. A case can be made for support on any of the 633 transport methods depending on the desired features. 635 The IB specification requires Unreliable Datagram mode to be 636 supported by all the IB nodes. The host channel adapters(HCAs) 637 are specifically required to support Reliable connected(RC) 638 and Unreliable connected(UC) modes but the same is not the 639 case with target channel adapters(TCAs). Support for the two 640 Raw Datagram modes is entirely optional. The Raw Datagram mode 641 supports a 16-bit CRC as against the better protection 642 provided by the use of a 32-bit CRC in other modes. 644 For the sake of simplicity, ease of implementation and 645 integration with existing stacks, it is desirable that the 646 fabric support multicasting. This is possible only in 647 Unreliable datagram (UD) and IB's Raw datagram modes. 649 Thus it is only the UD mode that is universal, supports 650 multicast, and a robust CRC. Given these conditions it is the 651 obvious choice for IP over InfiniBand [IPOIB_ENCAP]. 653 Future documents might consider the connected modes. In 654 contrast to the limited link MTU offered by UD mode, the 655 connected modes can offer significant benefit in terms of 656 performance by utilising a larger MTU. Reliability is also 657 enhanced if the underlying feature of automatic path migration 658 of connected modes is utilised. 660 3.2 Multicast Support 662 InfiniBand specification makes support of multicasting in the 663 switches optional. Multicast however, is a basic requirement 664 in IP networks. Therefore, IPoIB requires that multicast 665 capable InfiniBand fabrics be used to implement IPoIB 666 subnets. 668 3.2.1 Mapping IP Multicast to IB Multicast 670 Well known IP multicast groups are defined for both IPv4 and 671 IPv6 (RFC_1700, RFC_2373). Multicast groups may also be 672 dynamically created at any time. To avoid creating unnecessary 673 duplicates of multicast packets in the fabric, and to avoid 674 unnecessary handling of such packets at the hosts each of the 675 IP multicast groups needs to be associated with a different IB 676 multicast group as far as possible. A process is defined in 677 [IPOIB_ENCAP] for mapping the IP multicast addresses to unique 678 IB multicast addresses. 680 3.2.2 Transient Flag in IB MGIDs 682 The IB specification describes the flag bits as discussed in 683 section 1.3. The IB specification also defines some well known 684 IB multicast GIDs(MGIDs). The MGIDs are reserved for the IB's 685 Raw datagram mode which is incompatible with the other 686 transports of IB. Any mapping that is defined from IP 687 multicast addresses therefore must not fall into IB's 688 definition of a well-known address. 690 Therefore all IPoIB related multicast GIDs always set the 691 transient bit. 693 3.3 IP Subnets Across IB Subnets ? 695 Some implementations may wish to support multiple clusters of 696 machines in their own IB subnets but otherwise be part of a 697 common IP subnet. For such a solution the IB specification 698 needs multiple upgrades. Some of the required enhancements 699 are: 701 1) A method for creating IB multicast GIDs that span multiple 702 IB subnets. The partition keys and other parameters need to 703 be consistent across IB subnets. 705 2) Develop IB routing protocol to determine the IB topology 706 across IB subnets. 708 3) Define the process and protocols needed between IB nodes 709 and IB routers 711 Until the above conditions are met it is not possible to 712 implement IPoIB subnets that span IB subnets. The IPoIB 713 standards have however been defined with this possibility in 714 mind. 716 4.0 IP Subnets in InfiniBand Fabrics 718 The IPoIB subnet is overlaid over the IB subnet. The IPoIB 719 subnet is brought up in the following steps: 721 Note: the join/leave operation at the IP level will be 722 referred to as IP_join/IP_leave and the join/leave 723 operations at the IB level will be referred to as 724 IB_join in this document. 726 1. The all-IPoIB nodes IB multicast group is created 728 The fabric administrator creates an IB multicast 729 group(henceforth called 'broadcast group') when the IP subnet 730 is setup. The 'broadcast group' is defined in [IPOIB_ENCAP]. 731 The method by which the broadcast group is setup is not 732 defined by IPoIB. The group may be setup at the SM by the 733 administrator or by the first IB_join. 735 As noted earlier, at the time of creating an IB multicast 736 group, multiple values such as the P_Key, Q_Key, Service 737 Level, Hop Limit, Flow ID, TClass, MTU etc., have to be 738 specified. These values should be such that all potential 739 members of the IB multicast group are be able to communicate 740 with one another when using them. In the future, as the IB 741 specification associates more meaning with the various 742 parameters and defines IB QoS, different values for IP 743 multicast traffic may be possible. All unicast packets also 744 need to use the P_Key and Q_Key specified in the broadcast 745 group [IPOIB_ENCAP]. It is obvious that a thought out 746 configuration is required for a successful setup of the IPoIB 747 subnet. 749 2. All IPoIB interfaces IB_join the broadcast group 751 The broadcast group defines the span and the members of the 752 IPoIB link. This link gets built up as IPoIB nodes IB_join the 753 broadcast group. 755 The IB_join to the broadcast group has the additional benefit 756 of distributing the above mentioned multicast group parameters 757 to all the members of the subnet. 759 Note that this IB_join to the broadcast group is a FullMember 760 join. If any of the ports or the switches linking the port to 761 the rest of the IPoIB subnet cannot support the 762 parameters(e.g. path MTU or P_Key) associated with the 763 broadcast group, then the IB_join request will fail and the 764 requesting port will not become part of the IPoIB subnet. 766 3. Configuration Parameters 768 As noted above, parameters such as, Q_Key, Path MTU, needed 769 for all IPoIB communication are returned to the IPoIB node on 770 IB_joining the 'broadcast group'. [IPOIB_ENCAP] also notes 771 that the parameters used in the broadcast group are used when 772 creating other multicast groups. 774 However, the P_Key must still be known to the IPoIB endnode 775 before it can join the broadcast-group. The P_Key is included 776 in the mapping of the broadcast group[IPOIB_ENCAP]. Another 777 parameter, the scope of the broadcast group, also needs to be 778 known to the endnode before it can join the broadcast group. 780 It is an implementation choice on how the P_Key and the scope 781 bits related to the IPoIB subnet are determined by the 782 implementation. These could be configuration parameters 783 initialised by some means by the administrator. 785 The methods employed by an implementation to determine the 786 P_Key and scope bits are not specified by IPoIB. 788 4.1 IPoIB VLANs 790 The endpoints in an IB subnet must have compatible P_Keys to 791 communicate with one another. Thus the administrator when 792 setting up an IP subnet over an IB subnet must ensure that all 793 the members have compatible P_Keys. An IP subnet can have only 794 one P_Key associated with it to ensure that all IP nodes in it 795 can talk to one another. An endpoint may however have multiple 796 P_Keys. 798 The IB architecture specifies that there can be only one MGID 799 associated with a multicast group in the IB subnet. The P_Key 800 is included in the MGID mappings from the IP multicast 801 addresses[IPOIB_ENCAP]. Since the P_Key is unique in the IB 802 subnet the inclusion of the P_Key in the IB MGIDs ensures that 803 unique MGID mappings are created. Every unique broadcast group 804 MGID so formed creates a separate abstract IPoIB link and 805 hence an IPoIB VLAN. 807 4.2 Multicast in IPoIB subnets 809 IP multicast on InfiniBand subnets follows the same concepts 810 and rules as on any other media. However, unlike most other 811 media multicast over InfiniBand requires interaction with 812 another entity, the IB subnet manager. This section describes 813 the outline of the process and suggests some guidelines. 815 IB architecture specifies the following format for IB 816 multicast packets when used over unreliable datagram(UD) 817 mode: 819 +--------+-------+---------+---------+-------+---------+---------+ 820 |Local |Global |Base |Datagram |Packet |Invariant| Variant | 821 |Routing |Routing|Transport|Extended |Payload| CRC | CRC | 822 |Header |Header |Header |Transport| (IP) | | | 823 | | | |Header | | | | 824 +--------+-------+---------+---------+-------+---------+---------+ 826 For details about the various headers please refer to 827 InfiniBand Architecture Specification[IB_ARCH]. 829 The Global routing header (GRH) includes the IB multicast 830 group GID. The Local routing header (LRH) includes the local 831 identifier (LID). The IB switches in the fabric route the 832 packet based on the LID. 834 The GID is made available to the receiving IB user (the IPoIB 835 interface driver for example). The driver can therefore 836 determine the IB group the packet belongs to. 838 IPv4 defines three levels of multicast compliance. These are: 840 Level 0: No support for IP multicasting 842 Level 1: Support for sending but not receiving multicasts 844 Level 2: Full support for IP multicasting 846 In IPv6 there is no such distinction. Full multicast support 847 is mandatory. Additionally, all IPv4 subnets support 848 broadcast(255.255.255.255). IPv4 broadcast can always be 849 sent/received by all IPv4 interfaces. 851 Every IPoIB subnet requires the broadcast GID to be defined. 853 Thus a packet can always be broadcast. 855 4.2.1 Sending IP Multicast Datagrams 857 An IP host may send a multicast packet at any time to any 858 multicast address. 860 The IP layer conveys the multicast packet to the IPoIB 861 interface driver/module. This module attempts to IB_join the 862 relevant IB multicast group. This is required since otherwise 863 InfiniBand architecture does not guarantee that the packet 864 will reach its destinations. 866 A pure sender may choose to join the multicast group as a 867 FullMember. In such a case the sender will receive all the 868 multicast packets transmitted to the IB group. Additionally, 869 the IB group will not be deleted until the sender leaves the 870 group. 872 Alternatively, a sender might IB_join as a SendOnlyNonMember. 873 In such a case the packets are not routed to the sender though 874 packets transmitted by it can reach the other group members. 875 Additionally, the group can be deleted when all FullMembers 876 have left the group. The sender can further request delete 877 updates from the SM. 879 If the sender does not find the group in existence it is 880 recommended in [IPOIB_ENCAP] that the packets be sent to the 881 MGID corresponding to the all-IP routers address. A sender 882 could also send the packets to the broadcast group. The 883 sender might also choose to request 'creation' reports from 884 the SM. 886 4.2.2 Receiving Multicast Packets 888 The IP host must join the IB multicast group corresponding to 889 the IP address. This follows from the IBA requirement that the 890 receiver must join the relevant IB multicast group. The group 891 is automatically created if it does not exist [IB_ARCH]. 893 The IP receivers must IB_leave the IB group when the IP layer 894 stops listening of the corresponding IP address. The SM can 895 then choose to delete the group. 897 4.2.3 Router considerations for IPoIB 899 IP routers know of the new IP groups created in the subnet by 900 the use of protocols such as IGMP/MLD. However, this is not 901 enough for IPoIB since the router needs to IB_join the 902 relevant IB groups to be able to receive and transmit the 903 packets. There is no promiscuous mode for listening to all 904 packets. 906 The IPoIB routers therefore need to request the SM to report 907 all creations of IB groups in the fabric. The IPoIB router can 908 then IB_join the reported group. It is not desirable that the 909 router's IB_joining of a multicast group be considered the 910 same as the IB_join from a receiver - the router's IB_join 911 shouldn't disallow the group's deletion when all receivers 912 leave. To overcome just this type of situations, IBA provides 913 the NonMember IB_join mode. 915 The NonMember IB_join mode can be used by IP routers when they 916 join in response to the create reports. A router should 917 ideally request the delete reports too so that it can release 918 all the resources associated with the group. The MLID 919 associated with a deleted MGID can be reassigned by the SM and 920 therefore there is a possibility of erroneous transmissions if 921 the MLID is cached. A router that does not request delete 922 reports will still work correctly since it will receive the 923 correct MLID , and purge any old cached value, when it 924 IB_joins the IB group in response to a create report. 926 It is reasonable for a router to IB_join as a FullMember if it 927 is joining the IB group in response to an application/routing 928 daemon request. In such a case the router might end up 929 controlling the existence of the IB group (since it is a 930 FullMember of the group). 932 4.2.4 Impact of InfiniBand Architecture Limits 934 An HCA or TCA may have a limit on the number of MGIDs it can 935 support. Thus, even though the groups may not be limited at 936 the subnet manager and in the subnet as such, they may be 937 limited at a particular interface. It is advisable to choose 938 an adequately provisioned HCA/TCA when setting up an IPoIB 939 subnet. 941 4.2.5 Leaving/Deleting a Multicast Group 943 An IPv4 sender (level 1 compliance) IB_joins the IB multicast 944 group only because that is the only way to guarantee reception 945 of the packets by all the group recipients. The sender must 946 however IB_leave the group at some time. A sender could, when 947 not a receiver on the group, start a timer per multicast group 948 sent to. The sender leaves the IB group when the timer goes 949 off. It restarts the timer if another message is sent. 951 This suggestion doesn't apply to the IB broadcast group. It 952 also doesn't apply to the IB group corresponding to the 953 all-hosts multicast group. An IPv4 host must always remain a 954 member of the broadcast group. 956 An IP multicast receiver IB_leaves the corresponding IB 957 multicast group when it IP_leaves the IP multicast group. In 958 the case of IPv4 implementation the receiver may choose to 959 continue to be a sender (level 1 compliance). In which case it 960 may choose not to IB_leave the IB group but start a timer as 961 explained above. 963 As noted elsewhere, the SM can choose to free up the 964 resources(e.g. routing entries in the switches) associated 965 with the IB group when the last FullMember IB_leave the group. 966 The MLID therefore becomes invalid for the group. The MLID can 967 be reassigned when a new group is created. 969 SendOnlyNonMember/NonMember ports caching the MLID need to 970 avoid this possibility. The way out is for them to request 971 group delete reports. An IP router requesting reports for all 972 groups need not request the delete report since an IB_join in 973 response to a create report will return the new MLID 974 association to it. 976 A router might prefer to IB_leave the IB multicast group when 977 there are no members of the IP multicast address in the subnet 978 and it has no explicit knowledge of any need to forward such 979 packets. 981 4.3 Transmission of IPoIB packets 983 The encapsulation of IP packets in InfiniBand is described 984 in[IPOIB_ENCAP]. 986 It specifies the use of an 'Ethertype' value [IANA] in all 987 IPoIB communication packets. The link-layer address is 988 comprised of the Global Identifier(GID) and the Queue Pair 989 Number(QPN) [IPOIB_ENCAP]. 991 To allow for multiple IB subnet based IPoIB subnets, the 992 specification utilises the Global Identifier(GID) as part of 993 the link-layer address. Since all packets in IB have to use 994 the Local Identifier(LID) the address resolution process has 995 the additional step of resolving the destination GID, returned 996 in response to ARP/ND request, to the LID[IPOIB_ENCAP]. This 997 phase of address resolution might also be used to determine 998 other essential parameters (e.g. the SL, path rate etc.)for 999 successful IB communication between two peers. 1001 As noted earlier, all communication in the IPoIB subnet 1002 derives the Q_Key to use from the Q_Key specified in the 1003 broadcast group. 1005 4.4 RARP and Static ARP entries 1007 RARP entries or static ARP entries are based on invariant 1008 link-addresses. In the case of IPoIB, the link-address 1009 includes the QPN which might not be constant across reboots or 1010 even across network interface resets. Therefore, static ARP 1011 entries or RARP server entries will only work if the 1012 implementation(s) using these options can ensure that the QPN 1013 associated with an interface is invariant across 1014 reboots/network resets[IPOIB_ENCAP]. 1016 4.5 DHCPv4 and IPoIB 1018 DHCPv4 [RFC_2131] utilises a 'client identifier' field 1019 (expected to hold the link-layer address) of 16 bytes. The 1020 address in the case of IPoIB is 20-bytes. To get around this 1021 problem IPoIB specifies [IPOIB_DHCP] that the 'broadcast flag' 1022 be used by the client when requesting an IP address. 1024 5.0 QoS and Related Issues 1026 The IB specification suggests the use of service levels for 1027 load balancing, QoS and deadlock avoidance within an IB 1028 subnet. But the IB specification leaves the usage and mode of 1029 determination of the SL for the application to decide. The SL 1030 and list of SLs are available in the SA but it is up to the 1031 endnode's application to choose the 'right' value. 1033 Every IPoIB implementation will determine the relevant SL 1034 value based on its own policy. No method or process for 1035 choosing the SL has been defined by the IPoIB standards. 1037 6.0 Security Considerations 1039 This document describes the IB architecture as relevant to 1040 IPoIB. It further restates issues specified in other 1041 documents. It does not itself specify any requirements. There 1042 are no security issues introduced by this document. IPoIB 1043 related security issues are described in 1044 [IPOIB_ENCAP] and [IPOIB_DHCP]. 1046 7.0 Acknowledgements 1048 This document has benefited from the comments and suggestion 1049 of the members of the IPoIB working group and the members of 1050 the InfiniBand(SM) Trade Association. 1052 8.0 References 1054 [IB_ARCH] InfiniBand Architecture Specification, Volume 1.1 1056 [RFC_2373] IP Version 6 Addressing Architecture 1058 [RFC_2375] IPv6 Multicast Address Assignments 1060 [RFC_1700] Assigned Numbers 1062 [RFC_1112] Host extensions for IP multicasting 1064 [RFC_2236] Internet Group Management Protocol, Version 2 1066 [RFC_2710] Multicast Listener Discovery 1068 [IPOIB_ENCAP] draft-ietf-ipoib-ip-over-infiniband-05.txt 1070 [IPOIB_DHCP] draft-ietf-ipoib-dhcp-over-infiniband-05.txt 1072 9.0 Author's Address 1074 Vivek Kashyap 1076 IBM 1077 15450, SW Koll Parkway 1078 Beaverton, OR 97006 1080 Phone: +1 503 578 3422 1081 Email: vivk@us.ibm.com 1083 Full Copyright Statement 1085 Copyright (C) The Internet Society (2001). All Rights Reserved. 1087 This document and translations of it may be copied and 1088 furnished to others, and derivative works that comment on or 1089 otherwise explain it or assist in its implementation may be 1090 prepared, copied, published and distributed, in whole or in 1091 part, without restriction of any kind, provided that the above 1092 copyright notice and this paragraph are included on all such 1093 copies and derivative works. However, this document itself may 1094 not be modified in any way, such as by removing the copyright 1095 notice or references to the Internet Society or other Internet 1096 organizations, except as needed for the purpose of developing 1097 Internet standards in which case the procedures for copyrights 1098 defined in the Internet Standards process must be followed, or 1099 as required to translate it into languages other than 1100 English. 1102 The limited permissions granted above are perpetual and will 1103 not be revoked by the Internet Society or its successors or 1104 assigns. 1106 This document and the information contained herein is provided 1107 on an "AS IS" basis and THE INTERNET SOCIETY AND THE INTERNET 1108 ENGINEERING TASK FORCE DISCLAIMS ALL WARRANTIES, EXPRESS OR 1109 IMPLIED, INCLUDING BUT NOT LIMITED TO ANY WARRANTY THAT THE 1110 USE OF THE INFORMATION HEREIN WILL NOT INFRINGE ANY RIGHTS OR 1111 ANY IMPLIED WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A 1112 PARTICULAR PURPOSE.