idnits 2.17.1 draft-ietf-idr-add-paths-guidelines-05.txt: Checking boilerplate required by RFC 5378 and the IETF Trust (see https://trustee.ietf.org/license-info): ---------------------------------------------------------------------------- No issues found here. Checking nits according to https://www.ietf.org/id-info/1id-guidelines.txt: ---------------------------------------------------------------------------- No issues found here. Checking nits according to https://www.ietf.org/id-info/checklist : ---------------------------------------------------------------------------- == There are 5 instances of lines with non-RFC6890-compliant IPv4 addresses in the document. If these are example addresses, they should be changed. Miscellaneous warnings: ---------------------------------------------------------------------------- == The copyright year in the IETF Trust and authors Copyright Line does not match the current year == Line 591 has weird spacing: '...ltipath selec...' -- 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: 'RFC 4271' is mentioned on line 141, but not defined == Missing Reference: 'RFC-2119' is mentioned on line 163, but not defined == Missing Reference: 'RFC4364' is mentioned on line 788, but not defined == Unused Reference: 'RFC2119' is defined on line 860, but no explicit reference was found in the text == Unused Reference: 'RFC4271' is defined on line 887, but no explicit reference was found in the text == Outdated reference: A later version (-15) exists of draft-ietf-idr-add-paths-07 Summary: 0 errors (**), 0 flaws (~~), 9 warnings (==), 1 comment (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 1 Network Working Group J. Uttaro 2 Internet Draft AT&T 3 Expires: Dec 21, 2013 P. Francois 4 Intended status: Standards Track IMDEA Networks 5 June 21, 2013 R. Fragassi 6 A. Simpson 7 Alcatel-Lucent 8 K. Patel 9 Cisco Systems 11 Best Practices for Advertisement of Multiple Paths in IBGP 12 draft-ietf-idr-add-paths-guidelines-05.txt 14 Status of this Memo 16 This Internet-Draft is submitted in full conformance with the 17 provisions of BCP 78 and BCP 79. 19 Internet-Drafts are working documents of the Internet Engineering 20 Task Force (IETF), its areas, and its working groups. Note that 21 other groups may also distribute working documents as Internet- 22 Drafts. 24 Internet-Drafts are draft documents valid for a maximum of six months 25 and may be updated, replaced, or obsoleted by other documents at any 26 time. It is inappropriate to use Internet-Drafts as reference 27 material or to cite them other than as "work in progress." 29 The list of current Internet-Drafts can be accessed at 30 http://www.ietf.org/ietf/1id-abstracts.txt 32 The list of Internet-Draft Shadow Directories can be accessed at 33 http://www.ietf.org/shadow.html 35 This Internet-Draft will expire on Dec 21, 2013. 37 Copyright Notice 39 Copyright (c) 2012 IETF Trust and the persons identified as the 40 document authors. All rights reserved. 42 This document is subject to BCP 78 and the IETF Trust's Legal 43 Provisions Relating to IETF Documents 44 (http://trustee.ietf.org/license-info) in effect on the date of 45 publication of this document. Please review these documents 46 carefully, as they describe your rights and restrictions with respect 47 to this document. Code Components extracted from this document must 48 include Simplified BSD License text as described in Section 4.e of 49 the Trust Legal Provisions and are provided without warranty as 50 described in the Simplified BSD License. 52 Abstract 54 Add-Paths is a BGP enhancement that allows a BGP router to advertise 55 multiple distinct paths for the same prefix/NLRI. This provides a 56 number of potential benefits, including reduced routing churn, faster 57 convergence and better loadsharing. 59 This document provides recommendations to implementers of Add-Paths 60 so that network operators have the tools needed to address their 61 specific applications and to manage the scalability impact of Add- 62 Paths. A router implementing Add-Paths may learn many paths for a 63 prefix and must decide which of these to advertise to peers. This 64 document analyses different algorithms for making this selection and 65 provides recommendations based on the target application. 67 Table of Contents 69 1. Introduction...................................................4 70 2. Terminology....................................................4 71 3. Add-Paths Applications.........................................5 72 3.1. Fast Connectivity Restoration.............................5 73 3.2. Load Balancing............................................7 74 3.3. Churn Reduction...........................................7 75 3.4. Suppression of MED-Related Persistent Route Oscillation...7 76 4. Implementation Guidelines......................................8 77 4.1. Capability Negotiation....................................8 78 4.2. Receiving Multiple Paths..................................9 79 4.3. Advertising Multiple Paths................................9 80 4.3.1. Path Selection Modes................................11 81 4.3.1.1. Advertise All Paths............................11 82 4.3.1.2. Advertise N Paths..............................12 83 4.3.1.3. Advertise All AS-Wide Best Paths...............12 84 4.3.1.4. Advertise ALL AS-Wide Best and Next-Best Paths 85 (Double AS Wide)........................................13 86 4.3.1.5. Advertise Used Multipaths......................14 87 4.3.2. Derived Modes from Bounding the Number of Advertised 88 Paths......................................................14 89 4.3.3. Derived Modes from Adding N More Paths..............15 90 5. Deployment Considerations.....................................15 91 5.1. Introducing Add-Paths into an Existing Network...........15 92 5.2. Scalability Considerations...............................18 93 5.3. Routing Consistency Considerations.......................18 94 5.4. Consistency between Advertised Paths and Forwarding Paths19 95 5.5. Routing Churn............................................20 96 6. Security Considerations.......................................20 97 7. Acknowledgments...............................................20 98 8. Contributors..................................................20 99 9. IANA Considerations...........................................20 100 10. References...................................................21 101 10.1. Normative References....................................21 102 10.2. Informative References..................................21 103 Appendix A. Other Path Selection Modes...........................22 104 A.1. Advertise Neighbor-AS Group Best Path....................22 105 A.2. Best LocPref/Second LocPref..............................22 106 A.3. Advertise Paths at decisive step -1......................23 108 1. Introduction 110 The BGP Add-Paths capability enhances current BGP implementations by 111 allowing a BGP router to exchange with its BGP peers more than one 112 path for the same destination/NLRI. The base BGP standard [RFC 4271] 113 does not provide for such a capability. If a BGP router learns 114 multiple paths for the same NLRI (from multiple peers), it selects 115 only one as its best path and advertises the best path to its peers. 116 The primary goal of Add-Paths is to increase the visibility of paths 117 within an iBGP system. This has the effect of improving robustness 118 in case of failure, reducing the number of BGP messages exchanged 119 during such an event, and offering the potential for faster re- 120 convergence. Through careful selection of the paths to be advertised, 121 Add-Paths can also prevent routing oscillations. 123 The purpose of this document is to provide the necessary 124 recommendations to the implementers of Add-Paths so that network 125 operators have the tools needed to address their specific 126 applications and to manage the scalability impact of Add-Paths while 127 maintaining routing consistency. A router implementing Add-Paths may 128 learn many paths for a prefix and must decide which of these to 129 advertise to peers. This document analyses different algorithms for 130 making this selection and provides recommendations based on the 131 target application. 133 2. Terminology 135 In this document the following terms are used: 137 Add-Paths peer: refers a peer with which the local system has agreed 138 to receive and/or send NLRI with path identifiers 140 Primary path: A path toward a prefix that is considered a best path 141 by the BGP decision process [RFC 4271] and actively used for 142 forwarding traffic to that prefix. A router may have multiple primary 143 paths for a prefix if it implements multipath. 145 Diverse path: A BGP path associated with a different BGP next-hop and 146 BGP router than some other set of paths. The BGP router associated 147 with a path is inferred from the ORIGINATOR_ID attribute or, if there 148 is none, the BGP Identifier of the peer that advertised the path. 150 Backup path: A diverse path with respect to the primary paths toward 151 a prefix. The backup path can be used to forward traffic to the 152 destination if the primary paths fail. 154 Optimal backup path: The backup path that will be selected as the new 155 best path for a prefix when all primary paths are removed/withdrawn. 157 AS-Wide preferred paths: All paths that are considered as best when 158 applying rules of the BGP decision process up to the IGP tie-break. 160 The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", 161 "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this 162 document are to be interpreted as described in [RFC-2119]. 164 3. Add-Paths Applications 166 [draft-pmohapat] presents the applications that would benefit from 167 multiple paths advertisement in iBGP. They are summarized in the 168 following subsections. 170 3.1. Fast Connectivity Restoration 172 With the dissemination of backup paths, fast connectivity restoration 173 and convergence can be achieved. If a router has a backup path, it 174 can directly select that path as best upon failure of the primary 175 path. This minimizes packet loss in the dataplane. Sending multiple 176 paths in iBGP allows routers to receive backup paths when path 177 visibility is not sufficient with classical BGP. This is especially 178 useful when Route Reflection is used. 180 Consider a network such as the one depicted in Figure 1 and suppose 181 that none of the routers support Add-Paths. AS1 receives from AS3 2 182 paths (A and B) to a particular destination XYZ. Suppose path A is 183 preferred over path B due to path A having a lower MED (multi-exit 184 discriminator). 186 AS1 uses a route reflector RR1 to reduce the scale of its IBGP mesh. 187 If the routers in AS1 are not configured for best-external then RR1 188 knows about only path A during steady state because router B 189 suppresses/withdraws its advertisement of path (B) to RR1. If the 190 routers in AS1 do support best-external then RR1 may have both paths 191 in its Adj-RIB-IN, but regardless of the best-external configuration 192 RR1 can only advertise its best path A to its peers, including router 193 D. 195 ======== ===================== 196 = +---+ +---+ +---+ 197 = |RTR|________|RTR| |RTR| 198 = | E | | A | | C | 199 = +---+Path A->+---+ AS1 +---+ 200 = = = \ / = 201 = = = \ / = 202 = = = \ / = 203 = = = \ / = 204 = AS3 = = +---+ = 205 = = = |RR | = 206 = = = | 1 | = 207 = = = +---+ = 208 = = = / \ = 209 = = = / \ = 210 = = = / \ = 211 = = = / \ = 212 = +---+Path B->+---+ +---+ 213 = |RTR| ______|RTR| |RTR| 214 = | F | | B | | D | 215 = +---+ +---+ +---+ 216 ======== ===================== 218 Figure 1: Example Topology 220 Under these circumstances consider the steps required to restore 221 traffic from router D to destination XYZ when the link between Router 222 A and Router E fails. (Assume that router A set next-hop to self when 223 advertising path A and that router B is not configured for best- 224 external). 226 1. Router A sends a BGP UPDATE message withdrawing its advertisement 227 of path (A). 229 2. RR1 receives the withdrawal, and propagates it to its other client 230 peers, routers B, C and D. 232 3. When router B receives the withdrawal of path (A) it reruns its 233 decision process and selects path (B) as its new best path. Router 234 B advertises path (B) to RR1. 236 4. RR1 reruns its decision process and selects path (B) as its new 237 best path. RR1 advertises path (B) to client peers A, C and D. 239 5. Router D reruns its decisions process, determines path (B) to be 240 the best path, and updates its forwarding table. After this step 241 traffic from router D to destination XYZ is restored (the traffic 242 path has changed from A to B). 244 With the use of Add-Paths, the convergence time for the above path 245 failure example can be reduced considerably. The main reason for the 246 improvement is that Add-Paths allows router D to be aware of more 247 than one path to destination XYZ prior to the failure of the best 248 path (A). In steady-state (with no failures) router B decides, as 249 before, that path (A) is its best path but because of its Add-Paths 250 (or best-external) configuration it also advertises path (B) to RR1. 251 Using Add-Paths RR1 can advertise both learned paths to its IBGP 252 peers, including router D. Now consider again the scenario where the 253 link between Router A and Router E fails. In this case, with Add- 254 Paths, fewer steps are required to achieve re-convergence: 256 1. Router A sends a BGP UPDATE message withdrawing its advertisement 257 of path (A). 259 2. RR1 receives the withdrawal, and propagates it to its other client 260 peers, routers B, C and D. 262 3. Router D receives the withdrawal, reruns the decision process and 263 updates the forwarding entry for destination XYZ. 265 3.2. Load Balancing 267 Increased path diversity allows routers to install several paths in 268 their forwarding tables in order to load balance traffic across those 269 paths. 271 3.3. Churn Reduction 273 When Add-Paths is used in an AS, the availability of additional 274 backup paths means failures can be recovered locally with much less 275 path exploration in iBGP and therefore less updates disseminated in 276 eBGP. When the preferred backup path is the post-convergence path, 277 churn is minimized. 279 3.4. Suppression of MED-Related Persistent Route Oscillation 281 As described in [oscillation], Add-Paths is a valuable tool in 282 helping to stop persistent route oscillations caused by comparison of 283 paths based on MED in topologies where route reflectors or the 284 confederation structure hide some paths. With the appropriate path 285 selection algorithm Add-Paths stops these route oscillations because 286 the same set of paths are consistently advertised by the route 287 reflector or the confederation border router and the routers 288 receiving this set of paths make stable routing decisions about the 289 best path. 291 4. Implementation Guidelines 293 This section discusses recommendations for the implementation of Add- 294 Paths. The following topics are addressed: 296 . Considerations related to Add-Paths capability negotiation 298 . Receiving BGP routes from Add-Paths peers 300 . Advertising BGP routes to Add-Paths peers. This section 301 discusses various path selection algorithms, which are the 302 procedures available to an Add-Paths speaker for deciding which 303 set of paths to advertise to an Add-Paths peer for particular 304 prefixes. 306 4.1. Capability Negotiation 308 +---+ +---+ 309 |RTR|___________|RTR| 310 | A | <-BGP-> | B | 311 +---+ +---+ 313 Figure 2: BGP Peering Example 315 In Figure 2, in order for a router A to receive multiple paths per 316 NLRI from peer B, for a particular address family (AFI=x, SAFI=y), 317 the BGP capabilities advertisements during session setup must 318 indicate that peer B wants to send multiple paths for AFI=x, SAFI=y 319 and that router A is willing to receive multiple paths for AFI=x, 320 SAFI=y. Similarly, in order for router A to send multiple paths per 321 NLRI to peer B, for a particular address family (AFI=x, SAFI=y), the 322 BGP capabilities advertisements must indicate that router A wants to 323 send multiple paths for AFI=x, SAFI=y and peer B is willing to 324 receive multiple paths for AFI=x, SAFI=y. Refer to [Add-Paths] for 325 details of the Add-Paths capabilities advertisement. 327 The capabilities of the local router MUST be configurable per peer 328 and per address family, and SHOULD support the ability to configure 329 send-only operation or receive-only operation. The default mode of 330 operation shall be to both send and receive. 332 4.2. Receiving Multiple Paths 334 Currently, per standard BGP behavior, if a BGP router receives an 335 advertisement of an NLRI and path from a specific peer and that peer 336 subsequently advertises the same NLRI with different path information 337 (e.g. a different NEXT_HOP and/or different path attributes) the new 338 path effectively overwrites the existing path. 340 When Add-Paths has been negotiated with the peer, the newly 341 advertised path should be stored in the RIB-IN along with all of the 342 paths previously advertised (and not withdrawn) by the peer. 344 When an Add-Paths speaker has negotiated to receive multiple paths 345 for (AFIx, SAFIy) from a peer all advertisements and withdrawals of 346 NLRI within that address family from that peer MUST include a path 347 identifier, as described in [Add-Paths]. The path identifiers have no 348 significance to the receiving peer. If the combination of NLRI and 349 path identifier in an advertisement from a peer is unique (does not 350 match an existing route in the RIB-IN from that peer) then the route 351 is added to the RIB-IN. If the combination of NLRI and path 352 identifier in a received advertisement is the same as an existing 353 route in the RIB-IN from the peer then the new route replaces the 354 existing one. If the combination of NLRI and path identifier in a 355 received withdrawal matches an existing route in the RIB-IN from the 356 peer then that route shall be removed from the RIB-IN. 358 A BGP UPDATE message from an Add-Paths peer may advertise and 359 withdraw more than one NLRI belonging to one or more address 360 families. In this case Add-Paths may be supported for some of the 361 address families and not others. In this situation the receiving BGP 362 router should not expect that all of the path identifiers in the 363 UPDATE message will be the same. 365 4.3. Advertising Multiple Paths 367 [Add-Paths] specifies how to encode the advertisement of multiple 368 paths towards the same NLRI over an iBGP session, but provides no 369 details about which set of multiple paths should be advertised. In 370 this section, four path selection algorithms are described and 371 compared with each other. These 4 algorithms are considered to be the 372 most useful across the widest range of deployment scenarios. The list 373 of possible path selection algorithms is much larger and for the 374 interested reader Appendix A provides information about other path 375 selection modes that were considered in historical versions of this 376 document. 378 In comparing any two path selection algorithms the following factors 379 should be taken into account: 381 Control Plane Load: When a router receives multiples paths for a 382 prefix from an iBGP client it has to store more paths in its Adj-Rib- 383 Ins. 385 Control Plane Stress: Coping with multiple iBGP paths has two 386 implications on the computation that a router has to handle. First, 387 it has to compute the paths to send to its peers, i.e. more than the 388 best path. Second, it also has to handle the potential churn related 389 to the exchange of those multiple paths. 391 MED/IGP oscillations: BGP sometimes suffers from routing oscillations 392 when the physical topology differs from the logical topology, or when 393 the MED attribute is used. This is due to the limited path 394 visibility when a single path is advertised and Route Reflection is 395 used. Increasing the path visibility by advertising multiple paths 396 can help solve this issue. 398 Path optimality: When a single path is advertised, border routers do 399 not always receive the optimal path. As an example, Route Reflectors 400 typically send a single path chosen based on their own IGP tie-break 401 (although modifications to this are proposed in [BGP-ORR]). 402 Increasing path visibility would also help routers to learn the path 403 that is best suited for them w.r.t. the IGP tie-break. 405 Backup path optimality: Multiple paths advertisement gives routers 406 the opportunity to have a backup path. However, some backup paths 407 are better than others. Indeed, when a link failure occurs, if a 408 router already knows its post-convergence path, the BGP re- 409 convergence is straightforward and traffic is less impacted by the 410 transient use of non-best forwarding paths. 412 Convergence time: Advertising multiple paths in iBGP has an impact on 413 the convergence time of the BGP system. More paths need to be 414 exchanged, but on the other hand, the routing information is 415 propagated faster. With an increased path visibility, there is less 416 path exploration during the convergence. Also, with the availability 417 of backup paths, convergence time in case of failure is also reduced. 419 Target application: Depending on the application type, the number of 420 paths to advertise for a prefix will vary. For example, for fast 421 connectivity restoration, it may be sufficient to advertise only 2 422 paths to a peer so that it will have the best path and the optimal 423 backup path. For load balancing purposes, it may be desirable to 424 advertise more paths, but inclusion of the optimal backup path in the 425 set may be less critical. For route oscillation elimination, it is 426 required to advertise all group-best paths for a prefix. 428 4.3.1. Path Selection Modes 430 The following subsections describe the 4 main path selection modes 431 considered in this draft. Each mode is considered either MANDATORY or 432 OPTIONAL. A MANDATORY mode MUST be supported by any implementation 433 that claims compliance with this document. An OPTIONAL made may be 434 supported by some but not all implementations. 436 The path selection mode and any parameters applicable to the mode 437 MUST be configurable per AFI/SAFI and per peer and SHOULD be 438 configurable per prefix. To illustrate the value of this flexibility, 439 consider a prefix P that belongs to an address family F requiring 440 path IDs to be included with every NLRI (e.g. due to the Add-Paths 441 capability negotiation with the peer). If P is one of a number of 442 prefixes that would not benefit from the advertisement of multiple 443 paths then it is perfectly valid to send only the best path. 445 4.3.1.1. Advertise All Paths 447 A simple rule for advertising multiple paths in iBGP is to advertise 448 to iBGP peers all received paths minus those blocked by export 449 filters or applicable split horizon rules. This solution is easy to 450 implement, but the counterpart is that all those paths need to be 451 stored by all routers that receive them, which can be quite 452 expensive. If a path to a prefix P is advertised to N border 453 routers, with a Full Mesh of iBGP sessions, all routers have N paths 454 in their Adj-RIB-Ins. If Route Reflection is used and each client is 455 connected to 2 Route Reflectors, it may learn up to 2*N paths. 457 This solution gives a perfect path visibility to all routers, thus 458 limiting churn and losses of connectivity in case of failure. Indeed, 459 this allows routers to select their optimal primary path, and to 460 switch on their optimal backup path in case of failure. 462 However, as more paths are exchanged, the number of BGP messages 463 disseminated during the initial iBGP convergence can be high, and 464 convergence may be slower. 466 Routing oscillations are prevented with this rule, because a router 467 won't need to withdraw a previously advertised path when its best 468 path changes. 470 This path selection mode is OPTIONAL. 472 4.3.1.2. Advertise N Paths 474 Another solution is for a router to advertise a maximum of N paths to 475 iBGP peers. Here, the computational cost is the selection of the N 476 paths. Indeed, there must be a ranking of the paths in order to 477 advertise the most interesting ones. A way for a router to select N 478 paths is to run N times its decision process. At each iteration of 479 the process only those paths not selected during a previous iteration 480 and those with a different NEXT_HOP and BGP Identifier (or Originator 481 ID) combination from previously-selected paths are eligible for 482 consideration. In this mode the paths actually advertised to a peer 483 are the eligible paths (up to N) minus those blocked by export 484 filters or applicable split horizon rules. The memory cost is 485 bounded: a router receives a maximum of N paths for each prefix from 486 each peer. With N equal to 2, all routers know at least two paths and 487 can provide local recovery in case of failure. If multipath routing 488 is to be deployed in the AS, N can be increased to provide more 489 alternate paths to the routers. 491 Path optimality and backup path optimality are not guaranteed, i.e. 492 it is possible that the optimal path of a router (w.r.t. IGP tie- 493 break) is not contained in the set of paths advertised by its Route 494 Reflector. However, as the number of paths that it receives is higher 495 than without Add-Paths, it is possible that the chosen nexthop is 496 closer to the router in terms of IGP cost than the nexthop that would 497 have been chosen without Add-Paths. 499 This solution helps to reduce routing oscillations, but not in all 500 cases. Indeed, path visibility is still constrained by the maximum 501 number of paths, and configurations with routing oscillations still 502 exist. 504 This path selection mode is MANDATORY. The default value of N MUST be 505 2. The value of N MUST be configurable and MAY be upper bounded by 506 an implementation. 508 The default value of 2 ensures the availability of a backup path (if 509 2 or more paths have been received) while maintaining minimum impact 510 to memory and churn. If Add-N with N equal to 2 is insufficient to 511 meet another objective (e.g. loadsharing or MED/IGP oscillation) 512 there is always a large enough value of N that can selected, if N is 513 configurable, to meet that objective. 515 4.3.1.3. Advertise All AS-Wide Best Paths 517 Another choice is to consider the set of paths with the same AS-wide 518 preference [Basu-ibgp-osc], i.e. the paths that all routers would 519 select based on the rules of the decision process that are not 520 router-dependent (i.e. Local-preference, ASPath length and MED 521 rules). Thus, for a given router, those paths only differ by the IGP 522 cost to the nexthop or by the tie-breaking rules. The paths actually 523 advertised to a peer are the set of AS-wide best paths minus those 524 blocked by export filters or applicable split horizon rules. 526 The computational cost is reduced, as a router only has to send the 527 paths remaining before applying the IGP tie-breaking rule. However, 528 it is difficult to predict how many paths will be stored, as it 529 depends on the number of eBGP sessions on which this prefix is 530 advertised with the best AS-wide preference. 532 With this rule, the routing system is optimal: all routers can choose 533 their best path (or best paths if multipath is used) based on their 534 router-specific preferences, i.e. the IGP cost to the nexthop. Hot 535 potato routing is respected. Also, MED oscillations are prevented, 536 because the path visibility among the AS-wide preferred paths is 537 total. 539 The existence of a backup path is not guaranteed. If only one path 540 with the AS-wide best attributes exists, there is no backup path 541 disseminated. However, if such a path exists, it is optimal as it 542 has the same AS-wide preference as the primary 544 This path selection mode is OPTIONAL. 546 4.3.1.4. Advertise ALL AS-Wide Best and Next-Best Paths (Double 547 AS Wide) 549 This variant of "Advertise All AS Wide Best Paths" trades-off the 550 number of paths being propagated within the iBGP system for post- 551 convergence alternate paths availability and routing stability. A BGP 552 speaker running this mode will select, as candidates for 553 advertisement, its AS Wide Best paths, plus all the AS Wide Best 554 paths obtained when removing the first ones from consideration. The 555 paths actually advertised to a peer are the double-AS_wide candidate 556 paths minus those blocked by export filters or applicable split 557 horizon rules. 559 Under this mode, a BGP speaker knows multiple AS-Wide best paths or 560 the AS-Wide best path and all the second AS-Wide best paths, so that 561 routing optimality and backup path availability are ensured. Note 562 that the post-convergence paths will be known by each BGP node in an 563 AS supporting this mode. 565 The computation complexity of this mode is relatively low as it 566 requires to run the usual BGP Decision Process up to and including 567 the MED rule. The set of paths remaining after that step form the AS- 568 Wide best paths. Next, a best path selection algorithm is run up to 569 and including the MED rule, based on the paths that are not in the 570 set of AS-Wide best paths. 572 The number of paths for a prefix p, known by a given router of the 573 AS, is the number of AS-Wide best and second AS-Wide best paths found 574 at the Borders of the AS. 576 MED Oscillations are avoided by this mode, both for the primary and 577 alternate paths being picked under this mode. 579 This path selection mode is OPTIONAL. 581 4.3.1.5. Advertise Used Multipaths 583 Many BGP implementations support BGP Multipath, allowing a BGP router 584 to use multiple BGP next-hops for forwarding towards a prefix/NLRI 585 when the corresponding paths are considered equally preferred. In 586 cases where the deployment of Add-Paths is mostly aimed at providing 587 multiple paths for load balancing with BGP Multipath, a natural 588 approach for a BGP speaker supporting Add-paths is to advertise the 589 paths that are selected by its BGP multipath selection algorithm. 591 BGP Multipath selection algorithms can vary depending on the 592 implementation and configuration options. An Add-Paths mode based on 593 BGP multipath is considered practical because it lets the BGP path 594 propagation be aligned with the load balancing objectives expressed 595 by the operator configuring BGP multipath. 597 In some deployment scenarios, it is likely that such a mode leads to 598 the selection and advertisement of a large number of paths for some 599 NLRI, and hence should be controlled as per the mechanism described 600 in section 4.3.2. In case the number of multipaths exceeds the upper 601 bound on the number of advertised paths the ones that should be 602 advertised are those with the highest degree of preference by the BGP 603 decision process. This can be achieved if the advertising router has 604 strictly ordered all of its paths. 606 This path selection mode is OPTIONAL. 608 4.3.2. Derived Modes from Bounding the Number of Advertised Paths 610 For some of the modes discussed in section 4.3.1 the number of paths 611 selected by the algorithm (M) is not predictable in advance, and 612 depends on factors such as network topology. For such modes, 613 implementations MAY support the ability to limit the number of 614 advertised paths to some value N that is less than M. 616 It must be noted that the resulting derivative mode may no longer 617 meet the properties stated in section 4.3.1 (which assumes N=M). This 618 is particularly true for the MED oscillation avoidance property. The 619 use of such bounds thus needs to be considered carefully in 620 deployments where MED oscillation avoidance is a key goal of 621 deploying Add-path. If fast recovery is the main objective then it is 622 reasonable and sufficient to set N to 2. If the main goal is 623 improved load-balancing then limiting N to number of ECMP paths 624 supported by the forwarding planes of the receiving routers is also a 625 reasonable practice. 627 4.3.3. Derived Modes from Adding N More Paths 629 Some modes discussed in section 4.3.1 may result in only one or a few 630 selected paths, depending on network topology and/or router 631 configuration, and this small number of paths may not meet minimum 632 requirements for backup path or load balancing purposes. When using 633 such modes implementations MAY support the ability to add N more 634 paths to the set returned by the basic selection algorithm as 635 described in section 4.3.1. The N more paths should be the N next- 636 best paths, as determined by the BGP decision process. 638 It must be noted that the resulting derivative mode may no longer 639 meet the properties stated in section 4.3.1 (which assumes N=0). 641 5. Deployment Considerations 643 This section proposes a potential strategy for introducing Add-Paths 644 into an existing network and discusses considerations related to 645 scalability, routing consistency and routing churn. 647 5.1. Introducing Add-Paths into an Existing Network 649 There are many possible ways that Add-Paths can be introduced into an 650 existing deployed network. It is not a practical goal for this 651 document to list all of these options and discuss the pros and cons 652 for each one. It is however valuable to consider an example migration 653 strategy that may be relatively common among layer 3 service 654 providers that currently use route reflectors for scaling. This 655 example migration strategy is attractive for several reasons: 657 1. It involves incremental steps that allow the impact of Add- 658 Paths to be carefully evaluated before proceeding to the next 659 step. 661 2. It recognizes the fact that many routers will require at least 662 a software upgrade to support Add-Paths, and it will not be 663 practical to upgrade all of these routers all at once. 665 3. It reduces convergence time (in stages) with a relatively 666 moderate increase in router memory and CPU demands. 668 The example migration strategy assumes a starting point of a deployed 669 network with one or more RR clusters. None of the routers in the 670 network support Add-Paths without an upgrade, but some do support 671 best-external. Two of the clusters in this network are shown in 672 Figure 3. In cluster 2, PE1, PE2, RRy and RRz are configured for 673 best-external. This makes RRy and RRz aware of all external paths 674 received by PEs in cluster 2 and ensures that RRy and RRz can 675 advertise a path to the RRs in cluster 1 if it happens that the best 676 overall route is learned from cluster 1. It doesn't however allow 677 other clusters to be aware of more than one path per prefix learned 678 by cluster 2. 680 ========== ================== 681 = = = = 682 = +---+ +---+ +---+ = 683 = |RR |---------------|RR | <-BE| | = 684 = |a | |y |------|PE1| = 685 = | | | | | | = 686 = +---+ +---+ +---+ = 687 = | = = | \ / = 688 = | = = | \ / = 689 = | = = | \/ = 690 = | = = | /\ = 691 = | = = | / \ = 692 = | = = | / \ = 693 = +---+ +---+ +---+ = 694 = |RR |---------------|RR |------| | = 695 = |b | |z | <-BE|PE2| = 696 = | | | | | | = 697 = +---+ +---+ +---+ = 698 = = = = 699 ========== ================== 700 RR Cluster 1 RR Cluster 2 702 Figure 3: RR Cluster Before Add-Paths 704 The following sequence of steps occurs in the example migration 705 strategy: 707 1. The route reflectors are upgraded in each cluster, one by one, to 708 support Add-Paths. This allows the intra- and (eventually) inter- 709 cluster RR-to-RR sessions to start using Add-Paths. All RRs are 710 configured to use the Add-N, N=2 path selection algorithm. The 711 effect of this step is to slightly reduce convergence time when 712 the best and second-best paths for a prefix are learned by a 713 single cluster (such as cluster 2 in Figure 3). 715 2. The clients are upgraded in each cluster, one by one, to support 716 Add-Paths. On the RRs Add-Paths is configured to use the Add-N, 717 N=2 path selection algorithm towards upgraded client peers. At 718 this step clients are configured in the receive-only Add-Paths 719 mode. This means that best-external continues to operate as 720 before in the client-to-RR direction. The effect of this step is 721 to ensure that all clients have two paths per prefix for ECMP or 722 fast failover, assuming at least 2 paths are available. 724 3. The clients are re-configured to use Add-Paths in the transmit 725 direction towards their RR peers. This causes Add-Paths to replace 726 the best-external behavior. The effect of this step is to free up 727 CPU and memory resources related to the storage of paths that are 728 third best or worse. If a cluster such as the one in Figure 3 had 729 50 clients, and 10 of these learned an external route for the same 730 prefix, then the RRs in that cluster would need to store up to 12 731 paths for that prefix. This would be true even if the 2 best 732 overall paths came from another cluster. Contrast this with the 733 use of Add-Paths in the client-to-RR direction. For the same case 734 the route reflectors need only store the 2 paths learned from non- 735 client peers. 737 5.2. Scalability Considerations 739 In terms of scalability, we note that advertising multiple paths per 740 prefix requires more memory and state than the current behavior of 741 advertising the best path only. A BGP speaker that does not implement 742 Add-Paths maintains send state information in its prefix data 743 structure per neighbor as a way to determine that the prefix has been 744 advertised to the neighbor. With Add-Paths, this information has to 745 be replicated on a per path basis that needs to be advertised. 746 Mathematically, if "send state" size per prefix is 's' bytes, number 747 of neighbors is 'n', and number of paths being advertised is 'p', 748 then the current memory requirement for BGP "send state" = n * s 749 bytes; with Add-Paths, it becomes n * s * p bytes. In practice, this 750 value may be reduced with implementation optimizations similar to 751 attribute sharing. Receiving multiple paths per prefix also requires 752 more memory and state since each path is a separate entry in the Adj- 753 RIB-Ins. 755 5.3. Routing Consistency Considerations 757 As discussed in previous sections Add-Paths can help routers select 758 more optimal paths and it can help deal with certain route 759 oscillation conditions arising from incomplete knowledge of the 760 available paths. But depending on the path selection algorithm and 761 how it is used Add-Paths is not immune to its own cases of routing 762 inconsistencies. If the BGP routers within an AS do not make 763 consistent routing decisions about how to reach a particular 764 destination, route oscillations may occur and these route 765 oscillations may result in traffic loss. 767 Optimizing an Add-Paths deployment for scalability may run counter to 768 routing consistency goals, and in these circumstances operators have 769 to decide the correct tradeoff for their particular deployment. For 770 example the Advertise All Paths mode, if applied to many prefixes, is 771 far from ideal from a scalability perspective but it does guarantee 772 routing consistency and correctness. A path selection mode that 773 allows better control over scalability is the Advertise N paths mode, 774 but this is susceptible to routing inconsistency. First, if the N 775 paths do not include the best path from each neighbor AS group then 776 route oscillation cannot be precluded. Second, if the advertising 777 router (e.g. an RR) advertises N paths to peer_n and M paths to 778 peer_m, and N < M, care must be exercised to ensure that all paths 779 advertised to peer_n are included in the paths advertised to peer_m. 780 This can be assured as long as the advertising router has strictly 781 ordered all of its paths. 783 5.4. Consistency between Advertised Paths and Forwarding Paths 785 When using Add-Paths, routers may advertise paths that they have not 786 selected as best, and that they are thus not using for traffic 787 forwarding. This is generally not an issue if encapsulation is used 788 in the AS as described in [RFC4364] and all forwarding decisions, 789 including by the tunnel egress router, are based on label information 790 - i.e. if only the ingress router performs an IP FIB lookup. In this 791 situation the dataplane path followed by the packets is the one 792 intended by the ingress router, and corresponds to the control plane 793 path it selected. 795 On the other hand, if Add-Paths is used in a network without 796 encapsulation, some scenarios can result in forwarding deflection or 797 loops. Such forwarding anomalies already occur without Add-Paths, 798 when the routers on the forwarding path do not have a synchronized 799 view of the best path. They will deflect the traffic to their own 800 local view of the best path, and, when multiple deflections occur, 801 forwarding loops can occur. With Add-Paths, the issue can be 802 exacerbated due to routers advertising non-best paths. As discussed 803 above, encapsulation can help with this issue, but only to the extent 804 that it allows downstream routers to forward without an IP FIB 805 lookup. 807 A first example of such issue is when the Local-Pref of non-primary 808 paths received over iBGP sessions is modified. The ingress router 809 may thus select as best a path non-preferred by the egress, and the 810 egress router will thus deflect the traffic. 812 Another example is when the best path is selected based on tie- 813 breaking rule. When the ingress and the egress base their path 814 selection on the router-id of the neighbor that advertised the path 815 to them, the result may be different for each of them. This specific 816 issue is described and solved in [draft-pmohapat]. 818 5.5. Routing Churn 820 As noted in section 3.3 using Add-Paths between IBGP peers can help 821 to reduce routing churn with EBGP peers. This benefit does however 822 come at the cost of potentially increased churn between the IBGP Add- 823 Paths peers. In a non Add-Paths deployment a change in the preference 824 order of non-best paths requires no updates to be sent to peers. But 825 when a router has Add-Paths peers changes in non-best path preference 826 may no longer be invisible and increased route churn may be 827 observable. Choosing the right path selection mode and parameters - 828 for example not setting N unnecessarily large in the Add-N mode, is 829 important to minimizing this additional churn. 831 6. Security Considerations 833 TBD 835 7. Acknowledgments 837 This document was prepared using 2-Word-v2.0.template.dot. 839 8. Contributors 841 Virginie Van den Schrieck 842 Email: v.vandenschrieck@gmail.com 844 Pradosh Mohapatra 845 Cumulus Networks 846 Email: pmohapat@cumulusnetworks.com 848 Rohit Gupta 849 Apple Inc 850 Email: rxgupta@apple.com 852 9. IANA Considerations 854 TBD 856 10. References 858 10.1. Normative References 860 [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate 861 Requirement Levels", BCP 14, RFC 2119, March 1997. 863 10.2. Informative References 865 [Add-Paths] Walton, D., Retana, A., Chen E., Scudder J., 866 "Advertisement of Multiple Paths in BGP", draft- 867 ietf-idr-add-paths-07, June 17, 2012. 869 [draft-pmohapat] Mohapatra, P., Fernando, R., Filsfils, C., and R. 870 Raszuk, "Fast Connectivity Restoration Using BGP 871 Add-path", draft-pmohapat-idr-fast-conn-restore- 872 02.txt, Oct 3, 2011. 874 [oscillation] Walton, D., Retana, A., Chen, E., Scudder, J., "BGP 875 Persistent Route Oscillation Solutions", draft- 876 walton-bgp-route-oscillation-stop-06.txt, June 14, 877 2012. 879 [Basu-ibgp-osc] Basu, A., Ong, C., Rasala, A., Sheperd, B., and G. 880 Wilfong, "Route oscillations in iBGP with Route 881 Reflection", Sigcomm 2002. 883 [BGP-ORR] Raszuk, R., Cassar, C., Aman, E., Decraene, B., "BGP 884 Optimal Route Reflection", draft-raszuk-bgp-optimal- 885 route-reflection-01, March 11, 2011. 887 [RFC4271] Rekhter, Y., Li, T., Hares, S., "A Border Gateway 888 Protocol 4 (BGP-4), January 2006. 890 Appendix A. Other Path Selection Modes 892 A.1. Advertise Neighbor-AS Group Best Path 894 [walton-osc] proposes that a router groups its paths based on the 895 neighbor AS from which it was learned, and to advertise the best path 896 in each of those groups. 898 The control plane stress induced by this solution is the computation 899 of the per-neighbor path group, and the application of the decision 900 process to each of them. The Control-Plane load is bounded by the 901 number of neighboring ASes advertising a prefix, which cannot be 902 known a-priori. 904 Path optimality and backup path optimality are not guaranteed, as the 905 paths advertised are not all the AS-wide preferred paths. Backup path 906 availability is not guaranteed. Indeed, if only one AS advertises 907 this prefix, even on multiple eBGP sessions, only one of the paths 908 may be selected and advertised. 910 A.2. Best LocPref/Second LocPref 912 This selection method consists in grouping the paths by Local 913 Preference. A router sends to its peers all paths with the highest 914 Local Preference. If there is only a single path with the highest 915 Local Preference, it also sends all paths with the second best Local 916 Preference. 918 This method ensures that all routers know all paths with the best 919 local preference. As local preference are often related to the type 920 of peering of the peer the path comes from, this ensures that in case 921 of failure, routers have a backup path of equivalent quality. This 922 prevents for example that a router switches temporarily on a peer 923 path while an alternate path from a customer is available but hidden 924 at the border of the AS. Such a situation could result in a 925 temporary withdrawal of the prefix on some eBGP sessions when the 926 router selects the path via the peer. 928 The advertisement of the Second Local Preference occurs when there is 929 no alternate path with the same quality as the best path. This way, 930 fast convergence is still ensured. Backup path is optimal, as it has 931 the second AS-Wide preference, which becomes the AS-wide best 932 preference upon failure of the primary one. 934 Sending all the paths with a given Local Preference also has a 935 positive impact on routing optimality. Indeed, this allows border 936 routers to have an increased path visibility and to choose their best 937 path based on their own criteria. 939 The computational cost of this solution is reduced when there are 940 several paths with the best local preference. In this case, it is 941 sufficient to stop the decision process after the first rule to have 942 the set of paths to be advertised. When it is necessary to advertise 943 the paths with second local-preference, the additional cost is to 944 apply a second time the first rule of the decision process, which is 945 still reasonable. The memory cost depends on the number of paths 946 with the best local preference. 948 A.3. Advertise Paths at decisive step -1 950 When the goal is to provide fast recovery by advertising candidate 951 post-reconvergence paths, one can choose to stop the decision process 952 just before the step where only one path remains. If the decision 953 process comes to IGP tie-break, all remaining paths are advertised. 954 This way, routers advertise as many paths as possible with a quality 955 as similar as possible. 957 This path selection is an intermediary solution between the two 958 preceding ones. Here, instead of stopping the decision process at 959 the local preference step or the IGP step, we stop it before the rule 960 that removes the best potential backup paths. This way, we minimize 961 the number of paths to advertise while guaranteeing the presence of a 962 backup path. Primary and backup path optimality is ensured, as all 963 paths with the same AS-wide preference as the best paths are included 964 in the set of paths advertised. 966 Authors' Addresses 968 Jim Uttaro 969 AT&T 970 200 S. Laurel Avenue 971 Middletown, NJ 07748 USA 972 Email: uttaro@att.com 974 Pierre Francois 975 IMDEA Networks 976 Avenida del Mar Mediterraneo 977 Leganes 28919 978 Spain 979 Email: pierre.francois@imdea.org 981 Roberto Fragassi 982 Alcatel-Lucent 983 600 Mountain Avenue 984 Murray Hill, New Jersey 985 Email: roberto.fragassi@alcatel-lucent.com 987 Adam Simpson 988 Alcatel-Lucent 989 600 March Road 990 Ottawa, Ontario K2K 2E6 991 Canada 992 Email: adam.simpson@alcatel-lucent.com 994 Keyur Patel 995 Cisco Systems 996 170 W. Tasman Drive 997 San Jose, CA 95134 USA 998 Email: keyupate@cisco.com