idnits 2.17.1 draft-ietf-spring-segment-routing-msdc-00.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 : ---------------------------------------------------------------------------- No issues found here. Miscellaneous warnings: ---------------------------------------------------------------------------- == The copyright year in the IETF Trust and authors Copyright Line does not match the current year -- The document date (October 12, 2015) is 3090 days in the past. Is this intentional? Checking references for intended status: Informational ---------------------------------------------------------------------------- -- Looks like a reference, but probably isn't: '16000' on line 272 -- Looks like a reference, but probably isn't: '23999' on line 272 -- Looks like a reference, but probably isn't: '1000' on line 862 -- Looks like a reference, but probably isn't: '1999' on line 862 -- Looks like a reference, but probably isn't: '2000' on line 862 -- Looks like a reference, but probably isn't: '2999' on line 862 ** Obsolete normative reference: RFC 3107 (Obsoleted by RFC 8277) == Outdated reference: A later version (-11) exists of draft-ietf-rtgwg-bgp-routing-large-dc-07 Summary: 1 error (**), 0 flaws (~~), 2 warnings (==), 7 comments (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 2 Network Working Group C. Filsfils, Ed. 3 Internet-Draft S. Previdi, Ed. 4 Intended status: Informational Cisco Systems, Inc. 5 Expires: April 14, 2016 J. Mitchell 6 Unaffiliated 7 E. Aries 8 P. Lapukhov 9 Facebook 10 October 12, 2015 12 BGP-Prefix Segment in large-scale data centers 13 draft-ietf-spring-segment-routing-msdc-00 15 Abstract 17 This document describes the motivation and benefits for applying 18 segment routing in the data-center. It describes the design to 19 deploy segment routing in the data-center, for both the MPLS and IPv6 20 dataplanes. 22 Requirements Language 24 The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", 25 "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this 26 document are to be interpreted as described in RFC 2119 [RFC2119]. 28 Status of This Memo 30 This Internet-Draft is submitted in full conformance with the 31 provisions of BCP 78 and BCP 79. 33 Internet-Drafts are working documents of the Internet Engineering 34 Task Force (IETF). Note that other groups may also distribute 35 working documents as Internet-Drafts. The list of current Internet- 36 Drafts is at http://datatracker.ietf.org/drafts/current/. 38 Internet-Drafts are draft documents valid for a maximum of six months 39 and may be updated, replaced, or obsoleted by other documents at any 40 time. It is inappropriate to use Internet-Drafts as reference 41 material or to cite them other than as "work in progress." 43 This Internet-Draft will expire on April 14, 2016. 45 Copyright Notice 47 Copyright (c) 2015 IETF Trust and the persons identified as the 48 document authors. All rights reserved. 50 This document is subject to BCP 78 and the IETF Trust's Legal 51 Provisions Relating to IETF Documents 52 (http://trustee.ietf.org/license-info) in effect on the date of 53 publication of this document. Please review these documents 54 carefully, as they describe your rights and restrictions with respect 55 to this document. Code Components extracted from this document must 56 include Simplified BSD License text as described in Section 4.e of 57 the Trust Legal Provisions and are provided without warranty as 58 described in the Simplified BSD License. 60 Table of Contents 62 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 3 63 2. Large Scale Data Center Network Design Summary . . . . . . . 3 64 2.1. Reference design . . . . . . . . . . . . . . . . . . . . 4 65 3. Some open problems in large data-center networks . . . . . . 5 66 4. Applying Segment Routing in the DC with MPLS dataplane . . . 6 67 4.1. BGP Prefix Segment . . . . . . . . . . . . . . . . . . . 6 68 4.2. eBGP Labeled Unicast (RFC3107) . . . . . . . . . . . . . 7 69 4.2.1. Control Plane . . . . . . . . . . . . . . . . . . . . 7 70 4.2.2. Data Plane . . . . . . . . . . . . . . . . . . . . . 9 71 4.2.3. Network Design Variation . . . . . . . . . . . . . . 10 72 4.2.4. Global BGP Prefix Segment through the fabric . . . . 10 73 4.2.5. Incremental Deployments . . . . . . . . . . . . . . . 11 74 4.3. iBGP Labeled Unicast (RFC3107) . . . . . . . . . . . . . 12 75 5. Applying Segment Routing in the DC with IPv6 dataplane . . . 12 76 6. Communicating path information to the host . . . . . . . . . 13 77 7. Addressing the open problems . . . . . . . . . . . . . . . . 14 78 7.1. Per-packet and flowlet switching . . . . . . . . . . . . 14 79 7.2. Performance-aware routing . . . . . . . . . . . . . . . . 15 80 7.3. Non-oblivious routing . . . . . . . . . . . . . . . . . . 16 81 7.4. Deterministic network probing . . . . . . . . . . . . . . 16 82 8. Additional Benefits . . . . . . . . . . . . . . . . . . . . . 16 83 8.1. MPLS Dataplane with operational simplicity . . . . . . . 16 84 8.2. Minimizing the FIB table . . . . . . . . . . . . . . . . 17 85 8.3. Egress Peer Engineering . . . . . . . . . . . . . . . . . 17 86 8.4. Incremental Deployments . . . . . . . . . . . . . . . . . 18 87 8.5. Anycast . . . . . . . . . . . . . . . . . . . . . . . . . 18 88 9. Preferred SRGB Allocation . . . . . . . . . . . . . . . . . . 18 89 10. Alternative Options . . . . . . . . . . . . . . . . . . . . . 19 90 11. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 20 91 12. Manageability Considerations . . . . . . . . . . . . . . . . 20 92 13. Security Considerations . . . . . . . . . . . . . . . . . . . 20 93 14. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . 20 94 15. References . . . . . . . . . . . . . . . . . . . . . . . . . 20 95 15.1. Normative References . . . . . . . . . . . . . . . . . . 20 96 15.2. Informative References . . . . . . . . . . . . . . . . . 20 97 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 21 99 1. Introduction 101 Segment Routing (SR), as described in 102 [I-D.filsfils-spring-segment-routing] leverages the source routing 103 paradigm. A node steers a packet through an ordered list of 104 instructions, called segments. A segment can represent any 105 instruction, topological or service-based. A segment can have a 106 local semantic to an SR node or global within an SR domain. SR 107 allows to enforce a flow through any topological path and service 108 chain while maintaining per-flow state only at the ingress node to 109 the SR domain. Segment Routing can be applied to the MPLS and IPv6 110 data-planes. 112 The use-case use-cases described in this document should be 113 considered in the context of the BGP-based large-scale data-center 114 (DC) design described in[I-D.ietf-rtgwg-bgp-routing-large-dc]We 115 extend it by applying SR both with IPv6 and MPLS dataplane. 117 2. Large Scale Data Center Network Design Summary 119 This section provides a brief summary of the informational document 120 [I-D.ietf-rtgwg-bgp-routing-large-dc] that outlines a practical 121 network design suitable for data-centers of various scales: 123 o Data-center networks have highly symmetric topologies with 124 multiple parallel paths between two server attachment points. The 125 well-known Clos topology is most popular among the operators. In 126 a Clos topology, the minimum number of parallel paths between two 127 elements is determined by the "width" of the middle stage. See 128 Figure 1 below for an illustration of the concept. 130 o Large-scale data-centers commonly use a routing protocol, such as 131 BGP4 [RFC4271] in order to provide endpoint connectivity. 132 Recovery after a network failure is therefore driven either by 133 local knowledge of directly available backup paths or by 134 distributed signaling between the network devices. 136 o Within data-center networks, traffic is load-shared using the 137 Equal Cost Multipath (ECMP) mechanism. With ECMP, every network 138 device implements a pseudo-random decision, mapping packets to one 139 of the parallel paths by means of a hash function calculated over 140 certain parts of the packet, typically a combination of various 141 packet header fields. 143 The following is a schematic of a five-stage Clos topology, with four 144 devices in the middle stage. Notice that number of paths between 145 Node1 and Node12 equals to four: the paths have to cross all of 146 Tier-1 devices. At the same time, the number of paths between Node1 147 and Node2 equals two, and the paths only cross Tier-2 devices. Other 148 topologies are possible, but for simplicity we'll only look into the 149 topologies that have a single path from Tier-1 to Tier-3. The rest 150 could be treated similarly, with a few modifications to the logic. 152 2.1. Reference design 154 Tier-1 155 +-----+ 156 |NODE | 157 +->| 5 |--+ 158 | +-----+ | 159 Tier-2 | | Tier-2 160 +-----+ | +-----+ | +-----+ 161 +------------>|NODE |--+->|NODE |--+--|NODE |-------------+ 162 | +-----| 3 |--+ | 6 | +--| 9 |-----+ | 163 | | +-----+ +-----+ +-----+ | | 164 | | | | 165 | | +-----+ +-----+ +-----+ | | 166 | +-----+---->|NODE |--+ |NODE | +--|NODE |-----+-----+ | 167 | | | +---| 4 |--+->| 7 |--+--| 10 |--+ | | | 168 | | | | +-----+ | +-----+ | +-----+ | | | | 169 | | | | | | | | | | 170 +-----+ +-----+ | +-----+ | +-----+ +-----+ 171 |NODE | |NODE | Tier-3 +->|NODE |--+ Tier-3 |NODE | |NODE | 172 | 1 | | 2 | | 8 | | 11 | | 12 | 173 +-----+ +-----+ +-----+ +-----+ +-----+ 174 | | | | | | | | 175 A O B O <- Servers -> Z O O O 177 Figure 1: 5-stage Clos topology 179 In the reference topology illustrated in Figure 1, we assume: 181 o Each node is its own AS (Node X has AS X) 183 * For simple and efficient route propagation filtering, Nodes 5, 184 6, 7 and 8 share the same AS, Nodes 3 and 4 share the same AS, 185 Nodes 9 and 10 share the same AS. 187 * For efficient usage of the scarce 2-byte Private Use AS pool, 188 different Tier-3 nodes might share the same AS. 190 * Without loss of generality, we will simplify these details in 191 this document and assume that each node has its own AS. 193 o Each node peers with its neighbors via BGP session 195 * If not specified, eBGP is assumed. In a specific use-case, 196 iBGP will be used but this will be called out explicitly in 197 that case. 199 o Each node originates the IPv4 address of it's loopback interface 200 into BGP and announces it to its neighbors. 202 * The loopback of Node X is 192.0.2.x/32. 204 In this document, we also refer to the Tier-1, Tier-2 and Tier-3 205 switches respectively as Spine, Leaf and ToR (top of rack) switches. 206 When a ToR switch acts as a gateway to the "outside world", we call 207 it a border switch. 209 3. Some open problems in large data-center networks 211 The data-center network design summarized above provides means for 212 moving traffic between hosts with reasonable efficiency. There are 213 few open performance and reliability problems that arise in such 214 design: 216 o ECMP routing is most commonly realized per-flow. This means that 217 large, long-lived "elephant" flows may affect performance of 218 smaller, short-lived "mouse" flows and reduce efficiency of per- 219 flow load-sharing. In other words, per-flow ECMP that does not 220 perform efficiently when flow life-time distribution is heavy- 221 tailed. Furthermore, due to hash-function inefficiencies it is 222 possible to have frequent flow collisions, where more flows get 223 placed on one path over the others 225 o Shortest-path routing with ECMP implements oblivious routing 226 model, which is not aware of the network imbalances. If the 227 network symmetry is broken, for example due to link failures, 228 utilization hotspots may appear. For example, if a link fails 229 between Tier-1 and Tier-2 devices (e.g. "Node5" and "Node9"), 230 Tier-3 devices "Node1" and "Node2" will not be aware of that, 231 since there are other paths available from perspective of "Node3". 232 They will continue sending roughly equal traffic to Node3 and 233 Node4 as if the failure didn't exist which may cause a traffic 234 hotspot. 236 o Absence of path visibility leaves transport protocols, such as 237 TCP, with a "blackbox" view of the network. Some TCP metrics, 238 such as SRTT, MSS, CWND and few others could be inferred and 239 cached based on past history, but those apply to destinations, 240 regardless of the path that has been chosen to get there. Thus, 241 for instance, TCP is not capable of remembering "bad" paths, such 242 as those that exhibited poor performance in the past. This means 243 that every new connection will be established obliviously (memory- 244 less) with regards to the paths chosen before, or chosen by other 245 nodes. 247 o Isolating faults in the network with multiple parallel paths and 248 ECMP-based routing is non-trivial due to lack of determinism. 249 Specifically, the connections from HostA to HostB may take a 250 different path every time a new connection is formed, thus making 251 consistent reproduction of a failure much more difficult. This 252 complexity scales linearly with the number of parallel paths in 253 the network, and stems from the random nature of path selection by 254 the network devices. 256 Further in this document, we are going to demonstrate how these 257 problems could be addressed within the framework of Segment Routing. 259 First, we will explain how to apply SR in the DC, for MPLS and IPv6 260 data-planes. 262 4. Applying Segment Routing in the DC with MPLS dataplane 264 4.1. BGP Prefix Segment 266 A BGP-Prefix Segment is a segment associated with a BGP prefix. A 267 BGP-Prefix Segment is a network-wide instruction to forward the 268 packet along the ECMP-aware best path to the related prefix 269 ([I-D.keyupate-idr-bgp-prefix-sid]). 271 In this document, we make the network design decision to assume that 272 all the nodes are allocated the same SRGB, e.g. [16000, 23999]. This 273 is important to fulfill the recommendation for operational 274 simplification as explained in [I-D.filsfils-spring-segment-routing]. 276 Note well that the use of a common SRGB in all nodes is not a 277 requirement, one could use a different SRGB at every node. However, 278 this would make the operation of the DC fabric more complex as the 279 label allocated to the loopback of a remote switch is then different 280 at every node. This also may increase the complexity of the 281 centralized controller. 283 For illustration purpose, when considering an MPLS data-plane, we 284 assume that the segment index allocated to prefix 192.0.2.x/32 is X. 285 As a result, a local label 1600x is allocated for prefix 192.0.2.x/32 286 by each node throughout the DC fabric. 288 When IPv6 data-plane is considered, we assume that Node X is 289 allocated IPv6 address (segment) 2001:DB8::X. 291 4.2. eBGP Labeled Unicast (RFC3107) 293 Referring to Figure 1 and [[I-D.ietf-rtgwg-bgp-routing-large-dc], the 294 following design modifications are introduced: 296 o Each node peers with its neighbors via eBGP3107 session 298 o The forwarding plane at Tier-2 and Tier-1 is MPLS. 300 o The forwarding plane at Tier-3 is either IP2MPLS (if the host 301 sends IP traffic) or MPLS2MPLS (if the host sends MPLS- 302 encapsulated traffic). 304 Figure 2 zooms on a path from server A to server Z within the 305 topology of Figure 1. 307 +-----+ +-----+ +-----+ 308 +---------->|NODE | |NODE | |NODE | 309 | | 4 |--+->| 7 |--+--| 10 |---+ 310 | +-----+ +-----+ +-----+ | 311 | | 312 +-----+ +-----+ 313 |NODE | |NODE | 314 | 1 | | 11 | 315 +-----+ +-----+ 316 | | 317 A <- Servers -> Z 319 Figure 2: Path from A to Z via nodes 1, 4, 7, 10 and 11 321 Referring to Figure 1 and Figure 2 and assuming the IP address, AS 322 and index allocation previously described, the following sections 323 detail the control plane operation and the data plane states for the 324 prefix 192.0.2.11/32 (loopback of Node11) 326 4.2.1. Control Plane 328 Node11 originates 192.0.2.11/32 in BGP and allocates to it the BGP- 329 Prefix Segment attribute (index11). 331 Node11 sends the following eBGP3107 update to Node10: 333 . NLRI: 192.0.2.11/32 334 . Label: Implicit-Null 335 . Next-hop: Node11's interface address on the link to Node10 336 . AS Path: {11} 337 . BGP-Prefix Attribute: Index 11 339 Node10 receives the above update. As it is SR capable, Node10 is 340 able to interpret the BGP-Prefix Attribute and hence understands that 341 it should allocate the label LOCAL-SRGB (16000) + "index" 11 (hence 342 16011) to the NLRI instead of allocating an nondeterministic label 343 out of a dynamically allocated portion of the local label space. The 344 implicit-null label in the NLRI tells Node10 that it is the 345 penultimate hop and MUST pop the top label on the stack before 346 forwarding traffic for this prefix to Node11. 348 Then, Node10 sends the following eBGP3107 update to Node7: 350 . NLRI: 192.0.2.11/32 351 . Label: 16011 352 . Next-hop: Node10's interface address on the link to Node7 353 . AS Path: {10, 11} 354 . BGP-Prefix Attribute: Index 11 356 Node7 receives the above update. As it is SR capable, Node7 is able 357 to interpret the BGP-Prefix Attribute and hence allocates the local 358 (incoming) label 16011 (16000 + 11) to the NLRI (instead of 359 allocating a "dynamic" local label from its label manager). Node7 360 uses the label in the received eBGP3107 NLRI as the outgoing label 361 (the index is only used to derive the local/incoming label). 363 Node7 sends the following eBGP3107 update to Node4: 365 . NLRI: 192.0.2.11/32 366 . Label: 16011 367 . Next-hop: Node7's interface address on the link to Node4 368 . AS Path: {7, 10, 11} 369 . BGP-Prefix Attribute: Index 11 371 Node4 receives the above update. As it is SR capable, Node4 is able 372 to interpret the BGP-Prefix Attribute and hence allocates the local 373 (incoming) label 16011 to the NLRI (instead of allocating a "dynamic" 374 local label from its label manager). Node4 uses the label in the 375 received eBGP3107 NLRI as outgoing label (the index is only used to 376 derive the local/incoming label). 378 Node4 sends the following eBGP3107 update to Node1: 380 . NLRI: 192.0.2.11/32 381 . Label: 16011 382 . Next-hop: Node4's interface address on the link to Node1 383 . AS Path: {4, 7, 10, 11} 384 . BGP-Prefix Attribute: Index 11 386 Node1 receives the above update. As it is SR capable, Node1 is able 387 to interpret the BGP-Prefix Attribute and hence allocates the local 388 (incoming) label 16011 to the NLRI (instead of allocating a "dynamic" 389 local label from its label manager). Node1 uses the label in the 390 received eBGP3107 NLRI as outgoing label (the index is only used to 391 derive the local/incoming label). 393 4.2.2. Data Plane 395 Referring to Figure 1Referring to Figure 1, and assuming all nodes 396 apply the same advertisement rules described above and all nodes have 397 the same SRGB (16000-23999), here are the IP/MPLS forwarding tables 398 for prefix 192.0.2.11/32 at Nodes 1, 4, 7 and 10. 400 ----------------------------------------------- 401 Incoming label | outgoing label | Outgoing 402 or IP destination | | Interface 403 ------------------+----------------+----------- 404 16011 | 16011 | ECMP{3, 4} 405 192.0.2.11/32 | 16011 | ECMP{3, 4} 406 ------------------+----------------+----------- 408 Figure 3: Node1 Forwarding Table 410 ----------------------------------------------- 411 Incoming label | outgoing label | Outgoing 412 or IP destination | | Interface 413 ------------------+----------------+----------- 414 16011 | 16011 | ECMP{7, 8} 415 192.0.2.11/32 | 16011 | ECMP{7, 8} 416 ------------------+----------------+----------- 418 Figure 4: Node4 Forwarding Table 420 ----------------------------------------------- 421 Incoming label | outgoing label | Outgoing 422 or IP destination | | Interface 423 ------------------+----------------+----------- 424 16011 | 16011 | 10 425 192.0.2.11/32 | 16011 | 10 426 ------------------+----------------+----------- 428 Figure 5: Node7 Forwarding Table 430 ----------------------------------------------- 431 Incoming label | outgoing label | Outgoing 432 or IP destination | | Interface 433 ------------------+----------------+----------- 434 16011 | POP | 11 435 192.0.2.11/32 | N/A | 11 436 ------------------+----------------+----------- 438 Node10 Forwarding Table 440 4.2.3. Network Design Variation 442 A network design choice could consist of switching all the traffic 443 through Tier-1 and Tier-2 as MPLS traffic. In this case, one could 444 filter away the IP entries at Nodes 4, 7 and 10. This might be 445 beneficial in order to optimize the forwarding table size. 447 A network design choice could consist in allowing the hosts to send 448 MPLS-encapsulated traffic (based on EPE use-case, 449 [I-D.filsfils-spring-segment-routing-central-epe]). For example, 450 applications at HostA would send their Z-destined traffic to Node1 451 with an MPLS label stack where the top label is 16011 and the next 452 label is an EPE peer segment at Node11 directing the traffic to Z. 454 4.2.4. Global BGP Prefix Segment through the fabric 456 When the previous design is deployed, the operator enjoys global BGP 457 prefix segment (label) allocation throughout the DC fabric. 459 A few examples follow: 461 o Normal forwarding to Node11: a packet with top label 16011 462 received by any switch in the fabric will be forwarded along the 463 ECMP-aware BGP best-path towards Node11 and the label 16011 is 464 penultimate-popped at Node10. 466 o Traffic-engineered path to Node11: an application on a host behind 467 Node1 might want to restrict its traffic to paths via the Spine 468 switch Node5. The application achieves this by sending its 469 packets with a label stack of {16005, 16011}. BGP Prefix segment 470 16005 directs the packet up to Node5 along the path (Node1, Node3, 471 Node5). BGP Prefix Segment 16011 then directs the packet down to 472 Node11 along the path (Node5, Node9, Node11). 474 4.2.5. Incremental Deployments 476 The design previously described can be deployed incrementally. Let 477 us assume that Node7 does not support the BGP-Prefix Segment 478 attribute and let us show how the fabric connectivity is preserved. 480 From a signaling viewpoint, nothing would change: if Node7 does not 481 understand the BGP-Prefix Segment attribute, it does propagate the 482 attribute unmodified to its neighbors. 484 From a label allocation viewpoint, the only difference is that Node7 485 would allocate a dynamic (random) label to the prefix 192.0.2.11/32 486 (e.g. 123456) instead of the "hinted" label as instructed by the BGP 487 Prefix Segment attribute. The neighbors of Node7 adapt automatically 488 as they always use the label in the BGP3107 NLRI as outgoing label. 490 Node4 does understand the BGP-Prefix Segment attribute and hence 491 allocates the indexed label in the SRGB (16011) for 192.0.2.11/32. 493 As a result, all the data-plane entries across the network would be 494 unchanged except the entries at Node7 and its neighbor Node4 as shown 495 in the figures below. 497 The key point is that the end-to-end LSP is preserved because the 498 outgoing label is always derived from the received label within the 499 BGP3107 NLRI. The index in the BGP Prefix SID is only used as a hint 500 on how to allocate the local label (the incoming label) but never for 501 the outgoing label. 503 ------------------------------------------ 504 Incoming label | outgoing | Outgoing 505 or IP destination | label | Interface 506 -------------------+---------------------- 507 12345 | 16011 | 10 509 Figure 7: Node7 Forwarding Table 511 ------------------------------------------ 512 Incoming label | outgoing | Outgoing 513 or IP destination | label | Interface 514 -------------------+---------------------- 515 16011 | 12345 | 7 517 Figure 8: Node4 Forwarding Table 519 The BGP-Prefix Segment functionality can thus be deployed 520 incrementally one node at a time. 522 When deployed together with a homogeneous SRGB (same SRGB across the 523 fabric), the operator incrementally enjoys the global prefix segment 524 benefits as the deployment progresses through the fabric. 526 4.3. iBGP Labeled Unicast (RFC3107) 528 The same exact design as eBGP3107 is used with the following 529 modifications: 531 All switches share the same AS 533 iBGP3107 reflection with nhop-self is used instead of eBGP3107 535 For simple and efficient route propagation filtering, Nodes 5, 6, 536 7 and 8 share the same Cluster ID, Nodes 3 and 4 share the same 537 Cluster ID, Nodes 9 and 10 share the same Cluster ID. 539 AIGP metric ([RFC7311]) is likely applied to the BGP prefix 540 segments as part of a large-scale multi-domain design such as 541 Seamless MPLS [I-D.ietf-mpls-seamless-mpls]. 543 The control-plane behavior is mostly the same as described in the 544 previous section: the only difference is that the eBGP3107 path 545 propagation is simply replaced by an iBGP3107 path reflection with 546 next-hop changed to self. 548 The data-plane tables are exactly the same. 550 5. Applying Segment Routing in the DC with IPv6 dataplane 552 The design described in I-D.ietf-rtgwg-bgp-routing-large-dc 553 [I-D.ietf-rtgwg-bgp-routing-large-dc] is reused with one single 554 modification. We highlight it using the example of the reachability 555 to Node11 via spine switch Node5. 557 Spine5 originates 2001:DB8::5/128 with the attached BGP Prefix 558 Attribute adverting the support of the Segment Routing extension 559 header (SRH, [I-D.previdi-6man-segment-routing-header]) for IPv6 560 packets destined to segment 2001:DB8::5. 562 Tor11 originates 2001:DB8::11/128 with the attached BGP Prefix 563 Attribute adverting the support of the Segment Routing extension 564 header (SRH, [I-D.previdi-6man-segment-routing-header]) for IPv6 565 packets destined to segment 2001:DB8::11. 567 The control-plane and data-plane processing of all the other nodes in 568 the fabric is unchanged. Specifically, the routes to 2001:DB8::5 and 569 2001:DB8::11 are installed in the FIB along the eBGP best-path to 570 Node5 (spine node) and Node11 (ToR node) respectively. 572 An application on HostA which needs to send traffic to HostZ via only 573 Node5 (spine node) can do so by sending IPv6 packets with a SR 574 extension header. The destination address and active segment is set 575 to 2001:DB8::5. The next and last segment is set to 2001:DB8::11. 577 The application must only use IPv6 addresses that have been 578 advertised as capable for SRv6 segment processing (e.g. for which the 579 BGP prefix segment capability has been advertised). How applications 580 learn this (e.g.: centralized controller and orchestration) is 581 outside the scope of this document. 583 6. Communicating path information to the host 585 There are two general methods for communicating path information to 586 the end-hosts: "proactive" and "reactive", aka "push" and "pull" 587 models. There are multiple ways to implement either of these 588 methods. Here, we note that one way could be using a centralized 589 controller: the controller either tells the hosts of the prefix-to- 590 path mappings beforehand and updates them as needed (network event 591 driven push), or responds to the hosts making request for a path to 592 specific destination (host event driven pull). It is also possible 593 to use a hybrid model, i.e., pushing some state from the controller 594 in response to particular network events, while the host pulls other 595 state on demand. 597 We note, that when disseminating network-related data to the end- 598 hosts a trade-off is made to balance the amount of information vs the 599 level of visibility in the network state. This applies both to push 600 and pull models. In the extreme case, the host would request path 601 information on every flow, and keep no local state at all. On the 602 other end of the spectrum, information for every prefix in the 603 network along with available paths could be pushed and continuously 604 updated on all hosts. 606 7. Addressing the open problems 608 This section demonstrates how the problems describe above could be 609 solved using the segment routing concept. It is worth noting that 610 segment routing signaling and data-plane are only parts of the 611 solution. Additional enhancements, e.g. such as the centralized 612 controller mentioned previously, and host networking stack support 613 are required to implement the proposed solutions. 615 7.1. Per-packet and flowlet switching 617 With the ability to choose paths on the host, one may go from per- 618 flow load-sharing in the network to per-packet or per-flowlet (see 619 [KANDULA04] for information on flowlets). The host may select 620 different segment routing instructions either per packet, or per 621 flowlet, and route them over different paths. This allows for 622 solving the "elephant flow" problem in the data-center and avoiding 623 link imbalances. 625 Note that traditional ECMP routing could be easily simulated with on- 626 host path selection, using method proposed in VL2 (see 627 [GREENBERG09]). The hosts would randomly pick a Tier-2 or Tier-1 628 device to "bounce" the packet off of, depending on whether the 629 destination is under the same Tier-2 switches, or has to be reached 630 across Tier-1. The host would use a hash function that operates on 631 per-flow invariants, to simulate per-flow load-sharing in the 632 network. 634 Using Figure 1 as reference, let's illustrate this assuming that 635 HostA has an elephant flow to Z called Flow-f. 637 Normally, a flow is hashed on to a single path. Let's assume HostA 638 sends its packets associated with Flow-f with top label 16011 (the 639 label for the remote ToR, Node11, where HostZ is connected) and Node1 640 would hash all the packets of Flow-F via the same nhop (e.g. Node3). 641 Similarly, let's assume that leaf Node3 would hash all the packets of 642 Flow-F via the same next-hop (e.g.: spine switch Node1). This normal 643 operation would restrict the elephant flow on a small subset of the 644 ECMP paths to HostZ and potentially create imbalance and congestion 645 in the fabric. 647 Leveraging the flowlet proposal, assuming A is made aware of 4 648 disjoint paths via intermediate segment 16005, 16006, 16007 and 16008 649 (the BGP prefix SID's of the 4 spine switches) and also made aware of 650 the prefix segment of the remote ToR connected to the destination 651 (16011), then the application can break the elephant flow F into 652 flowlets F1, F2, F3, F4 and associate each flowlet with one of the 653 following 4 label stacks: {16005, 16011}, {16006, 16011}, {16007, 654 16011} and {16008, 16011}. This would spread the load of the elephant 655 flow through all the ECMP paths available in the fabric and rebalance 656 the load. 658 7.2. Performance-aware routing 660 Knowing the path associated with flows/packets, the end host may 661 deduce certain characteristics of the path on its own, and 662 additionally use the information supplied with path information 663 pushed from the controller or received via pull request. The host 664 may further share its path observations with the centralized agent, 665 so that the latter may keep up-to-date network health map to assist 666 other hosts with this information. 668 For example, an application A.1 at HostA may pin a TCP flow destined 669 to HostZ via Spine switch Node5 using label stack {16005, 16011}. The 670 application A.1 may collect information on packet loss, deduced from 671 TCP retransmissions and other signals (e.g. RTT increases). A.1 may 672 additionally publish this information to a centralized agent, e.g. 673 after a flow completes, or periodically for longer lived flows. 674 Next, using both local and/or global performance data, application 675 A.1 as well as other applications sharing the same resources in the 676 DC fabric may pick up the best path for the new flow, or update an 677 existing path (e.g.: when informed of congestion on an existing 678 path). 680 One particularly interesting instance of performance-aware routing is 681 dynamic fault-avoidance. If some links or devices in the network 682 start discarding packets due to a fault, the end-hosts could detect 683 the path(s) being affected and steer their flows away from the 684 problem spot. Similar logic applies to failure cases where packets 685 get completely black-holed, e.g. when a link goes down. 687 For example, an application A.1 informed about 5 paths to Z {16005, 688 16011}, {16006, 16011}, {16007, 16011}, {16008, 16011} and {16011} 689 might use the latter one by default (for simplicity). When 690 performance is degrading, A.1 might then start to pin TCP flows to 691 each of the 4 other paths (each via a distinct spine) and monitor the 692 performance. It would then detect the faulty path and assign a 693 negative preference to the faulty path to avoid further flows using 694 it. Gradually, over time, it may re-assign flows on the faulty path 695 to eventually detect the resolution of the trouble and start reusing 696 the path. 698 7.3. Non-oblivious routing 700 By leveraging Segment Routing, one avoids issues associated with 701 oblivious ECMP hashing. For example, if in the topology depicted on 702 Figure 1 a link between spine switch Node5 and leaf node Node9 fails, 703 HostA may exclude the segment corresponding to Node5 from the prefix 704 matching the servers under Tier-2 devices Node9. In the push path 705 discovery model, the affected path mappings may be explicitly pushed 706 to all the servers for the duration of the failure. The new mapping 707 would instruct them to avoid the particular Tier-1 switch until the 708 link has recovered. Alternatively, in pull path, the centralized 709 controller may start steering new flows immediately after it 710 discovers the issue. Until then, the existing flows may recover 711 using local detection of the path issues, as described in 712 Section 7.2. 714 7.4. Deterministic network probing 716 Active probing is a well-known technique for monitoring network 717 elements health, constituting of sending continuous packet streams 718 simulating network traffic to the hosts in the data-center. Segment 719 routing makes possible to prescribe the exact paths that each probe 720 or series of probes would be taking toward their destination. This 721 allows for fast correlation and detection of failed paths, by 722 processing information from multiple actively probing agents. This 723 complements the data collected from the hosts routing stacks as 724 described inSection 7.2. 726 For example, imagine a probe agent sending packets to all machines in 727 the data-center. For every host, it may send packets over each of 728 the possible paths, knowing exactly which links and devices these 729 packets will be crossing. Correlating results for multiple 730 destinations with the topological data, it may automatically isolate 731 possible problem to a link or device in the network. 733 8. Additional Benefits 735 8.1. MPLS Dataplane with operational simplicity 737 As required by [I-D.ietf-rtgwg-bgp-routing-large-dc], no new 738 signaling protocol is introduced. The Prefix Segment is a 739 lightweight extension to BGP Labelled Unicast (RFC3107 [RFC3107]). 740 It applies either to eBGP or iBGP based designs. 742 Specifically, LDP and RSVP-TE are not used. These protocols would 743 drastically impact the operational complexity of the Data Center and 744 would not scale. This is in line with the requirements expressed in 745 [I-D.ietf-rtgwg-bgp-routing-large-dc] 746 A key element of the operational simplicity is the deployment of the 747 design with a single and consistent SRGB across the DC fabric. 749 At every node in the fabric, the same label is associated to a given 750 BGP prefix segment and hence a notion of global prefix segment 751 arises. 753 When a controller programs HostA to send traffic to HostZ via the 754 normally available BGP ECMP paths, the controller uses label 16011 755 associated with the ToR switch connected to the HostZ. The 756 controller does not need to pick the label based on the ToR that the 757 source host is connected to. 759 In a classic BGP Labelled Unicast design applied to the DC fabric 760 illustrated in Figure 1, the ToR Node1 connected to HostA would most 761 likely allocate a different label for 192.0.2.11/32 than the one 762 allocated by ToR Node2. As a consequence, the controller would need 763 to adapt the SR policy to each host, based on the ToR switch that 764 they are connected to. This adds state maintenance and 765 synchronization problems. All of this unnecessary complexity is 766 eliminated if a single consistent SRGB is utilized across the fabric. 768 8.2. Minimizing the FIB table 770 The designer may decide to switch all the traffic at Tier-1 and Tier- 771 2's based on MPLS, hence drastically decreasing the IP table size at 772 these nodes. 774 This is easily accomplished by encapsulating the traffic either 775 directly at the host or at the source ToR switch by pushing the BGP- 776 Prefix Segment of the destination ToR for intra-DC traffic or border 777 switch for inter-DC or DC-to-outside-world traffic. 779 8.3. Egress Peer Engineering 781 It is straightforward to combine the design illustrated in this 782 document with the Egress Peer Engineering (EPE) use-case described in 783 [I-D.filsfils-spring-segment-routing-central-epe]. 785 In such case, the operator is able to engineer its outbound traffic 786 on a per host-flow basis, without incurring any additional state at 787 intermediate points in the DC fabric. 789 For example, the controller only needs to inject a per-flow state on 790 the HostA to force it to send its traffic destined to a specific 791 Internet destination D via a selected border switch (say Node12 in 792 Figure 1 instead of another border switch Node11) and a specific 793 egress peer of Node12 (say peer AS 9999 of local PeerNode segment 794 9999 at Node12 instead of any other peer which provides a path to the 795 destination D). Any packet matching this state at host A would be 796 encapsulated with SR segment list (label stack) {16012, 9999}. 16012 797 would steer the flow through the DC fabric, leveraging any ECMP, 798 along the best path to border switch Node12. Once the flow gets to 799 border switch Node12, the active segment is 9999 (thanks to PHP on 800 the upstream neighbor of Node12). This EPE PeerNode segment forces 801 border switch Node12 to forward the packet to peer AS 9999, without 802 any IP lookup at the border switch. There is no per-flow state for 803 this engineered flow in the DC fabric. A benefit of segment routing 804 is the per-flow state is only required at the source. 806 As well as allowing full traffic engineering control such a design 807 also offers FIB table minimization benefits as the Internet- scale 808 FIB at border switch Node12 is not required if all FIB lookups are 809 avoided there by using EPE. 811 8.4. Incremental Deployments 813 As explained in Section 4.2.5, this design can be deployed 814 incrementally. 816 8.5. Anycast 818 The design presented in this document preserves the availability and 819 load-balancing properties of the base design presented in 820 [I-D.filsfils-spring-segment-routing]. 822 For example, one could assign an anycast loopback 192.0.2.20/32 and 823 associate segment index 20 to it on the border switches 11 and 12 (in 824 addition to their node-specific loopbacks). Doing so, the EPE 825 controller could express a default "go-to-the- Internet via any 826 border switch" policy as segment list {16020}. Indeed, from any host 827 in the DC fabric or from any ToR switch, 16020 steers the packet 828 towards the border switches 11 or 12 leveraging ECMP where available 829 along the best paths to these switches. 831 9. Preferred SRGB Allocation 833 In the MPLS case, we do not recommend to use different SRGBs at each 834 node. 836 Different SRGBs in each node likely increase the complexity of the 837 solution both from an operation viewpoint and from a controller 838 viewpoint. 840 From an operation viewpoint, it is much simpler to have the same 841 global label at every node for the same destination (the MPLS 842 troubleshooting is then similar to the IPv6 troubleshooting where 843 this global property is a given). 845 From a controller viewpoint, this allows to construct simple policies 846 applicable across the fabric. 848 Let us consider two applications A and B respectively connected to 849 ToR1 and ToR2. A has two flows FA1 and FA2 destined to Z. B has two 850 flows FB1 and FB2 destined to Z. The controller wants FA1 and FB1 to 851 be load-shared across the fabric while FA2 and FB2 must be 852 respectively steered via Spine5 and spine 8. 854 Assuming a consistent unique SRGB across the fabric as described in 855 the document, the controller can simply do it by instructing A and B 856 to use {16011} respectively for FA1 and FB1 and by instructing A and 857 B to use {16005 16011} and {16008 16011} respectively for FA2 and 858 FB2. 860 Let us assume a design where the SRGB is different at every node: 861 SRGB of Node K starts at value K*1000 and the SRGB length is 1000 862 (e.g. ToR1's SRGB is [1000, 1999], ToR2's SRGB is [2000, 2999]...). 864 In this case, not only the controller would need to collect and store 865 all of these different SRGB's, furthermore it would need to adapt the 866 policy for each host. Indeed, the controller would instruct A to use 867 {1011} for FA1 while it would have to instruct B to use {2011} for 868 FB1 (while with the same SRGB, both policies are the same {16011}). 870 Even worse, the controller would instruct A to use {1005, 5011} for 871 FA1 while it would instruct B to use {2011, 8011} for FB1 (while with 872 the same SRGB, the second segment is the same across both policies: 873 16011). When combining segments to create a policy, one need to 874 carefully update the label of each segment. This is obviously more 875 error-prone, more complex and more difficult to troubleshoot. 877 10. Alternative Options 879 In order to support all the requirements and get consensus, the BGP 880 Prefix SID attribute has been extended to allow this design. 882 Specifically, the ORIGINATOR_SRGB TLV in the BGP Prefix SID signals 883 the SRGB of the switch that originated the BGP Prefix Segment. 885 This allows to determine the local label allocated by any switch for 886 any BGP Prefix Segment, despite the lack of a consistent unique SRGB 887 in the domain. 889 11. IANA Considerations 891 TBD 893 12. Manageability Considerations 895 TBD 897 13. Security Considerations 899 TBD 901 14. Acknowledgements 903 The authors would like to thank Benjamin Black, Arjun Sreekantiah, 904 Keyur Patel and Acee Lindem for their comments and review of this 905 document. 907 15. References 909 15.1. Normative References 911 [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate 912 Requirement Levels", BCP 14, RFC 2119, 913 DOI 10.17487/RFC2119, March 1997, 914 . 916 [RFC3107] Rekhter, Y. and E. Rosen, "Carrying Label Information in 917 BGP-4", RFC 3107, DOI 10.17487/RFC3107, May 2001, 918 . 920 [RFC4271] Rekhter, Y., Ed., Li, T., Ed., and S. Hares, Ed., "A 921 Border Gateway Protocol 4 (BGP-4)", RFC 4271, 922 DOI 10.17487/RFC4271, January 2006, 923 . 925 [RFC7311] Mohapatra, P., Fernando, R., Rosen, E., and J. Uttaro, 926 "The Accumulated IGP Metric Attribute for BGP", RFC 7311, 927 DOI 10.17487/RFC7311, August 2014, 928 . 930 15.2. Informative References 932 [GREENBERG09] 933 Greenberg, A., Hamilton, J., Jain, N., Kadula, S., Kim, 934 C., Lahiri, P., Maltz, D., Patel, P., and S. Sengupta, 935 "VL2: A Scalable and Flexible Data Center Network", 2009. 937 [I-D.filsfils-spring-segment-routing] 938 Filsfils, C., Previdi, S., Bashandy, A., Decraene, B., 939 Litkowski, S., Horneffer, M., Milojevic, I., Shakir, R., 940 Ytti, S., Henderickx, W., Tantsura, J., and E. Crabbe, 941 "Segment Routing Architecture", draft-filsfils-spring- 942 segment-routing-04 (work in progress), July 2014. 944 [I-D.filsfils-spring-segment-routing-central-epe] 945 Filsfils, C., Previdi, S., Patel, K., Shaw, S., Ginsburg, 946 D., and D. Afanasiev, "Segment Routing Centralized Egress 947 Peer Engineering", draft-filsfils-spring-segment-routing- 948 central-epe-05 (work in progress), August 2015. 950 [I-D.ietf-mpls-seamless-mpls] 951 Leymann, N., Decraene, B., Filsfils, C., Konstantynowicz, 952 M., and D. Steinberg, "Seamless MPLS Architecture", draft- 953 ietf-mpls-seamless-mpls-07 (work in progress), June 2014. 955 [I-D.ietf-rtgwg-bgp-routing-large-dc] 956 Lapukhov, P., Premji, A., and J. Mitchell, "Use of BGP for 957 routing in large-scale data centers", draft-ietf-rtgwg- 958 bgp-routing-large-dc-07 (work in progress), August 2015. 960 [I-D.keyupate-idr-bgp-prefix-sid] 961 Patel, K., Previdi, S., Filsfils, C., Sreekantiah, A., 962 Ray, S., and H. Gredler, "Segment Routing Prefix SID 963 extensions for BGP", draft-keyupate-idr-bgp-prefix-sid-05 964 (work in progress), July 2015. 966 [I-D.previdi-6man-segment-routing-header] 967 Previdi, S., Filsfils, C., Field, B., Leung, I., Linkova, 968 J., Kosugi, T., Vyncke, E., and D. Lebrun, "IPv6 Segment 969 Routing Header (SRH)", draft-previdi-6man-segment-routing- 970 header-08 (work in progress), October 2015. 972 [KANDULA04] 973 Sinha, S., Kandula, S., and D. Katabi, "Harnessing TCP's 974 Burstiness with Flowlet Switching", 2004. 976 Authors' Addresses 978 Clarence Filsfils (editor) 979 Cisco Systems, Inc. 980 Brussels 981 BE 983 Email: cfilsfil@cisco.com 984 Stefano Previdi (editor) 985 Cisco Systems, Inc. 986 Rome 00142 987 Italy 989 Email: sprevidi@cisco.com 991 Jon Mitchell 992 Unaffiliated 993 US 995 Email: jrmitche@puck.nether.net 997 Ebben Aries 998 Facebook 999 US 1001 Email: exa@fb.com 1003 Petr Lapukhov 1004 Facebook 1005 US 1007 Email: petr@fb.com 1009 Dmitry Afanasiev 1010 Yandex 1011 RU 1013 Email: fl0w@yandex-team.ru 1015 Tim Laberge 1016 Microsoft 1017 US 1019 Email: tim.laberge@microsoft.com 1021 Edet Nkposong 1022 Microsoft 1023 US 1025 Email: edetn@microsoft.com 1026 Mohan Nanduri 1027 Microsoft 1028 US 1030 Email: mnanduri@microsoft.com 1032 James Uttaro 1033 ATT 1034 US 1036 Email: ju1738@att.com 1038 Saikat Ray 1039 Unaffiliated 1040 US 1042 Email: raysaikat@gmail.com