idnits 2.17.1 draft-ietf-rift-applicability-10.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 (16 December 2021) is 862 days in the past. Is this intentional? Checking references for intended status: Informational ---------------------------------------------------------------------------- == Outdated reference: A later version (-21) exists of draft-ietf-rift-rift-13 Summary: 0 errors (**), 0 flaws (~~), 2 warnings (==), 1 comment (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 2 RIFT WG Yuehua. Wei, Ed. 3 Internet-Draft Zheng. Zhang 4 Intended status: Informational ZTE Corporation 5 Expires: 19 June 2022 Dmitry. Afanasiev 6 Yandex 7 P. Thubert 8 Cisco Systems 9 T. Przygienda 10 Juniper Networks 11 16 December 2021 13 RIFT Applicability 14 draft-ietf-rift-applicability-10 16 Abstract 18 This document discusses the properties, applicability and operational 19 considerations of RIFT in different network scenarios. It intends to 20 provide a rough guide how RIFT can be deployed to simplify routing 21 operations in Clos topologies and their variations. 23 Status of This Memo 25 This Internet-Draft is submitted in full conformance with the 26 provisions of BCP 78 and BCP 79. 28 Internet-Drafts are working documents of the Internet Engineering 29 Task Force (IETF). Note that other groups may also distribute 30 working documents as Internet-Drafts. The list of current Internet- 31 Drafts is at https://datatracker.ietf.org/drafts/current/. 33 Internet-Drafts are draft documents valid for a maximum of six months 34 and may be updated, replaced, or obsoleted by other documents at any 35 time. It is inappropriate to use Internet-Drafts as reference 36 material or to cite them other than as "work in progress." 38 This Internet-Draft will expire on 19 June 2022. 40 Copyright Notice 42 Copyright (c) 2021 IETF Trust and the persons identified as the 43 document authors. All rights reserved. 45 This document is subject to BCP 78 and the IETF Trust's Legal 46 Provisions Relating to IETF Documents (https://trustee.ietf.org/ 47 license-info) in effect on the date of publication of this document. 48 Please review these documents carefully, as they describe your rights 49 and restrictions with respect to this document. Code Components 50 extracted from this document must include Revised BSD License text as 51 described in Section 4.e of the Trust Legal Provisions and are 52 provided without warranty as described in the Revised BSD License. 54 Table of Contents 56 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 3 57 2. Terminology . . . . . . . . . . . . . . . . . . . . . . . . . 3 58 3. Problem Statement of Routing in Modern IP Fabric Fat Tree 59 Networks . . . . . . . . . . . . . . . . . . . . . . . . 5 60 4. Applicability of RIFT to Clos IP Fabrics . . . . . . . . . . 5 61 4.1. Overview of RIFT . . . . . . . . . . . . . . . . . . . . 5 62 4.2. Applicable Topologies . . . . . . . . . . . . . . . . . . 8 63 4.2.1. Horizontal Links . . . . . . . . . . . . . . . . . . 8 64 4.2.2. Vertical Shortcuts . . . . . . . . . . . . . . . . . 9 65 4.2.3. Generalizing to any Directed Acyclic Graph . . . . . 9 66 4.2.4. Reachability of Internal Nodes in the Fabric . . . . 11 67 4.3. Use Cases . . . . . . . . . . . . . . . . . . . . . . . . 11 68 4.3.1. Data Center Topologies . . . . . . . . . . . . . . . 11 69 4.3.2. Metro Fabrics . . . . . . . . . . . . . . . . . . . . 12 70 4.3.3. Building Cabling . . . . . . . . . . . . . . . . . . 13 71 4.3.4. Internal Router Switching Fabrics . . . . . . . . . . 13 72 4.3.5. CloudCO . . . . . . . . . . . . . . . . . . . . . . . 13 73 5. Operational Considerations . . . . . . . . . . . . . . . . . 15 74 5.1. South Reflection . . . . . . . . . . . . . . . . . . . . 16 75 5.2. Suboptimal Routing on Link Failures . . . . . . . . . . . 16 76 5.3. Black-Holing on Link Failures . . . . . . . . . . . . . . 18 77 5.4. Zero Touch Provisioning (ZTP) . . . . . . . . . . . . . . 19 78 5.5. Mis-cabling Examples . . . . . . . . . . . . . . . . . . 20 79 5.6. Positive vs. Negative Disaggregation . . . . . . . . . . 22 80 5.7. Mobile Edge and Anycast . . . . . . . . . . . . . . . . . 24 81 5.8. IPv4 over IPv6 . . . . . . . . . . . . . . . . . . . . . 26 82 5.9. In-Band Reachability of Nodes . . . . . . . . . . . . . . 26 83 5.10. Dual Homing Servers . . . . . . . . . . . . . . . . . . . 28 84 5.11. Fabric With A Controller . . . . . . . . . . . . . . . . 28 85 5.11.1. Controller Attached to ToFs . . . . . . . . . . . . 29 86 5.11.2. Controller Attached to Leaf . . . . . . . . . . . . 29 87 5.12. Internet Connectivity Within Underlay . . . . . . . . . . 29 88 5.12.1. Internet Default on the Leaf . . . . . . . . . . . . 30 89 5.12.2. Internet Default on the ToFs . . . . . . . . . . . . 30 90 5.13. Subnet Mismatch and Address Families . . . . . . . . . . 30 91 5.14. Anycast Considerations . . . . . . . . . . . . . . . . . 30 92 5.15. IoT Applicability . . . . . . . . . . . . . . . . . . . . 31 93 5.16. Key Management . . . . . . . . . . . . . . . . . . . . . 32 94 6. Security Considerations . . . . . . . . . . . . . . . . . . . 32 95 7. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 33 96 8. Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . 33 97 9. Contributors . . . . . . . . . . . . . . . . . . . . . . . . 33 98 10. Normative References . . . . . . . . . . . . . . . . . . . . 33 99 11. Informative References . . . . . . . . . . . . . . . . . . . 34 100 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 35 102 1. Introduction 104 This document discusses the properties and applicability of "Routing 105 in Fat Trees" [RIFT] in different deployment scenarios and highlights 106 the operational simplicity of the technology compared to traditional 107 routing solutions. It also documents special considerations when 108 RIFT is used with or without overlays and/or controllers, and how 109 RIFT identifies topology mis-cablings and reroutes around node and 110 link failures. 112 2. Terminology 114 Clos/Fat Tree: 116 This document uses the terms Clos and Fat Tree interchangeably 117 whereas it always refers to a folded spine-and-leaf topology with 118 possibly multiple Points of Delivery (PoDs) and one or multiple Top 119 of Fabric (ToF) planes. 121 Directed Acyclic Graph (DAG): 123 A finite directed graph with no directed cycles (loops). If links in 124 a Clos are considered as either being all directed towards the top or 125 vice versa, each of such two graphs is a DAG. 127 Disaggregation: 129 Process in which a node decides to advertise more specific prefixes 130 Southwards, either positively to attract the corresponding traffic, 131 or negatively to repel it. Disaggregation is performed to prevent 132 black-holing and suboptimal routing to the more specific prefixes. 134 TIE: 136 This is an acronym for a "Topology Information Element". TIEs are 137 exchanged between RIFT nodes to describe parts of a network such as 138 links and address prefixes. A TIE has always a direction and a type. 139 North TIEs (sometimes abbreviated as N-TIEs) are used when dealing 140 with TIEs in the northbound representation and South-TIEs (sometimes 141 abbreviated as S-TIEs) for the southbound equivalent. TIEs have 142 different types such as node and prefix TIEs. 144 Node TIE: 146 This stands as acronym for a "Node Topology Information Element", 147 which contains all adjacencies the node discovered and information 148 about the node itself. Node TIE should NOT be confused with a North 149 TIE since "node" defines the type of TIE rather than its direction. 150 Consequently North Node TIEs and South Node TIEs exist. 152 Prefix TIE: 154 This is an acronym for a "Prefix Topology Information Element" and it 155 contains all prefixes directly attached to this node in case of a 156 North TIE and in case of South TIE the necessary default routes and 157 disaggregated routes the node advertises southbound. 159 South Reflection: 161 Often abbreviated just as "reflection", it defines a mechanism where 162 South Node TIEs are "reflected" from the level south back up north to 163 allow nodes in the same level without East- West links to "see" each 164 other's node Topology Information Elements (TIEs). 166 LIE: 168 This is an acronym for a "Link Information Element" exchanged on all 169 the system's links running RIFT to form ThreeWay adjacencies and 170 carry information used to perform Zero Touch Provisioning (ZTP) of 171 levels. 173 Shortest-Path First (SPF): 175 A well-known graph algorithm attributed to Dijkstra that establishes 176 a tree of shortest paths from a source to destinations on the graph. 177 SPF acronym is used due to its familiarity as general term for the 178 node reachability calculations that RIFT can employ to ultimately 179 calculate routes of which Dijkstra algorithm is a possible one. 181 North SPF (N-SPF): 183 A reachability calculation that is progressing northbound, as example 184 SPF that is using South Node TIEs only. Normally it progresses a 185 single hop only and installs default routes. 187 South SPF (S-SPF): 189 A reachability calculation that is progressing southbound, as example 190 SPF that is using North Node TIEs only. 192 3. Problem Statement of Routing in Modern IP Fabric Fat Tree Networks 194 Clos [CLOS] topologies (called commonly a fat tree/network in modern 195 IP fabric considerations as homonym to the original definition of the 196 term Fat Tree [FATTREE]) have gained prominence in today's 197 networking, primarily as a result of the paradigm shift towards a 198 centralized data-center based architecture that deliver a majority of 199 computation and storage services. 201 Current routing protocols were geared towards a network with an 202 irregular topology with isotropic properties, and low degree of 203 connectivity. When applied to Fat Tree topologies: 205 * They tend to need extensive configuration or provisioning during 206 bring up and adding or removing Rift nodes from the fabric. 208 * All nodes including spine and leaf nodes learn the entire network 209 topology and routing information, which is in fact, not needed on 210 the leaf nodes during normal operation. 212 * They flood significant amounts of duplicate link state information 213 between spine and leaf nodes during topology updates and 214 convergence events, requiring that additional CPU and link 215 bandwidth be consumed. This may impact the stability and 216 scalability of the fabric, make the fabric less reactive to 217 failures, and prevent the use of cheaper hardware at the lower 218 levels (i.e. spine and leaf nodes). 220 4. Applicability of RIFT to Clos IP Fabrics 222 Further content of this document assumes that the reader is familiar 223 with the terms and concepts used in OSPF [RFC2328] and IS-IS 224 [ISO10589-Second-Edition] link-state protocols. The sections of RIFT 225 [RIFT] outline the requirements of routing in IP fabrics and RIFT 226 protocol concepts. 228 4.1. Overview of RIFT 230 RIFT is a dynamic routing protocol that is tailored for use in Clos, 231 Fat-Tree, and other anisotropic topologies. A core property of RIFT 232 is that its operation is sensitive to the structure of the fabric - 233 it is anisotropic. RIFT acts as a link-state protocol when "pointing 234 north" - advertising southwards routes to northwards peer routers 235 (parents) through flooding and database synchronization- but operates 236 hop-by-hop like a distance-vector protocol when "pointing south" - 237 typically advertising a fabric default route directed towards the Top 238 of Fabric (ToF, aka superspine) to southwards peer routers 239 (children). 241 The fabric default is typically the default route, as described in 242 Section 4.2.3.8 "Southbound Default Route Origination" of RIFT 243 [RIFT]. The ToF nodes may alternatively originate more specific 244 prefixes (P') southbound instead of the default route. In such a 245 scenario, all addresses carried within the RIFT domain must be 246 contained within P', and it is possible for a leaf that acts as 247 gateway to the internet to advertise the default route instead. 249 RIFT floods flat link-state information northbound only so that each 250 level obtains the full topology of levels south of it. That 251 information is never flooded east-west or back south again. So a top 252 tier node has full set of prefixes from the Shortest Path First (SPF) 253 calculation. 255 In the southbound direction, the protocol operates like a "fully 256 summarizing, unidirectional" path-vector protocol or rather a 257 distance-vector with implicit split horizon. Routing information, 258 normally just the default route, propagates one hop south and is "re- 259 advertised" by nodes at next lower level. 261 +---------------+ +----------------+ 262 | ToF | | ToF | LEVEL 2 263 + ++------+--+--+-+ ++-+--+----+-----+ 264 | | | | | | | | | ^ 265 + | | | +-------------------------+ | 266 Distance | +-------------------+ | | | | | 267 Vector | | | | | | | | + 268 South | | | | +--------+ | | | Link-State 269 + | | | | | | | | Flooding 270 | | | +----------------+ | | | North 271 v | | | | | | | | + 272 ++---+-+ +------+ +-+----+ ++----++ | 273 |SPINE | |SPINE | | SPINE| | SPINE| | LEVEL 1 274 + ++----++ ++---+-+ +-+--+-+ ++----++ | 275 + | | | | | | | | | ^ N 276 Distance | +-------+ | | +--------+ | | | E 277 Vector | | | | | | | | | +------> 278 South | +-------+ | | | +------+ | | | | 279 + | | | | | | | | | + 280 v ++--++ +-+-++ ++--++ ++--++ + 281 |LEAF| |LEAF| |LEAF| |LEAF| LEVEL 0 282 +----+ +----+ +----+ +----+ 284 Figure 1: RIFT overview 286 A spine node has only information necessary for its level, which is 287 all destinations south of the node based on SPF calculation, default 288 route, and potential disaggregated routes. 290 RIFT combines the advantage of both link-state and distance-vector: 292 * Fastest possible convergence 294 * Automatic detection of topology 296 * Minimal routes/info on Top-of-Rack (ToR) switches, aka leaf nodes 298 * High degree of ECMP 300 * Fast de-commissioning of nodes 302 * Maximum propagation speed with flexible prefixes in an update 304 So there are two types of link-state database which are "north 305 representation" North Topology Information Elements (N-TIEs) and 306 "south representation" South Topology Information Elements (S-TIEs). 307 The N-TIEs contain a link-state topology description of lower levels 308 and S-TIEs carry simply default and disaggregated routes for the 309 lower levels. 311 RIFT also eliminates major disadvantages of link-state and distance- 312 vector with: 314 * Reduced and balanced flooding 316 * Level constrained automatic neighbor detection 318 To achieve this, RIFT builds on the art of IGPs, not only OSPF and 319 IS-IS but also MANET and IoT, to provide unique features: 321 * Automatic (positive or negative) route disaggregation of 322 northwards routes upon fallen leaves 324 * Recursive operation in the case of negative route disaggregation 326 * Anisotropic routing that extends a principle seen in RPL [RFC6550] 327 to wide superspines 329 * Optimal flooding reduction that derives from the concept of a 330 "multipoint relay" (MPR) found in OLSR [RFC3626] and balances the 331 flooding load over northbound links and nodes. 333 Additional advantages that are unique to RIFT are listed below, the 334 details of which can be found in RIFT [RIFT]. 336 * True ZTP(Zero Touch Provisioning) 338 * Minimal blast radius on failures 340 * Can utilize all paths through fabric without looping 342 * Simple leaf implementation that can scale down to servers 344 * Key-Value store 346 * Horizontal links used for protection only 348 4.2. Applicable Topologies 350 Albeit RIFT is specified primarily for "proper" Clos or Fat Tree 351 topologies, the protocol natively supports Points of Delivery (PoD) 352 concepts, which, strictly speaking, are not found in the original 353 Clos concept. 355 Further, the specification explains and supports operations of multi- 356 plane Clos variants where the protocol recommends the use of inter- 357 plane rings at the Top-of-Fabric level to allow the reconciliation of 358 topology view of different planes to make the negative disaggregation 359 viable in case of failures within a plane. These observations hold 360 not only in case of RIFT but also in the generic case of dynamic 361 routing on Clos variants with multiple planes and failures in bi- 362 sectional bandwidth, especially on the leafs. 364 4.2.1. Horizontal Links 366 RIFT is not limited to pure Clos divided into PoD and multi-planes 367 but supports horizontal (East-West) links below the top of fabric 368 level. Those links are used only for last resort northbound 369 forwarding when a spine loses all its northbound links or cannot 370 compute a default route through them. 372 A full-mesh connectivity between nodes on the same level can be 373 employed and that allows N-SPF to provide for any node losing all its 374 northbound adjacencies (as long as any of the other nodes in the 375 level are northbound connected) to still participate in northbound 376 forwarding. 378 Note that a "ring" of horizontal links at any level below ToF does 379 not provide a "ring-based protection" scheme since the SPF 380 computation would have to deal necessarily with breaking of "loops" 381 in Dijkstra sense--an application for which RIFT is not intended. 383 4.2.2. Vertical Shortcuts 385 Through relaxations of the specified adjacency forming rules, RIFT 386 implementations can be extended to support vertical "shortcuts". The 387 RIFT specification itself does not provide the exact details since 388 the resulting solution suffers from either much larger blast radius 389 with increased flooding volumes or in case of maximum aggregation 390 routing, bow-tie problems. 392 4.2.3. Generalizing to any Directed Acyclic Graph 394 RIFT is an anisotropic routing protocol, meaning that it has a sense 395 of direction (northbound, southbound, east-west) and that it operates 396 differently depending on the direction. 398 * Northbound, RIFT operates as a link-state protocol, whereby the 399 control packets are reflooded first all the way north and only 400 interpreted later. All the individual fine grained routes are 401 advertised. 403 * Southbound, RIFT operates as a distance-vector protocol, whereby 404 the control packets are flooded only one-hop, interpreted, and the 405 consequence of that computation is what gets flooded one more hop 406 south. In the most common use-cases, a ToF node can reach most of 407 the prefixes in the fabric. If that is the case, the ToF node 408 advertises the fabric default and negatively disaggregates the 409 prefixes that it cannot reach. On the other hand, a ToF node that 410 can reach only a small subset of the prefixes in the fabric will 411 preferably advertise those prefixes and refrain from aggregating. 413 In the general case, what gets advertised south are: 415 1. A fabric default that aggregates all the prefixes that are 416 reachable within the fabric, and that could be a default route 417 or a prefix that is dedicated to this particular fabric. 419 2. The loopback addresses of the northbound nodes, e.g., for 420 inband management. 422 3. The disaggregated prefixes for the dynamic exceptions to the 423 fabric default, advertised to route around the black hole that 424 may form. 426 * East-West routing can optionally be used, with specific 427 restrictions. It is used when a sibling has access to the fabric 428 default but this node does not. 430 Since a Directed Acyclic Graph (DAG) provides a sense of north (the 431 direction of the DAG) and of south (the reverse), it can be used to 432 apply RIFT--an edge in the DAG that has only incoming vertices is a 433 ToF node. 435 There are a number of caveats though: 437 * The DAG structure must exist before RIFT starts, so there is a 438 need for a companion protocol to establish the logical DAG 439 structure. 441 * A generic DAG does not have a sense of east and west. The 442 operation specified for east-west links and the southbound 443 reflection between nodes are not applicable. Also ZTP(Zero Touch 444 Provisioning) will derive a sense of depth that will eliminate 445 some links. Variations of ZTP(Zero Touch Provisioning) could be 446 derived to meet specific objectives, e.g., make it so that most 447 routers have at least 2 parents to reach the ToF. 449 * RIFT applies to any Destination-Oriented DAG (DODAG) where there's 450 only one ToF node and the problem of disaggregation does not 451 exist. In that case, RIFT operates very much like RPL [RFC6550], 452 but using Link State for southbound routes (downwards in RPL's 453 terms). For an arbitrary DAG with multiple destinations (ToFs) 454 the way disaggregation happens has to be considered. 456 * Positive disaggregation expects that most of the ToF nodes reach 457 most of the leaves, so disaggregation is the exception as opposed 458 to the rule. When this is no more true, it makes sense to turn 459 off disaggregation and route between the ToF nodes over a ring, a 460 full mesh, transit network, or a form of area zero. There again, 461 this operation is similar to RPL operating as a single DODAG with 462 a virtual root. 464 * In order to aggregate and disaggregate routes, RIFT requires that 465 all the ToF nodes share the full knowledge of the prefixes in the 466 fabric. 468 * This can be achieved with a ring as suggested by "RIFT" [RIFT], by 469 some preconfiguration, or using a synchronization with a common 470 repository where all the active prefixes are registered. 472 4.2.4. Reachability of Internal Nodes in the Fabric 474 RIFT does not require that nodes have reachable addresses in the 475 fabric, though it is clearly desirable for operational purposes. 476 Under normal operating conditions this can be easily achieved by 477 injecting the node's loopback address into North and South Prefix 478 TIEs or other implementation specific mechanisms. 480 Special considerations arise when a node loses all northbound 481 adjacencies, but is not at the top of the fabric. If a spine node 482 loses all northbound links, the spine node doesn't advertise default 483 route. But if the level of the spine node is auto-determined by ZTP, 484 it will "fall down" as despicted in Figure 8. 486 4.3. Use Cases 488 4.3.1. Data Center Topologies 490 4.3.1.1. Data Center Fabrics 492 RIFT is suited for applying in data center (DC) IP fabrics underlay 493 routing, vast majority of which seem to be currently (and for the 494 foreseeable future) Clos architectures. It significantly simplifies 495 operation and deployment of such fabrics as described in Section 5 496 for environments compared to extensive proprietary provisioning and 497 operational solutions. 499 4.3.1.2. Adaptations to Other Proposed Data Center Topologies 500 . +-----+ +-----+ 501 . | | | | 502 .+-+ S0 | | S1 | 503 .| ++---++ ++---++ 504 .| | | | | 505 .| | +------------+ | 506 .| | | +------------+ | 507 .| | | | | 508 .| ++-+--+ +--+-++ 509 .| | | | | 510 .| | A0 | | A1 | 511 .| +-+--++ ++---++ 512 .| | | | | 513 .| | +------------+ | 514 .| | +-----------+ | | 515 .| | | | | 516 .| +-+-+-+ +--+-++ 517 .+-+ | | | 518 . | L0 | | L1 | 519 . +-----+ +-----+ 521 Figure 2: Level Shortcut 523 RIFT is not strictly limited to Clos topologies. The protocol only 524 requires a sense of "compass rose directionality" either achieved 525 through configuration or derivation of levels. So, conceptually, 526 shortcuts between levels could be included. Figure 2 depicts an 527 example of a shortcut between levels. In this example, sub-optimal 528 routing will occur when traffic is sent from L0 to L1 via S0's 529 default route and back down through A0 or A1. In order to avoid 530 that, only default routes from A0 or A1 are used, all leaves would be 531 required to install each others routes. 533 While various technical and operational challenges may require the 534 use of such modifications, discussion of those topics are outside the 535 scope of this document. 537 4.3.2. Metro Fabrics 539 The demand for bandwidth is increasing steadily, driven primarily by 540 environments close to content producers (server farms connection via 541 DC fabrics) but in proximity to content consumers as well. Consumers 542 are often clustered in metro areas with their own network 543 architectures that can benefit from simplified, regular Clos 544 structures and hence from RIFT. 546 4.3.3. Building Cabling 548 Commercial edifices are often cabled in topologies that are either 549 Clos or its isomorphic equivalents. The Clos can grow rather high 550 with many levels. That presents a challenge for traditional routing 551 protocols (except BGP and by now largely phased-out PNNI) which do 552 not support an arbitrary number of levels which RIFT does naturally. 553 Moreover, due to the limited sizes of forwarding tables in network 554 elements of building cabling, the minimum FIB size RIFT maintains 555 under normal conditions is cost-effective in terms of hardware and 556 operational costs. 558 4.3.4. Internal Router Switching Fabrics 560 It is common in high-speed communications switching and routing 561 devices to use fabrics when a crossbar is not feasible due to cost, 562 head-of-line blocking or size trade-offs. Normally such fabrics are 563 not self-healing or rely on 1:/+1 protection schemes but it is 564 conceivable to use RIFT to operate Clos fabrics that can deal 565 effectively with interconnections or subsystem failures in such 566 module. RIFT is not IP specific and hence any link addressing 567 connecting internal device subnets is conceivable. 569 4.3.5. CloudCO 571 The Cloud Central Office (CloudCO) is a new stage of telecom Central 572 Office. It takes the advantage of Software Defined Networking (SDN) 573 and Network Function Virtualization (NFV) in conjunction with general 574 purpose hardware to optimize current networks. The following figure 575 illustrates this architecture at a high level. It describes a single 576 instance or macro-node of cloud CO that provides a number of Value 577 Added Services (VAS), a Broadband Access Abstraction (BAA), and 578 virtualized nerwork services. An Access I/O module faces a Cloud CO 579 access node, and the Customer Premises Equipments (CPEs) behind it. 580 A Network I/O module is facing the core network. The two I/O modules 581 are interconnected by a leaf and spine fabric [TR-384]. 583 +---------------------+ +----------------------+ 584 | Spine | | Spine | 585 | Switch | | Switch | 586 +------+---+------+-+-+ +--+-+-+-+-----+-------+ 587 | | | | | | | | | | | | 588 | | | | | +-------------------------------+ | 589 | | | | | | | | | | | | 590 | | | | +-------------------------+ | | | 591 | | | | | | | | | | | | 592 | | +----------------------+ | | | | | | | | 593 | | | | | | | | | | | | 594 | +---------------------------------+ | | | | | | | 595 | | | | | | | | | | | | 596 | | | +-----------------------------+ | | | | | 597 | | | | | | | | | | | | 598 | | | | | +--------------------+ | | | | 599 | | | | | | | | | | | | 600 +--+ +-+---+--+ +-+---+--+ +--+----+--+ +-+--+--+ +--+ 601 |L | | Leaf | | Leaf | | Leaf | | Leaf | |L | 602 |S | | Switch | | Switch | | Switch | | Switch| |S | 603 ++-+ +-+-+-+--+ +-+-+-+--+ +--+-+--+--+ ++-+--+-+ +-++ 604 | | | | | | | | | | | | | | 605 | +-+-+-+--+ +-+-+-+--+ +--+-+--+--+ ++-+--+-+ | 606 | |Compute | |Compute | | Compute | |Compute| | 607 | |Node | |Node | | Node | |Node | | 608 | +--------+ +--------+ +----------+ +-------+ | 609 | || VAS5 || || vDHCP|| || vRouter|| ||VAS1 || | 610 | |--------| |--------| |----------| |-------| | 611 | |--------| |--------| |----------| |-------| | 612 | || VAS6 || || VAS3 || || v802.1x|| ||VAS2 || | 613 | |--------| |--------| |----------| |-------| | 614 | |--------| |--------| |----------| |-------| | 615 | || VAS7 || || VAS4 || || vIGMP || ||BAA || | 616 | |--------| |--------| |----------| |-------| | 617 | +--------+ +--------+ +----------+ +-------+ | 618 | | 619 ++-----------+ +---------++ 620 |Network I/O | |Access I/O| 621 +------------+ +----------+ 623 Figure 3: An example of CloudCO architecture 625 The Spine-Leaf architecture deployed inside CloudCO meets the network 626 requirements of adaptable, agile, scalable and dynamic. 628 5. Operational Considerations 630 RIFT presents the opportunity for organizations building and 631 operating IP fabrics to simplify their operation and deployments 632 while achieving many desirable properties of a dynamic routing on 633 such a substrate: 635 * RIFT only floods routing information to the devices that 636 absolutely need it. RIFT design follows minimum blast radius and 637 minimum necessary epistemological scope philosophy which leads to 638 good scaling properties while delivering maximum reactiveness. 640 * RIFT allows for extensive Zero Touch Provisioning within the 641 protocol. In its most extreme version RIFT does not rely on any 642 specific addressing and for IP fabric can operate using IPv6 ND 643 [RFC4861] only. 645 * RIFT has provisions to detect common IP fabric mis-cabling 646 scenarios. 648 * RIFT negotiates automatically BFD per link. This allows for IP 649 and micro-BFD [RFC7130] to replace Link Aggregation Groups (LAGs) 650 which do hide bandwidth imbalances in case of constituent 651 failures. Further automatic link validation techniques similar to 652 [RFC5357] could be supported as well. 654 * RIFT inherently solves many difficult problems associated with the 655 use of traditional routing topologies with dense meshes and high 656 degrees of ECMP by including automatic bandwidth balancing, flood 657 reduction and automatic disaggregation on failures while providing 658 maximum aggregation of prefixes in default scenarios. 660 * RIFT reduces FIB size towards the bottom of the IP fabric where 661 most nodes reside and allows with that for cheaper hardware on the 662 edges and introduction of modern IP fabric architectures that 663 encompass e.g. server multi-homing. 665 * RIFT provides valley-free routing and with that is loop free. 666 This allows the use of any such valley-free path in bi-sectional 667 fabric bandwidth between two destination irrespective of their 668 metrics which can be used to balance load on the fabric in 669 different ways. 671 * RIFT includes a key-value distribution mechanism which allows for 672 many future applications such as automatic provisioning of basic 673 overlay services or automatic key roll-overs over whole fabrics. 675 * RIFT is designed for minimum delay in case of prefix mobility on 676 the fabric. In conjunction with [RFC8505], RIFT can differentiate 677 anycast advertisements from mobility events and retain only the 678 most recent advertisement in the latter case. 680 * Many further operational and design points collected over many 681 years of routing protocol deployments have been incorporated in 682 RIFT such as fast flooding rates, protection of information 683 lifetimes and operationally easily recognizable remote ends of 684 links and node names. 686 5.1. South Reflection 688 South reflection is a mechanism that South Node TIEs are "reflected" 689 back up north to allow nodes in same level without east-west links to 690 "see" each other. 692 For example, Spine111\Spine112\Spine121\Spine122 reflects Node S-TIEs 693 from ToF21 to ToF22 separately. Respectively, 694 Spine111\Spine112\Spine121\Spine122 reflects Node S-TIEs from ToF22 695 to ToF21 separately. So ToF22 and ToF21 see each other's node 696 information as level 2 nodes. 698 In an equivalent fashion, as the result of the south reflection 699 between Spine121-Leaf121-Spine122 and Spine121-Leaf122-Spine122, 700 Spine121 and Spine 122 knows each other at level 1. 702 5.2. Suboptimal Routing on Link Failures 703 +--------+ +--------+ 704 | ToF21 | | ToF22 | LEVEL 2 705 ++--+-+-++ ++-+--+-++ 706 | | | | | | | + 707 | | | | | | | linkTS8 708 +------------+ | +-+linkTS3+-+ | | | +-------------+ 709 | | | | | | + | 710 | +---------------------------+ | linkTS7 | 711 | | | | + + + | 712 | | | +-------+linkTS4+------------+ | 713 | | | + + | | | 714 | | | +-------------+--+ | | 715 | | | | | linkTS6 | | 716 +-+----+-+ +-+----+-+ ++--------+ +-+----+-+ 717 |Spine111| |Spine112| |Spine121 | |Spine122| LEVEL 1 718 +-+---+--+ +-+----+-+ +-+---+---+ +-+----+-+ 719 | | | | | | | | 720 | +-------------+ | + ++XX+linkSL6+---+ + 721 | | | | linkSL5 | | linkSL8 722 | +-----------+ | | + +---+linkSL7+-+ | + 723 | | | | | | | | 724 +-+---+-+ +--+--+-+ +-+---+-+ +--+--+-+ 725 |Leaf111| |Leaf112| |Leaf121| |Leaf122| LEVEL 0 726 +-+-----+ +-+-----+ +-----+-+ +-+-----+ 727 + + + + 728 Prefix111 Prefix112 Prefix121 Prefix122 730 Figure 4: Suboptimal routing upon link failure use case 732 As shown in Figure 4, as the result of the south reflection between 733 Spine121-Leaf121-Spine122 and Spine121-Leaf122-Spine122, Spine121 and 734 Spine 122 knows each other at level 1. 736 Without disaggregation mechanism, when linkSL6 fails, the packet from 737 leaf121 to prefix122 will probably go up through linkSL5 to linkTS3 738 then go down through linkTS4 to linkSL8 to Leaf122 or go up through 739 linkSL5 to linkTS6 then go down through linkTS8 and linkSL8 to 740 Leaf122 based on pure default route. It's the case of suboptimal 741 routing or bow-tieing. 743 With disaggregation mechanism, when linkSL6 fails, Spine122 will 744 detect the failure according to the reflected node S-TIE from 745 Spine121. Based on the disaggregation algorithm provided by RIFT, 746 Spine122 will explicitly advertise prefix122 in Disaggregated Prefix 747 S-TIE PrefixesElement(prefix122, cost 1). The packet from leaf121 to 748 prefix122 will only be sent to linkSL7 following a longest-prefix 749 match to prefix 122 directly then go down through linkSL8 to Leaf122 750 . 752 5.3. Black-Holing on Link Failures 754 +--------+ +--------+ 755 | ToF 21 | | ToF 22 | LEVEL 2 756 ++-+--+-++ ++-+--+-++ 757 | | | | | | | + 758 | | | | | | | linkTS8 759 +--------------+ | +-+linkTS3+X+ | | | +--------------+ 760 linkTS1 | | | | | + | 761 + +-----------------------------+ | linkTS7 | 762 | | + | + + + | 763 | | linkTS2 +-------+linkTS4+X+----------+ | 764 | + + + + | | | 765 | linkTS5 +-+ +------------+--+ | | 766 | + | | | linkTS6 | | 767 +-+----+-+ +-+----+-+ ++-------+ +-+-----++ 768 |Spine111| |Spine112| |Spine121| |Spine122| LEVEL 1 769 +-+---+--+ ++----+--+ +-+---+--+ +-+----+-+ 770 | | | | | | | | 771 + +---------------+ | + +---+linkSL6+---+ + 772 linkSL1 | | | linkSL5 | | linkSL8 773 + +--+linkSL3+--+ | | + +---+linkSL7+-+ | + 774 | | | | | | | | 775 +-+---+-+ +--+--+-+ +-+---+-+ +--+--+-+ 776 |Leaf111| |Leaf112| |Leaf121| |Leaf122| LEVEL 0 777 +-+-----+ +-+-----+ +-----+-+ +-----+-+ 778 + + + + 779 Prefix111 Prefix112 Prefix121 Prefix122 781 Figure 5: Black-holing upon link failure use case 783 This scenario illustrates a case when double link failure occurs and 784 with that black-holing can happen. 786 Without disaggregation mechanism, when linkTS3 and linkTS4 both fail, 787 the packet from leaf111 to prefix122 would suffer 50% black-holing 788 based on pure default route. The packet supposed to go up through 789 linkSL1 to linkTS1 then go down through linkTS3 or linkTS4 will be 790 dropped. The packet supposed to go up through linkSL3 to linkTS2 791 then go down through linkTS3 or linkTS4 will be dropped as well. 792 It's the case of black-holing. 794 With disaggregation mechanism, when linkTS3 and linkTS4 both fail, 795 ToF22 will detect the failure according to the reflected node S-TIE 796 of ToF21 from Spine111\Spine112. Based on the disaggregation 797 algorithm provided by RIFT, ToF22 will explicitly originate an S-TIE 798 with prefix 121 and prefix 122, that is flooded to spines 111, 112, 799 121 and 122. 801 The packet from leaf111 to prefix122 will not be routed to linkTS1 or 802 linkTS2. The packet from leaf111 to prefix122 will only be routed to 803 linkTS5 or linkTS7 following a longest-prefix match to prefix122. 805 5.4. Zero Touch Provisioning (ZTP) 807 RIFT is designed to require a very minimal configuration to simplify 808 its operation and avoid human errors; based on that minimal 809 information, Zero Touch Provisioning (ZTP) autoconfigures the key 810 operational parameters of all the RIFT nodes, including the SystemID 811 of the node that must be unique in the RIFT network and the level of 812 the node in the Fat Tree, which determines which peers are northwards 813 "parents" and which are southwards "children". 815 ZTP is always on, but its decisions can be overridden when a network 816 administrator prefers to impose its own configuration. In that case, 817 it is the responsibility of the administrator to ensure that the 818 configured parameters are correct, in other words that the SystemID 819 of each node is unique, and that the administratively set levels 820 truly reflect the relative position of the nodes in the fabric. It 821 is recommended to let ZTP configure the network, and when not, it is 822 recommended to configure the level of all the nodes to avoid an 823 undesirable interaction between ZTP and the manual configuration. 825 ZTP requires that the administrator points out the Top-of-Fabric 826 (ToF) nodes to set the baseline from which the fabric topology is 827 derived. The Top-of-Fabric nodes are configured with TOP_OF_FABRIC 828 flag which are initial 'seeds' needed for other ZTP nodes to derive 829 their level in the topology. ZTP computes the level of each node 830 based on the Highest Available Level (HAL) of the potential parent(s) 831 nearest that baseline, which represents the superspine. In a 832 fashion, RIFT can be seen as a distance-vector protocol that computes 833 a set of feasible successors towards the superspine and auto- 834 configures the rest of the topology. 836 The autoconfiguration mechanism computes a global maximum of levels 837 by diffusion. The derivation of the level of each node happens then 838 based on Link Information Elements (LIEs) received from its neighbors 839 whereas each node (with possibly exceptions of configured leaves) 840 tries to attach at the highest possible point in the fabric. This 841 guarantees that even if the diffusion front reaches a node from 842 "below" faster than from "above", it will greedily abandon already 843 negotiated level derived from nodes topologically below it and 844 properly peer with nodes above. 846 The achieved equilibrium can be disturbed massively by all nodes with 847 highest level either leaving or entering the domain (with some finer 848 distinctions not explained further). It is therefore recommended 849 that each node is multi-homed towards nodes with respective HAL 850 offerings. Fortunately, this is the natural state of things for the 851 topology variants considered in RIFT. 853 A RIFT node may also be configured to confine it to the leaf role 854 with the LEAF_ONLY flag. A leaf node can also be configured to 855 support leaf-2-leaf procedures with the LEAF_2_LEAF flag. In either 856 case the node cannot be TOP_OF_FABRIC and its level cannot be 857 configured. RIFT will fully determine the node's level after it is 858 attached to the topology and ensure that the node is at the "bottom 859 of the hierarchy" (southernmost). 861 5.5. Mis-cabling Examples 863 +----------------+ +-----------------+ 864 | ToF21 | +------+ ToF22 | LEVEL 2 865 +-------+----+---+ | +----+---+--------+ 866 | | | | | | | | | 867 | | | +----------------------------+ | 868 | +---------------------------+ | | | | 869 | | | | | | | | | 870 | | | | +-----------------------+ | | 871 | | +------------------------+ | | | 872 | | | | | | | | | 873 +-+---+--+ +-+---+--+ | +--+---+-+ +--+---+-+ 874 |Spine111| |Spine112| | |Spine121| |Spine122| LEVEL 1 875 +-+---+--+ ++----+--+ | +--+---+-+ +-+----+-+ 876 | | | | | | | | | 877 | +---------+ | link-M | +---------+ | 878 | | | | | | | | | 879 | +-------+ | | | | +-------+ | | 880 | | | | | | | | | 881 +-+---+-+ +--+--+-+ | +-+---+-+ +--+--+-+ 882 |Leaf111| |Leaf112+-----+ |Leaf121| |Leaf122| LEVEL 0 883 +-------+ +-------+ +-------+ +-------+ 885 Figure 6: A single plane mis-cabling example 887 Figure 6 shows a single plane mis-cabling example. It's a perfect 888 Fat Tree fabric except link-M connecting Leaf112 to ToF22. 890 The RIFT control protocol can discover the physical links 891 automatically and be able to detect cabling that violates Fat Tree 892 topology constraints. It reacts accordingly to such mis-cabling 893 attempts, at a minimum preventing adjacencies between nodes from 894 being formed and traffic from being forwarded on those mis-cabled 895 links. Leaf112 will in such scenario use link-M to derive its level 896 (unless it is leaf) and can report links to Spine111 and Spine112 as 897 mis-cabled unless the implementations allows horizontal links. 899 Figure 7 shows a multiple plane mis-cabling example. Since Leaf112 900 and Spine121 belong to two different PoDs, the adjacency between 901 Leaf112 and Spine121 can not be formed. link-W would be detected and 902 prevented. 904 +-------+ +-------+ +-------+ +-------+ 905 |ToF A1| |ToF A2| |ToF B1| |ToF B2| LEVEL 2 906 +-------+ +-------+ +-------+ +-------+ 907 | | | | | | | | 908 | | | +-----------------+ | | | 909 | +--------------------------+ | | | | 910 | +------+ | | | +------+ | 911 | | +-----------------+ | | | | | 912 | | | +--------------------------+ | | 913 | A | | B | | A | | B | 914 +-----+--+ +-+---+--+ +--+---+-+ +--+-----+ 915 |Spine111| |Spine112| +---+Spine121| |Spine122| LEVEL 1 916 +-+---+--+ ++----+--+ | +--+---+-+ +-+----+-+ 917 | | | | | | | | | 918 | +---------+ | | | +---------+ | 919 | | | | link-W | | | | 920 | +-------+ | | | | +-------+ | | 921 | | | | | | | | | 922 +-+---+-+ +--+--+-+ | +-+---+-+ +--+--+-+ 923 |Leaf111| |Leaf112+------+ |Leaf121| |Leaf122| LEVEL 0 924 +-------+ +-------+ +-------+ +-------+ 925 +--------PoD#1----------+ +---------PoD#2---------+ 927 Figure 7: A multiple plane mis-cabling example 929 RIFT provides an optional level determination procedure in its Zero 930 Touch Provisioning mode. Nodes in the fabric without their level 931 configured determine it automatically. This can have possibly 932 counter-intuitive consequences however. One extreme failure scenario 933 is depicted in Figure 8 and it shows that if all northbound links of 934 spine11 fail at the same time, spine11 negotiates a lower level than 935 Leaf11 and Leaf12. 937 To prevent such scenario where leafs are expected to act as switches, 938 LEAF_ONLY flag can be set for Leaf111 and Leaf112. Since level -1 is 939 invalid, Spine11 would not derive a valid level from the topology in 940 Figure 8. It will be isolated from the whole fabric and it would be 941 up to the leafs to declare the links towards such spine as mis- 942 cabled. 944 +-------+ +-------+ +-------+ +-------+ 945 |ToF A1| |ToF A2| |ToF A1| |ToF A2| 946 +-------+ +-------+ +-------+ +-------+ 947 | | | | | | 948 | +-------+ | | | 949 + + | | ====> | | 950 X X +------+ | +------+ | 951 + + | | | | 952 +----+--+ +-+-----+ +-+-----+ 953 |Spine11| |Spine12| |Spine12| 954 +-+---+-+ ++----+-+ ++----+-+ 955 | | | | | | 956 | +---------+ | | | 957 | +-------+ | | +-------+ | 958 | | | | | | 959 +-+---+-+ +--+--+-+ +-----+-+ +-----+-+ 960 |Leaf111| |Leaf112| |Leaf111| |Leaf112| 961 +-------+ +-------+ +-+-----+ +-+-----+ 962 | | 963 | +--------+ 964 | | 965 +-+---+-+ 966 |Spine11| 967 +-------+ 969 Figure 8: Fallen spine 971 5.6. Positive vs. Negative Disaggregation 973 Disaggregation is the procedure whereby [RIFT] advertises a more 974 specific route southwards as an exception to the aggregated fabric- 975 default north. Disaggregation is useful when a prefix within the 976 aggregation is reachable via some of the parents but not the others 977 at the same level of the fabric. It is mandatory when the level is 978 the ToF since a ToF node that cannot reach a prefix becomes a black 979 hole for that prefix. The hard problem is to know which prefixes are 980 reachable by whom. 982 In the general case, [RIFT] solves that problem by interconnecting 983 the ToF nodes. So the ToF nodes can exchange the full list of 984 prefixes that exist in the fabric and figure out when a ToF node 985 lacks reachability to some prefixes. This requires additional ports 986 at the ToF, typically 2 ports per ToF node to form a ToF-spanning 987 ring. [RIFT] also defines the southbound reflection procedure that 988 enables a parent to explore the direct connectivity of its peers, 989 meaning their own parents and children; based on the advertisements 990 received from the shared parents and children, it may enable the 991 parent to infer the prefixes its peers can reach. 993 When a parent lacks reachability to a prefix, it may disaggregate the 994 prefix negatively, i.e., advertise that this parent can be used to 995 reach any prefix in the aggregation except that one. The Negative 996 Disaggregation signaling is simple and functions transitively from 997 ToF to top-of-pod (ToP) and then from ToP to Leaf. But it is hard 998 for a parent to figure which prefix it needs to disaggregate, because 999 it does not know what it does not know; it results that the use of a 1000 spanning ring at the ToF is required to operate the Negative 1001 Disaggregation. Also, though it is only an implementation problem, 1002 the programmation of the FIB is complex compared to normal routes, 1003 and may incur recursions. 1005 The more classical alternative is, for the parents that can reach a 1006 prefix that peers at the same level cannot, to advertise a more 1007 specific route to that prefix. This leverages the normal longest 1008 prefix match in the FIB, and does not require a special 1009 implementation. But as opposed to the Negative Disaggregation, the 1010 Positive Disaggregation is difficult and inefficient to operate 1011 transitively. 1013 Transitivity is not needed to a grandchild if all its parents 1014 received the Positive Disaggregation, meaning that they shall all 1015 avoid the black hole; when that is the case, they collectively build 1016 a ceiling that protects the grandchild. But until then, a parent 1017 that received a Positive Disaggregation may believe that some peers 1018 are lacking the reachability and readvertise too early, or defer and 1019 maintain a black hole situation longer than necessary. 1021 In a non-partitioned fabric, all the ToF nodes see one another 1022 through the reflection and can figure if one is missing a child. In 1023 that case it is possible to compute the prefixes that the peer cannot 1024 reach and disaggregate positively without a ToF-spanning ring. The 1025 ToF nodes can also ascertain that the ToP nodes are connected each to 1026 at least a ToF node that can still reach the prefix, meaning that the 1027 transitive operation is not required. 1029 The bottom line is that in a fabric that is partitioned (e.g., using 1030 multiple planes) and/or where the ToP nodes are not guaranteed to 1031 always form a ceiling for their children, it is mandatory to use the 1032 Negative Disaggregation. On the other hand, in a highly symmetrical 1033 and fully connected fabric, (e.g., a canonical Clos Network), the 1034 Positive Disaggregation methods allows to save the complexity and 1035 cost associated to the ToF-spanning ring. 1037 Note that in the case of Positive Disaggregation, the first ToF 1038 node(s) that announces a more-specific route attracts all the traffic 1039 for that route and may suffer from a transient incast. A ToP node 1040 that defers injecting the longer prefix in the FIB, in order to 1041 receive more advertisements and spread the packets better, also keeps 1042 on sending a portion of the traffic to the black hole in the 1043 meantime. In the case of Negative Disaggregation, the last ToF 1044 node(s) that injects the route may also incur an incast issue; this 1045 problem would occur if a prefix that becomes totally unreachable is 1046 disaggregated. 1048 5.7. Mobile Edge and Anycast 1050 When a physical or a virtual node changes its point of attachement in 1051 the fabric from a previous-leaf to a next-leaf, new routes must be 1052 installed that supersede the old ones. Since the flooding flows 1053 northwards, the nodes (if any) between the previous-leaf and the 1054 common parent are not immediately aware that the path via previous- 1055 leaf is obsolete, and a stale route may exist for a while. The 1056 common parent needs to select the freshest route advertisement in 1057 order to install the correct route via the next-leaf. This requires 1058 that the fabric determines the sequence of the movements of the 1059 mobile node. 1061 On the one hand, a classical sequence counter provides a total order 1062 for a while but it will eventually wrap. On the other hand, a 1063 timestamp provides a permanent order but it may miss a movement that 1064 happens too quickly vs. the granularity of the timing information. 1065 It is not envisioned that an average fabric supports Precision Time 1066 Protocol [IEEEstd1588] in the short term, nor that the precision 1067 available with the Network Time Protocol [RFC5905] (in the order of 1068 100 to 200ms) may not be necessarily enough to cover, e.g., the fast 1069 mobility of a Virtual Machine. 1071 Section 4.3.3. "Mobility" of [RIFT] specifies an hybrid method that 1072 combines a sequence counter from the mobile node and a timestamp from 1073 the network taken at the leaf when the route is injected. If the 1074 timestamps of the concurrent advertisements are comparable (i.e., 1075 more distant than the precision of the timing protocol), then the 1076 timestamp alone is used to determine the relative freshness of the 1077 routes. Otherwise, the sequence counter from the mobile node, if 1078 available, is used. One caveat is that the sequence counter must not 1079 wrap within the precision of the timing protocol. Another is that 1080 the mobile node may not even provide a sequence counter, in which 1081 case the mobility itself must be slower than the precision of the 1082 timing. 1084 Mobility must not be confused with anycast. In both cases, a same 1085 address is injected in RIFT at different leaves. In the case of 1086 mobility, only the freshest route must be conserved, since mobile 1087 node changed its point of attachment for a leaf to the next. In the 1088 case of anycast, the node may be either multihomed (attached to 1089 multiple leaves in parallel) or reachable beyond the fabric via 1090 multiple routes that are redistributed to different leaves; either 1091 way, in the case of anycast, the multiple routes are equally valid 1092 and should be conserved. Without further information from the 1093 redistributed routing protocol, it is impossible to sort out a 1094 movement from a redistribution that happens asynchronously on 1095 different leaves. [RIFT] expects that anycast addresses are 1096 advertised within the timing precision, which is typically the case 1097 with a low-precision timing and a multihomed node. Beyond that time 1098 interval, RIFT interprets the lag as a mobility and only the freshest 1099 route is retained. 1101 When using IPv6 [RFC8200], RIFT suggests to leverage "Registration 1102 Extensions for IPv6 over Low-Power Wireless Personal Area Network 1103 (6LoWPAN) Neighbor Discovery (ND)" [RFC8505] as the IPv6 ND 1104 interaction between the mobile node and the leaf. This provides not 1105 only a sequence counter but also a lifetime and a security token that 1106 may be used to protect the ownership of an address [RFC8928]. When 1107 using [RFC8505], the parallel registration of an anycast address to 1108 multiple leaves is done with the same sequence counter, whereas the 1109 sequence counter is incremented when the point of attachement 1110 changes. This way, it is possible to differentiate a mobile node 1111 from a multihomed node, even when the mobility happens within the 1112 timing precision. It is also possible for a mobile node to be 1113 multihomed as well, e.g., to change only one of its points of 1114 attachement. 1116 5.8. IPv4 over IPv6 1118 RIFT allows advertising IPv4 prefixes over IPv6 RIFT network. IPv6 1119 Address Family (AF) configures via the usual Neighbor Discovery (ND) 1120 mechanisms and then V4 can use V6 nexthops analogous to [RFC8950]. 1121 It is expected that the whole fabric supports the same type of 1122 forwarding of address families on all the links. RIFT provides an 1123 indication whether a node is v4 forwarding capable and 1124 implementations are possible where different routing tables are 1125 computed per address family as long as the computation remains loop- 1126 free. 1128 +-----+ +-----+ 1129 +---+---+ | ToF | | ToF | 1130 ^ +--+--+ +-----+ 1131 | | | | | 1132 | | +-------------+ | 1133 | | +--------+ | | 1134 + | | | | 1135 V6 +-----+ +-+---+ 1136 Forwarding |Spine| |Spine| 1137 + +--+--+ +-----+ 1138 | | | | | 1139 | | +-------------+ | 1140 | | +--------+ | | 1141 | | | | | 1142 v +-----+ +-+---+ 1143 +---+---+ |Leaf | | Leaf| 1144 +--+--+ +--+--+ 1145 | | 1146 IPv4 prefixes| |IPv4 prefixes 1147 | | 1148 +---+----+ +---+----+ 1149 | V4 | | V4 | 1150 | subnet | | subnet | 1151 +--------+ +--------+ 1153 Figure 9: IPv4 over IPv6 1155 5.9. In-Band Reachability of Nodes 1157 RIFT doesn't precondition that nodes of the fabric have reachable 1158 addresses. But the operational reasons to reach the internal nodes 1159 may exist. Figure 10 shows an example that the network management 1160 station (NMS) attaches to leaf1. 1162 +-------+ +-------+ 1163 | ToF1 | | ToF2 | 1164 ++---- ++ ++-----++ 1165 | | | | 1166 | +----------+ | 1167 | +--------+ | | 1168 | | | | 1169 ++-----++ +--+---++ 1170 |Spine1 | |Spine2 | 1171 ++-----++ ++-----++ 1172 | | | | 1173 | +----------+ | 1174 | +--------+ | | 1175 | | | | 1176 ++-----++ +--+---++ 1177 | Leaf1 | | Leaf2 | 1178 +---+---+ +-------+ 1179 | 1180 |NMS 1182 Figure 10: In-Band reachability of node 1184 If NMS wants to access Leaf2, it simply works. Because loopback 1185 address of Leaf2 is flooded in its Prefix North TIE. 1187 If NMS wants to access Spine2, it simply works too. Because spine 1188 node always advertises its loopback address in the Prefix North TIE. 1189 NMS may reach Spine2 from Leaf1-Spine2 or Leaf1-Spine1-ToF1/ 1190 ToF2-Spine2. 1192 If NMS wants to access ToF2, ToF2's loopback address needs to be 1193 injected into its Prefix South TIE. This TIE must be seen by all 1194 nodes at the level below - the spine nodes in Figure 10 - that must 1195 form a ceiling for all the traffic coming from below (south). 1196 Otherwise, the traffic from NMS may follow the default route to the 1197 wrong ToF Node, e.g., ToF1. 1199 In case of failure between ToF2 and spine nodes, ToF2's loopback 1200 address must be disaggregated recursively all the way to the leaves. 1201 In a partitioned ToF, even with recursive disaggregation a ToF node 1202 is only reachable within its plane. 1204 A possible alternative to recursive disaggregation is to use a ring 1205 that interconnects the ToF nodes to transmit packets between them for 1206 their loopback addresses only. The idea is that this is mostly 1207 control traffic and should not alter the load balancing properties of 1208 the fabric. 1210 5.10. Dual Homing Servers 1212 Each RIFT node may operate in Zero Touch Provisioning (ZTP) mode. It 1213 has no configuration (unless it is a Top-of-Fabric at the top of the 1214 topology or the must operate in the topology as leaf and/or support 1215 leaf-2-leaf procedures) and it will fully configure itself after 1216 being attached to the topology. 1218 +---+ +---+ +---+ 1219 |ToF| |ToF| |ToF| ToF 1220 +---+ +---+ +---+ 1221 | | | | | | 1222 | +----------------+ | | 1223 | +----------------+ | 1224 | | | | | | 1225 +----------+--+ +--+----------+ 1226 | ToR1 | | ToR2 | Spine 1227 +--+------+---+ +--+-------+--+ 1228 +---+ | | | | | | +---+ 1229 | +-----------------+ | | | 1230 | | | +-------------+ | | 1231 | | | | | +-----------------+ | 1232 | | | | +--------------+ | | | 1233 | | | | | | | | 1234 +---+ +---+ +---+ +---+ 1235 | | | | | | | | 1236 +---+ +---+ ............. +---+ +---+ 1237 SV(1) SV(2) SV(n-1) SV(n) Leaf 1239 Figure 11: Dual-homing servers 1241 Sometimes, people may prefer to disaggregate from ToR to servers from 1242 start on, i.e. the servers have couple tens of routes in FIB from 1243 start on beside default routes to avoid breakages at rack level. 1244 Full disaggregation of the fabric could be achieved by configuration 1245 supported by RIFT. 1247 5.11. Fabric With A Controller 1249 There are many different ways to deploy the controller. One 1250 possibility is attaching a controller to the RIFT domain from ToF and 1251 another possibility is attaching a controller from the leaf. 1253 +------------+ 1254 | Controller | 1255 ++----------++ 1256 | | 1257 | | 1258 +----++ ++----+ 1259 ------- | ToF | | ToF | 1260 | +--+--+ +-----+ 1261 | | | | | 1262 | | +-------------+ | 1263 | | +--------+ | | 1264 | | | | | 1265 +-----+ +-+---+ 1266 RIFT domain |Spine| |Spine| 1267 +--+--+ +-----+ 1268 | | | | | 1269 | | +-------------+ | 1270 | | +--------+ | | 1271 | | | | | 1272 | +-----+ +-+---+ 1273 ------- |Leaf | | Leaf| 1274 +-----+ +-----+ 1276 Figure 12: Fabric with a controller 1278 5.11.1. Controller Attached to ToFs 1280 If a controller is attaching to the RIFT domain from ToF, it usually 1281 uses dual-homing connections. The loopback prefix of the controller 1282 should be advertised down by the ToF and spine to leaves. If the 1283 controller loses link to ToF, make sure the ToF withdraw the prefix 1284 of the controller. 1286 5.11.2. Controller Attached to Leaf 1288 If the controller is attaching from a leaf to the fabric, no special 1289 provisions are needed. 1291 5.12. Internet Connectivity Within Underlay 1293 If global addressing is running without overlay, an external default 1294 route needs to be advertised through RIFT fabric to achieve internet 1295 connectivity. For the purpose of forwarding of the entire RIFT 1296 fabric, an internal fabric prefix needs to be advertised in the South 1297 Prefix TIE by ToF and spine nodes. 1299 5.12.1. Internet Default on the Leaf 1301 In case that the internet gateway is a leaf, the leaf node as the 1302 internet gateway needs to advertise a default route in its Prefix 1303 North TIE. 1305 5.12.2. Internet Default on the ToFs 1307 In case that the internet gateway is a ToF, the ToF and spine nodes 1308 need to advertise a default route in the Prefix South TIE. 1310 5.13. Subnet Mismatch and Address Families 1312 +--------+ +--------+ 1313 | | LIE LIE | | 1314 | A | +----> <----+ | B | 1315 | +---------------------+ | 1316 +--------+ +--------+ 1317 X/24 Y/24 1319 Figure 13: subnet mismatch 1321 LIEs are exchanged over all links running RIFT to perform Link 1322 (Neighbor) Discovery. A node must NOT originate LIEs on an address 1323 family if it does not process received LIEs on that family. LIEs on 1324 same link are considered part of the same negotiation independent on 1325 the address family they arrive on. An implementation must be ready 1326 to accept TIEs on all addresses it used as source of LIE frames. 1328 As shown in the above figure, without further checks adjacency of 1329 node A and B may form, but the forwarding between node A and node B 1330 may fail because subnet X mismatches with subnet Y. 1332 To prevent this a RIFT implementation should check for subnet 1333 mismatch just like e.g. ISIS does. This can lead to scenarios where 1334 an adjacency, despite exchange of LIEs in both address families may 1335 end up having an adjacency in a single AF only. This is a 1336 consideration especially in Section 5.8 scenarios. 1338 5.14. Anycast Considerations 1339 + traffic 1340 | 1341 v 1342 +------+------+ 1343 | ToF | 1344 +---+-----+---+ 1345 | | | | 1346 +------------+ | | +------------+ 1347 | | | | 1348 +---+---+ +-------+ +-------+ +---+---+ 1349 | | | | | | | | 1350 |Spine11| |Spine12| |Spine21| |Spine22| LEVEL 1 1351 +-+---+-+ ++----+-+ +-+---+-+ ++----+-+ 1352 | | | | | | | | 1353 | +---------+ | | +---------+ | 1354 | +-------+ | | | +-------+ | | 1355 | | | | | | | | 1356 +-+---+-+ +--+--+-+ +-+---+-+ +--+--+-+ 1357 | | | | | | | | 1358 |Leaf111| |Leaf112| |Leaf121| |Leaf122| LEVEL 0 1359 +-+-----+ ++------+ +-----+-+ +-----+-+ 1360 + + + ^ + 1361 PrefixA PrefixB PrefixA | PrefixC 1362 | 1363 + traffic 1365 Figure 14: Anycast 1367 If the traffic comes from ToF to Leaf111 or Leaf121 which has anycast 1368 prefix PrefixA, RIFT can deal with this case well. But if the 1369 traffic comes from Leaf122, it arrives Spine21 or Spine22 at level 1. 1370 But Spine21 or Spine22 doesn't know another PrefixA attaching 1371 Leaf111. So it will always get to Leaf121 and never get to Leaf111. 1372 If the intension is that the traffic should been offloaded to 1373 Leaf111, then use policy guided prefixes defined in RIFT [RIFT]. 1375 5.15. IoT Applicability 1377 The design of RIFT inherits from RPL [RFC6550] the anisotropic design 1378 of a default route upwards (northwards); it also inherits the 1379 capability to inject external host routes at the Leaf level using 1380 Wireless ND (WiND) [RFC8505][RFC8928] between a RIFT-agnostic host 1381 and a RIFT router. Both the RPL and the RIFT protocols are meant for 1382 large scale, and WiND enables device mobility at the edge the same 1383 way in both cases. 1385 The main difference between RIFT and RPL is that with RPL, there's a 1386 single Root, whereas RIFT has many ToF nodes. This adds huge 1387 capabilities for leaf-2-leaf ECMP paths, but additional complexity 1388 with the need to disaggregate. Also RIFT uses Link State flooding 1389 northwards, and is not designed for low-power operation. 1391 Still nothing prevents that the IP devices connected at the Leaf are 1392 IoT (Internet of Things) devices, which typically expose their 1393 address using WiND - which is an upgrade from 6LoWPAN ND [RFC6775]. 1395 A network that serves high speed/ high power IoT devices should 1396 typically provide deterministic capabilities for applications such as 1397 high speed control loops or movement detection. The Fat Tree is 1398 highly reliable, and in normal condition provides an equivalent 1399 multipath operation; but the ECMP doesn't provide hard guarantees for 1400 either delivery or latency. As long as the fabric is non-blocking 1401 the result is the same; but there can be load unbalances resulting in 1402 incast and possibly congestion loss that will prevent the delivery 1403 within bounded latency. 1405 This could be alleviated with Packet Replication, Elimination and 1406 Reordering (PREOF) [RFC8655] leaf-2-leaf but PREOF is hard to provide 1407 at the scale of all flows, and the replication may increase the 1408 probability of the overload that it attempts to solve. 1410 Note that the load balancing is not RIFT's problem, but it is key to 1411 serve IoT adequately. 1413 5.16. Key Management 1415 As outlined in Section "Security Considerations" of [RIFT], either a 1416 private shared key or a public/private key pair is used to 1417 authenticate the adjacency. Both the key distribution and key 1418 synchronization methods are out of scope for this document. Both 1419 nodes in the adjacency must share the same keys, key type, and 1420 algorithm for a given key ID. Mismatched keys will not inter-operate 1421 as their security envelopes will be unverifiable. 1423 Key roll-over while the adjacency is active may be supported. The 1424 specific mechanism is well documented in [RFC6518]. 1426 6. Security Considerations 1428 This document presents applicability of RIFT. As such, it does not 1429 introduce any security considerations. However, there are a number 1430 of security concerns at [RIFT]. 1432 7. IANA Considerations 1434 This document has no IANA actions. 1436 8. Acknowledgments 1438 The authors would like to thank Jaroslaw Kowalczyk and Jeffrey Zhang 1439 for providing invaluable concepts and content for this document. 1441 9. Contributors 1443 The following people (listed in alphabetical order) contributed 1444 significantly to the content of this document and should be 1445 considered co-authors: 1447 Tom Verhaeg 1449 Juniper Networks 1451 Email: tverhaeg@juniper.net 1453 10. Normative References 1455 [ISO10589-Second-Edition] 1456 International Organization for Standardization, 1457 "Intermediate system to Intermediate system intra-domain 1458 routeing information exchange protocol for use in 1459 conjunction with the protocol for providing the 1460 connectionless-mode Network Service (ISO 8473)", November 1461 2002. 1463 [TR-384] Broadband Forum Technical Report, "TR-384 Cloud Central 1464 Office Reference Architectural Framework", January 2018. 1466 [RFC2328] Moy, J., "OSPF Version 2", STD 54, RFC 2328, 1467 DOI 10.17487/RFC2328, April 1998, 1468 . 1470 [RFC4861] Narten, T., Nordmark, E., Simpson, W., and H. Soliman, 1471 "Neighbor Discovery for IP version 6 (IPv6)", RFC 4861, 1472 DOI 10.17487/RFC4861, September 2007, 1473 . 1475 [RFC5357] Hedayat, K., Krzanowski, R., Morton, A., Yum, K., and J. 1476 Babiarz, "A Two-Way Active Measurement Protocol (TWAMP)", 1477 RFC 5357, DOI 10.17487/RFC5357, October 2008, 1478 . 1480 [RFC7130] Bhatia, M., Ed., Chen, M., Ed., Boutros, S., Ed., 1481 Binderberger, M., Ed., and J. Haas, Ed., "Bidirectional 1482 Forwarding Detection (BFD) on Link Aggregation Group (LAG) 1483 Interfaces", RFC 7130, DOI 10.17487/RFC7130, February 1484 2014, . 1486 [RFC8950] Litkowski, S., Agrawal, S., Ananthamurthy, K., and K. 1487 Patel, "Advertising IPv4 Network Layer Reachability 1488 Information (NLRI) with an IPv6 Next Hop", RFC 8950, 1489 DOI 10.17487/RFC8950, November 2020, 1490 . 1492 [RFC6518] Lebovitz, G. and M. Bhatia, "Keying and Authentication for 1493 Routing Protocols (KARP) Design Guidelines", RFC 6518, 1494 DOI 10.17487/RFC6518, February 2012, 1495 . 1497 [RFC6550] Winter, T., Ed., Thubert, P., Ed., Brandt, A., Hui, J., 1498 Kelsey, R., Levis, P., Pister, K., Struik, R., Vasseur, 1499 JP., and R. Alexander, "RPL: IPv6 Routing Protocol for 1500 Low-Power and Lossy Networks", RFC 6550, 1501 DOI 10.17487/RFC6550, March 2012, 1502 . 1504 [RFC6775] Shelby, Z., Ed., Chakrabarti, S., Nordmark, E., and C. 1505 Bormann, "Neighbor Discovery Optimization for IPv6 over 1506 Low-Power Wireless Personal Area Networks (6LoWPANs)", 1507 RFC 6775, DOI 10.17487/RFC6775, November 2012, 1508 . 1510 [RFC8655] Finn, N., Thubert, P., Varga, B., and J. Farkas, 1511 "Deterministic Networking Architecture", RFC 8655, 1512 DOI 10.17487/RFC8655, October 2019, 1513 . 1515 [RIFT] Sharma, A., Thubert, P., Rijsman, B., and D. Afanasiev, 1516 "RIFT: Routing in Fat Trees", Work in Progress, Internet- 1517 Draft, draft-ietf-rift-rift-13, 12 July 2021, 1518 . 1521 11. Informative References 1523 [IEEEstd1588] 1524 IEEE standard for Information Technology, "IEEE Standard 1525 for a Precision Clock Synchronization Protocol for 1526 Networked Measurement and Control Systems", 1527 . 1529 [CLOS] Yuan, X., "On Nonblocking Folded-Clos Networks in Computer 1530 Communication Environments", IEEE International Parallel & 1531 Distributed Processing Symposium, 2011. 1533 [FATTREE] Leiserson, C. E., "Fat-Trees: Universal Networks for 1534 Hardware-Efficient Supercomputing", 1985. 1536 [RFC3626] Clausen, T., Ed. and P. Jacquet, Ed., "Optimized Link 1537 State Routing Protocol (OLSR)", RFC 3626, 1538 DOI 10.17487/RFC3626, October 2003, 1539 . 1541 [RFC5905] Mills, D., Martin, J., Ed., Burbank, J., and W. Kasch, 1542 "Network Time Protocol Version 4: Protocol and Algorithms 1543 Specification", RFC 5905, DOI 10.17487/RFC5905, June 2010, 1544 . 1546 [RFC8200] Deering, S. and R. Hinden, "Internet Protocol, Version 6 1547 (IPv6) Specification", STD 86, RFC 8200, 1548 DOI 10.17487/RFC8200, July 2017, 1549 . 1551 [RFC8505] Thubert, P., Ed., Nordmark, E., Chakrabarti, S., and C. 1552 Perkins, "Registration Extensions for IPv6 over Low-Power 1553 Wireless Personal Area Network (6LoWPAN) Neighbor 1554 Discovery", RFC 8505, DOI 10.17487/RFC8505, November 2018, 1555 . 1557 [RFC8928] Thubert, P., Ed., Sarikaya, B., Sethi, M., and R. Struik, 1558 "Address-Protected Neighbor Discovery for Low-Power and 1559 Lossy Networks", RFC 8928, DOI 10.17487/RFC8928, November 1560 2020, . 1562 Authors' Addresses 1564 Yuehua Wei (editor) 1565 ZTE Corporation 1566 No.50, Software Avenue 1567 Nanjing 1568 210012 1569 China 1571 Email: wei.yuehua@zte.com.cn 1573 Zheng Zhang 1574 ZTE Corporation 1575 No.50, Software Avenue 1576 Nanjing 1577 210012 1578 China 1580 Email: zhang.zheng@zte.com.cn 1582 Dmitry Afanasiev 1583 Yandex 1585 Email: fl0w@yandex-team.ru 1587 Pascal Thubert 1588 Cisco Systems, Inc 1589 Building D 1590 45 Allee des Ormes - BP1200 1591 06254 MOUGINS - Sophia Antipolis 1592 France 1594 Phone: +33 497 23 26 34 1595 Email: pthubert@cisco.com 1597 Tony Przygienda 1598 Juniper Networks 1599 1194 N. Mathilda Ave 1600 Sunnyvale, CA, 94089 1601 United States of America 1603 Email: prz@juniper.net