idnits 2.17.1 draft-ietf-mpls-spring-entropy-label-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 (April 27, 2018) is 2184 days in the past. Is this intentional? Checking references for intended status: Informational ---------------------------------------------------------------------------- -- Looks like a reference, but probably isn't: '1000' on line 395 -- Looks like a reference, but probably isn't: '1999' on line 395 == Outdated reference: A later version (-22) exists of draft-ietf-spring-segment-routing-mpls-13 == Outdated reference: A later version (-13) exists of draft-ietf-isis-mpls-elc-03 == Outdated reference: A later version (-15) exists of draft-ietf-ospf-mpls-elc-05 Summary: 0 errors (**), 0 flaws (~~), 4 warnings (==), 3 comments (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 2 Network Working Group S. Kini 3 Internet-Draft 4 Intended status: Informational K. Kompella 5 Expires: October 29, 2018 Juniper 6 S. Sivabalan 7 Cisco 8 S. Litkowski 9 Orange 10 R. Shakir 11 Google 12 J. Tantsura 13 April 27, 2018 15 Entropy label for SPRING tunnels 16 draft-ietf-mpls-spring-entropy-label-10 18 Abstract 20 Segment Routing (SR) leverages the source routing paradigm. A node 21 steers a packet through an ordered list of instructions, called 22 segments. Segment Routing can be applied to the Multi Protocol Label 23 Switching (MPLS) data plane. Entropy label (EL) is a technique used 24 in MPLS to improve load-balancing. This document examines and 25 describes how ELs are to be applied to Segment Routing MPLS. 27 Status of This Memo 29 This Internet-Draft is submitted in full conformance with the 30 provisions of BCP 78 and BCP 79. 32 Internet-Drafts are working documents of the Internet Engineering 33 Task Force (IETF). Note that other groups may also distribute 34 working documents as Internet-Drafts. The list of current Internet- 35 Drafts is at https://datatracker.ietf.org/drafts/current/. 37 Internet-Drafts are draft documents valid for a maximum of six months 38 and may be updated, replaced, or obsoleted by other documents at any 39 time. It is inappropriate to use Internet-Drafts as reference 40 material or to cite them other than as "work in progress." 42 This Internet-Draft will expire on October 29, 2018. 44 Copyright Notice 46 Copyright (c) 2018 IETF Trust and the persons identified as the 47 document authors. All rights reserved. 49 This document is subject to BCP 78 and the IETF Trust's Legal 50 Provisions Relating to IETF Documents 51 (https://trustee.ietf.org/license-info) in effect on the date of 52 publication of this document. Please review these documents 53 carefully, as they describe your rights and restrictions with respect 54 to this document. Code Components extracted from this document must 55 include Simplified BSD License text as described in Section 4.e of 56 the Trust Legal Provisions and are provided without warranty as 57 described in the Simplified BSD License. 59 Table of Contents 61 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 3 62 1.1. Requirements Language . . . . . . . . . . . . . . . . . . 4 63 2. Abbreviations and Terminology . . . . . . . . . . . . . . . . 4 64 3. Use-case requiring multipath load-balancing . . . . . . . . . 4 65 4. Entropy Readable Label Depth . . . . . . . . . . . . . . . . 6 66 5. Maximum SID Depth . . . . . . . . . . . . . . . . . . . . . . 7 67 6. LSP stitching using the binding SID . . . . . . . . . . . . . 9 68 7. Insertion of entropy labels for SPRING path . . . . . . . . . 10 69 7.1. Overview . . . . . . . . . . . . . . . . . . . . . . . . 10 70 7.1.1. Example 1 where the ingress node has a sufficient MSD 11 71 7.1.2. Example 2 where the ingress node has not a sufficient 72 MSD . . . . . . . . . . . . . . . . . . . . . . . . . 12 73 7.2. Considerations for the placement of entropy labels . . . 13 74 7.2.1. ERLD value . . . . . . . . . . . . . . . . . . . . . 14 75 7.2.2. Segment type . . . . . . . . . . . . . . . . . . . . 14 76 7.2.2.1. Node-SID . . . . . . . . . . . . . . . . . . . . 15 77 7.2.2.2. Adjacency-set SID . . . . . . . . . . . . . . . . 15 78 7.2.2.3. Adjacency-SID representing a single IP link . . . 15 79 7.2.2.4. Adjacency-SID representing a single link within a 80 L2 bundle . . . . . . . . . . . . . . . . . . . . 16 81 7.2.2.5. Adjacency-SID representing a L2 bundle . . . . . 16 82 7.2.3. Maximizing number of LSRs that will load-balance . . 16 83 7.2.4. Preference for a part of the path . . . . . . . . . . 16 84 7.2.5. Combining criteria . . . . . . . . . . . . . . . . . 17 85 8. A simple example algorithm . . . . . . . . . . . . . . . . . 17 86 9. Deployment Considerations . . . . . . . . . . . . . . . . . . 18 87 10. Options considered . . . . . . . . . . . . . . . . . . . . . 18 88 10.1. Single EL at the bottom of the stack . . . . . . . . . . 18 89 10.2. An EL per segment in the stack . . . . . . . . . . . . . 19 90 10.3. A re-usable EL for a stack of tunnels . . . . . . . . . 19 91 10.4. EL at top of stack . . . . . . . . . . . . . . . . . . . 20 92 10.5. ELs at readable label stack depths . . . . . . . . . . . 20 93 11. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . 21 94 12. Contributors . . . . . . . . . . . . . . . . . . . . . . . . 21 95 13. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 21 96 14. Security Considerations . . . . . . . . . . . . . . . . . . . 22 97 15. References . . . . . . . . . . . . . . . . . . . . . . . . . 22 98 15.1. Normative References . . . . . . . . . . . . . . . . . . 22 99 15.2. Informative References . . . . . . . . . . . . . . . . . 22 100 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 23 102 1. Introduction 104 Segment Routing [I-D.ietf-spring-segment-routing] is based on source 105 routed tunnels to steer a packet along a particular path. This path 106 is encoded as an ordered list of segments. When applied to the MPLS 107 dataplane [I-D.ietf-spring-segment-routing-mpls], each segment is an 108 LSP with an associated MPLS label value. Hence, label stacking is 109 used to represent the ordered list of segments and the label stack 110 associated with an SR tunnel can be seen as nested LSPs (LSP 111 hierarchy) in the MPLS architecture. 113 Using label stacking to encode the list of segment has implications 114 on the label stack depth. 116 Traffic load-balancing over ECMP (Equal Cost MultiPath) or LAGs (Link 117 Aggregation Groups) is usually based on a hashing function. The 118 local node who performs the load-balancing is required to read some 119 header fields in the incoming packets and then computes a hash based 120 on these fields. The result of the hash is finally mapped to a list 121 of outgoing nexthops. The hashing technique is required to perfom a 122 per-flow load-balancing and thus prevent packet disordering. For IP 123 traffic, the usual fields that are looked up are the source address, 124 the destination address, the protocol type, and, if the upper layer 125 is TCP or UDP, the source port and destination port can be added as 126 well in the hash. 128 The MPLS architecture brings some challenges on the load-balancing as 129 an LSR (Label Switch Router) should be able to look at header fields 130 that are beyond the MPLS label stack. An LSR must perform a deeper 131 inspection compared to an ingress router which could be challenging 132 for some hardware. Entropy label (EL) [RFC6790] is a technique used 133 in the MPLS data plane to provide entropy for load-balancing. The 134 idea behind entropy label is that the ingress router computes a hash 135 based on several fields from a given packet and place the result in 136 an additional label, named "entropy label". When using entropy 137 label, an LSR is only required to hash on the MPLS label stack or 138 solely on the entropy label to get a full benefit of load-balancing; 139 deep hashing is not required anymore. 141 When using LSP hierarchies, there are implications on how [RFC6790] 142 should be applied. The current document addresses the case where a 143 hierarchy is created at a single LSR as required by Segment Routing. 145 A use-case requiring load-balancing with SR is given in Section 3. A 146 recommended solution is described in Section 7 keeping in 147 consideration the limitations of implementations when applying 148 [RFC6790] to deeper label stacks. Options that were considered to 149 arrive at the recommended solution are documented for historical 150 purposes in Section 10. 152 1.1. Requirements Language 154 The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", 155 "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and 156 "OPTIONAL" in this document are to be interpreted as described in BCP 157 14 [RFC2119] [RFC8174] when, and only when, they appear in all 158 capitals, as shown here. 160 2. Abbreviations and Terminology 162 EL - Entropy Label 164 ELI - Entropy Label Identifier 166 ELC - Entropy Label Capability 168 ERLD - Entropy Readable Label Depth 170 SR - Segment Routing 172 ECMP - Equal Cost Multi Path 174 LSR - Label Switch Router 176 MPLS - Multiprotocol Label Switching 178 MSD - Maximum SID Depth 180 SID - Segment Identifier 182 RLD - Readable Label Depth 184 OAM - Operation, Administration and Maintenance 186 3. Use-case requiring multipath load-balancing 187 +------+ 188 | | 189 +-------| P3 |-----+ 190 | +-----| |---+ | 191 L3| |L4 +------+ L1| |L2 +----+ 192 | | | | +--| P4 |--+ 193 +-----+ +-----+ +-----+ | +----+ | +-----+ 194 | S |-----| P1 |------------| P2 |--+ +--| D | 195 | | | | | |--+ +--| | 196 +-----+ +-----+ +-----+ | +----+ | +-----+ 197 +--| P5 |--+ 198 +----+ 199 S=Source LSR, D=Destination LSR, P1,P2,P3,P4,P5=Transit LSRs, 200 L1,L2,L3,L4=Links 202 Figure 1: Traffic engineering use-case 204 Traffic-engineering is one of the applications of MPLS and is also a 205 requirement for source routed tunnels with label stacks [RFC7855]. 206 Consider the topology shown in Figure 1. The LSR S requires data to 207 be sent to LSR D along a traffic-engineered path that goes over the 208 link L1. Good load-balancing is also required across equal cost 209 paths (including parallel links). To engineer traffic along a path 210 that takes link L1, the label stack that LSR S creates consists of a 211 label to the node SID of LSR P3, stacked over the label for the 212 adjacency SID of link L1 and that in turn is stacked over the label 213 to the node SID of LSR D. For simplicity lets assume that all LSRs 214 use the same label space (SRGB) for source routed label stacks. Let 215 L_N-Px denote the label to be used to reach the node SID of LSR Px. 216 Let L_A-Ln denote the label used for the adjacency SID for link Ln. 217 The LSR S must use the label stack for 218 traffic-engineering. However to achieve good load-balancing over the 219 equal cost paths P2-P4-D, P2-P5-D and the parallel links L3, L4, a 220 mechanism such as Entropy labels [RFC6790] should be adapted for 221 source routed label stacks. Indeed, the SPRING architecture with the 222 MPLS dataplane ([I-D.ietf-spring-segment-routing-mpls]) uses nested 223 MPLS LSPs composing the source routed label stacks. 225 An MPLS node may have limitations in the number of labels it can 226 push. It may also have a limitation in the number of labels it can 227 inspect when looking for hash keys during load-balancing. While 228 entropy label is normally inserted at the bottom of the transport 229 tunnel, this may prevent an LSR to take into account the EL in its 230 load-balancing function if the EL is too deep in the stack. In a 231 segment routing environment, it is important to define the 232 considerations that needs to be taken into account when inserting EL. 233 Multiple ways to apply entropy labels were considered and are 234 documented in Section 10 along with their trade-offs. A recommended 235 solution is described in Section 7. 237 4. Entropy Readable Label Depth 239 The Entropy Readable Label Depth (ERLD) is defined as the number of 240 labels a router can both: 242 a. Read in an MPLS packet received on its incoming interface(s) 243 (starting from the top of the stack). 245 b. Use in its load-balancing function. 247 The ERLD means that the router will perform load-balancing using the 248 EL label if the EL is placed within the ERLD first labels. 250 A router capable of reading N labels but not using an EL located 251 within those N labels MUST consider its ERLD to be 0. In a 252 distributed switching architecture, each linecard may have a 253 different capability in terms of ERLD. For simplicity, an 254 implementation MAY use the minimum ERLD between each linecard as the 255 ERLD value for the system. 257 Examples: 259 | Payload | 260 +----------+ 261 | Payload | | EL | P7 262 +----------+ +----------+ 263 | Payload | | EL | | ELI | 264 +----------+ +----------+ +----------+ 265 | Payload | | EL | | ELI | | Label 50 | 266 +----------+ +----------+ +----------+ +----------+ 267 | Payload | | EL | | ELI | | Label 40 | | Label 40 | 268 +----------+ +----------+ +----------+ +----------+ +----------+ 269 | EL | | ELI | | Label 30 | | Label 30 | | Label 30 | 270 +----------+ +----------+ +----------+ +----------+ +----------+ 271 | ELI | | Label 20 | | Label 20 | | Label 20 | | Label 20 | 272 +----------+ +----------+ +----------+ +----------+ +----------+ 273 | Label 16 | | Label 16 | | Label 16 | | Label 16 | | Label 16 | P1 274 +----------+ +----------+ +----------+ +----------+ +----------+ 275 Packet 1 Packet 2 Packet 3 Packet 4 Packet 5 277 Figure 2: Label stacks with ELI/EL 279 In the figure 2, we consider the displayed packets received on a 280 router interface. We consider also a single ERLD value for the 281 router. 283 o If the router has an ERLD of 3, it will be able to load-balance 284 Packet 1 displayed in Figure 2 using the EL as part of the load- 285 balancing keys. The ERLD value of 3 means that the router can 286 read and take into account the entropy label for load-balancing if 287 it is placed between position 1 (top) and position 3. 289 o If the router has an ERLD of 5, it will be able to load-balance 290 Packets 1 to 3 in Figure 2 using the EL as part of the load- 291 balancing keys. Packets 4 and 5 have the EL placed at a position 292 greater than 5, so the router is not able to read it and use as 293 part of the load-balancing keys. 295 o If the router has an ERLD of 10, it will be able to load-balance 296 all the packets displayed in Figure 2 using the EL as part of the 297 load-balancing keys. 299 To allow an efficient load-balancing based on entropy labels, a 300 router running SPRING SHOULD advertise its ERLD (or ERLDs), so all 301 the other SPRING routers in the network are aware of its capability. 302 How this advertisement is done is outside the scope of this document. 304 To advertise an ERLD value, a SPRING router: 306 o MUST be entropy label capable and, as a consequence, MUST apply 307 the dataplane procedures defined in [RFC6790]. 309 o MUST be able to read an ELI/EL which is located within its ERLD 310 value. 312 o MUST take into account this EL in its load-balancing function. 314 5. Maximum SID Depth 316 The Maximum SID Depth defines the maximum number of labels that a 317 particular node can impose on a packet. This includes any kind of 318 labels (service, entropy, transport...). In an MPLS network, the MSD 319 is a limit of the Ingress LSR (I-LSR) or any stitching node that 320 would perform an imposition of additional labels on an existing label 321 stack. 323 Depending of the number of MPLS operations (POP, SWAP...) to be 324 performed before the PUSH, the MSD may vary due to the hardware or 325 software limitations. As for the ERLD, there may also be different 326 MSD limits based on the linecard type used in a distributed switching 327 system. 329 When an external controller is used to program a label stack on a 330 particular node, this node MAY advertise its MSD value or a subset of 331 its MSD value to the controller. How this advertisement is done is 332 outside the scope of this document. As the controller does not have 333 the knowledge of the entire label stack to be pushed by the node, the 334 node may advertise an MSD value which is lower than its actual limit. 335 This gives the ability for the controller to program a label stack up 336 to the advertised MSD value while leaving room for the local node to 337 add more labels (e.g., service, entropy, transport...) without 338 reaching the hardware/software limit. 340 P7 ---- P8 ---- P9 341 / \ 342 PE1 --- P1 --- P2 --- P3 --- P4 --- P5 --- P6 --- PE2 343 | \ | 344 ----> P10 \ | 345 IP Pkt | \ | 346 P11 --- P12 --- P13 347 100 10000 349 Figure 3 351 In the figure 3, an IP packet comes in the MPLS network at PE1. All 352 metrics are considered equal to 1 except P12-P13 which is 10000 and 353 P11-P12 which is 100. PE1 wants to steer the traffic using a SPRING 354 path to PE2 along 355 PE1->P1->P7->P8->P9->P4->P5->P10->P11->P12->P13->PE2. By using 356 adjacency SIDs only, PE1 (acting as an I-LSR) will be required to 357 push 10 labels on the IP packet received and thus requires an MSD of 358 10. If the IP packet should be carried over an MPLS service like a 359 regular layer 3 VPN, an additional service label should be imposed, 360 requiring an MSD of 11 for PE1. In addition, if PE1 wants to insert 361 an ELI/EL for load-balancing purpose, PE1 will need to push 13 labels 362 on the IP packet requiring an MSD of 13. 364 In the SPRING architecture, Node SIDs or Binding SIDs can be used to 365 reduce the label stack size. As an example, to steer the traffic on 366 the same path as before, PE1 may be able to use the following label 367 stack: . In this example we 368 consider a combination of Node SIDs and a Binding SID advertised by 369 P5 that will stitch the traffic along the path P10->P11->P12->P13. 370 The instruction associated with the binding SID at P5 is thus to swap 371 Binding_P5 to Adj_P12-P13 and then push . P5 372 acts as a stitching node that pushes additional labels on an existing 373 label stack, P5's MSD needs also to be taken into account and may 374 limit the number of labels that could be imposed. 376 6. LSP stitching using the binding SID 378 The binding SID allows binding a segment identifier to an existing 379 LSP. As examples, the binding SID can represent an RSVP-TE tunnel, 380 an LDP path (through the mapping server advertisement), or a SPRING 381 path. Each LSP associated with a binding SID has its own entropy 382 label capability. 384 In the figure 3, we consider that: 386 o P6, PE2, P10, P11, P12, P13 are pure LDP routers. 388 o PE1, P1, P2, P3, P4, P7, P8, P9 are pure SPRING routers. 390 o P5 is running SPRING and LDP. 392 o P5 acts as a mapping server and advertises Prefix SIDs for the LDP 393 FECs: an index value of 20 is used for PE2. 395 o All SPRING routers use an SRGB of [1000, 1999]. 397 o P6 advertises label 20 for the PE2 FEC. 399 o Traffic from PE1 to PE2 uses the shortest path. 401 PE1 ----- P1 -- P2 -- P3 -- P4 ---- P5 --- P6 --- PE2 403 --> +----+ +----+ +----+ +----+ 404 IP Pkt | IP | | IP | | IP | | IP | 405 +----+ +----+ +----+ +----+ 406 |1020| |1020| | 20 | 407 +----+ +----+ +----+ 408 SPRING LDP 410 In term of packet forwarding, by learning the mapping-server 411 advertisement from P5, PE1 imposes a label 1020 to an IP packet 412 destinated to PE2. SPRING routers along the shortest path to PE2 413 will switch the traffic until it reaches P5 which will perform the 414 LSP stitching. P5 will swap the SPRING label 1020 to the LDP label 415 20 advertised by the nexthop P6. P6 will then forward the packet 416 using the LDP label towards PE2. 418 PE1 cannot push an ELI/EL for the binding SID without knowing that 419 the tail-end of the LSP associated with the binding (PE2) is entropy 420 label capable. 422 To accomodate the mix of signaling protocols involved during the 423 stitching, the entropy label capability SHOULD be propagated between 424 the signaling domains. Each binding SID SHOULD have its own entropy 425 label capability that MUST be inherited from the entropy label 426 capability of the associated LSP. If the router advertising the 427 binding SID does not know the ELC state of the target FEC, it MUST 428 NOT set the ELC for the binding SID. An ingress node MUST NOT push 429 an ELI/EL associated with a binding SID unless this binding SID has 430 the entropy label capability. How the entropy label capability is 431 advertised for a binding SID is outside the scope of this document. 433 In our example, if PE2 is LDP entropy label capable, it will add the 434 entropy label capability in its LDP advertisement. When P5 receives 435 the FEC/label binding for PE2, it learns about the ELC and can set 436 the ELC in the mapping server advertisement. Thus PE1 learns about 437 the ELC of PE2 and may push an ELI/EL associated with the binding 438 SID. 440 The proposed solution only works if the SPRING router advertising the 441 binding SID is also performing the dataplane LSP stitching. In our 442 example, if the mapping server function is hosted on P8 instead of 443 P5, P8 does not know about the ELC state of PE2's LDP FEC. As a 444 consequence, it does not set the ELC for the associated binding SID. 446 7. Insertion of entropy labels for SPRING path 448 7.1. Overview 450 The solution described in this section follows the dataplane 451 processing defined in [RFC6790]. Within a SPRING path, a node may be 452 ingress, egress, transit (regarding the entropy label processing 453 described in [RFC6790]), or it can be any combination of those. For 454 example: 456 o The ingress node of a SPRING domain may be an ingress node from an 457 entropy label perspective. 459 o Any LSR terminating a segment of the SPRING path is an egress node 460 (because it terminates the segment) but may also be a transit node 461 if the SPRING path is not terminated because there is a subsequent 462 SPRING MPLS label in the stack. 464 o Any LSR processing a binding SID may be a transit node and an 465 ingress node (because it may push additional labels when 466 processing the binding SID). 468 As described earlier, an LSR may have a limitation, ERLD, on the 469 depth of the label stack that it can read and process in order to do 470 multipath load-balancing based on entropy labels. 472 If an EL does not occur within the ERLD of an LSR in the label stack 473 of an MPLS packet that it receives, then it would lead to poor load- 474 balancing at that LSR. Hence an ELI/EL pair must be within the ERLD 475 of the LSR in order for the LSR to use the EL during load-balancing. 477 Adding a single ELI/EL pair for the entire SPRING path may lead also 478 to poor load-balancing as well because the EL/ELI may not occur 479 within the ERLD of some LSR on the path (if too deep) or may not be 480 present in the stack when it reaches some LSRs if it is too shallow. 482 In order for the EL to occur within the ERLD of LSRs along the path 483 corresponding to a SPRING label stack, multiple pairs MAY 484 be inserted in this label stack. 486 The insertion of the ELI/EL SHOULD occur only with a SPRING label 487 advertised by an LSR that advertised an ERLD (the LSR is entropy 488 label capable) or with a SPRING label associated with a binding SID 489 that has the ELC set. 491 The ELs among multiple pairs inserted in the stack MAY be 492 the same or different. The LSR that inserts pairs MAY have 493 limitations on the number of such pairs that it can insert and also 494 the depth at which it can insert them. If, due to limitations, the 495 inserted ELs are at positions such that an LSR along the path 496 receives an MPLS packet without an EL in the label stack within that 497 LSR's ERLD, then the load-balancing performed by that LSR would be 498 poor. An implementation MAY consider multiple criteria when 499 inserting pairs. 501 7.1.1. Example 1 where the ingress node has a sufficient MSD 503 ECMP LAG LAG 504 PE1 --- P1 --- P2 --- P3 --- P4 --- P5 --- P6 --- PE2 506 Figure 4 508 In the figure 4, PE1 wants to forward some MPLS VPN traffic over an 509 explicit path to PE2 resulting in the following label stack to be 510 pushed onto the received IP header: . PE1 is limited 512 to push a maximum of 11 labels (MSD=11). P2, P3 and P6 have an ERLD 513 of 3 while others have an ERLD of 10. 515 PE1 can only add two ELI/EL pairs in the label stack due to its MSD 516 limitation. It should insert them strategically to benefit load- 517 balancing along the longest part of the path. 519 PE1 may take into account multiple parameters when inserting ELs, as 520 examples: 522 o The ERLD value advertised by transit nodes. 524 o The requirement of load-balancing for a particular label value. 526 o Any service provider preference: favor beginning of the path or 527 end of the path. 529 In the figure 4, a good strategy may be to use the following stack 530 . The original stack requests P2 to forward 532 based on a L3 adjacency set that will require load-balancing. 533 Therefore it is important to ensure that P2 can load-balance 534 correctly. As P2 has a limited ERLD of 3, ELI/EL must be inserted 535 just next to the label that P2 will use to forward. On the path to 536 PE2, P3 has also a limited ERLD, but P3 will forward based on a basic 537 adjacency segment that may require no load-balancing. Therefore it 538 does not seem important to ensure that P3 can do load-balancing 539 despite of its limited ERLD. The next nodes along the forwarding 540 path have a high ERLD that does not cause any issue, except P6, 541 moreover P6 is using some LAGs to PE2 and so is expected to load- 542 balance. It becomes important to insert a new ELI/EL just next to P6 543 forwarding label. 545 In the case above, the ingress node had enough label push capacity to 546 ensure end-to-end load-balancing taking into the path attributes. 547 There might be some cases, where the ingress node may not have the 548 necessary label imposition capacity. 550 7.1.2. Example 2 where the ingress node has not a sufficient MSD 552 ECMP LAG ECMP ECMP 553 PE1 --- P1 --- P2 --- P3 --- P4 --- P5 --- P6 --- P7 --- P8 --- PE2 555 Figure 5 557 In the figure 5, PE1 wants to forward MPLS VPN traffic over an 558 explicit path to PE2 resulting in the following label stack to be 559 pushed onto the IP header: . PE1 is limited to push a maximum of 11 labels, P2, P3 562 and P6 have an ERLD of 3 while others have an ERLD of 15. 564 Using a similar strategy as the previous case may lead to a dilemma, 565 as PE1 can only push a single ELI/EL while we may need a minimum of 566 three to load-balance the end-to-end path. An optimized stack that 567 would enable end-to-end load-balancing may be: . 571 A decision needs to be taken to favor some part of the path for load- 572 balancing considering that load-balancing may not work on the other 573 part. A service provider may decide to place the ELI/EL after the P6 574 forwarding label as it will allow P4 and P6 to load-balance. Placing 575 the ELI/EL at bottom of the stack is also a possibility enabling 576 load-balancing for P4 and P8. 578 7.2. Considerations for the placement of entropy labels 580 The sample cases described in the previous section showed that 581 placing the ELI/EL when the maximum number of labels to be pushed is 582 limited is not an easy decision and multiple criteria may be taken 583 into account. 585 This section describes some considerations that could be taken into 586 account when placing ELI/ELs. This list of criteria is not 587 considered as exhaustive and an implementation MAY take into account 588 additional criteria or tie-breakers that are not documented here. 590 An implementation SHOULD try to maximize the load-balancing where 591 multiple ECMP paths are available and minimize the number of EL/ELIs 592 that need to be inserted. In case of a trade-off, an implementation 593 MAY provide flexibility to the operator to select the criteria to be 594 considered when placing EL/ELIs or the sub-objective for which to 595 optimize. 597 2 2 598 PE1 -- P1 -- P2 --P3 --- P4 --- P5 -- ... -- P8 -- P9 -- PE2 599 | | 600 P3'--- P4'--- P5' 602 Figure 6 604 The figure above will be used as reference in the following 605 subsections. All metrics are equal to 1, except P3-P4 and P4-P5 606 which have a metric 2. 608 7.2.1. ERLD value 610 As mentioned in Section 7.1, the ERLD value is an important parameter 611 to consider when inserting ELI/EL. If an ELI/EL does not fall within 612 the ERLD of a node on the path, the node will not be able to load- 613 balance the traffic efficiently. 615 The ERLD value can be advertised via protocols and those extensions 616 are described in separate documents [I-D.ietf-isis-mpls-elc] and 617 [I-D.ietf-ospf-mpls-elc]. 619 Let's consider a path from PE1 to PE2 using the following stack 620 pushed by PE1: . 622 Using the ERLD as an input parameter may help to minimize the number 623 of required ELI/EL pairs to be inserted. An ERLD value must be 624 retrieved for each SPRING label in the label stack. 626 For a label bound to an adjacency segment, the ERLD is the ERLD of 627 the node that advertised the adjacency segment. In the example 628 above, the ERLD associated with Adj_P1P2 would be the ERLD of router 629 P1 as P1 will perform the forwarding based on the Adj_P1P2 label. 631 For a label bound to a node segment, multiple strategies MAY be 632 implemented. An implementation may try to evaluate the minimum ERLD 633 value along the node segment path. If an implementation cannot find 634 the minimum ERLD along the path of the segment, it can use the ERLD 635 of the starting node instead. In the example above, if the 636 implementation supports computation of minimum ERLD along the path, 637 the ERLD associated with label Node_P9 would be the minimum ERLD 638 between nodes {P2,P3,P4 ..., P8}. If an implementation does not 639 support the computation of minimum ERLD, it should consider the ERLD 640 of P2 (starting node that will forward based on the Node_P9 label). 642 For a label bound to a binding segment, if the binding segment 643 describes a path, an implementation may also try to evaluate the 644 minimum ERLD along this path. If the implementation cannot find the 645 minimum ERLD along the path of the segment, it can use the ERLD of 646 the starting node instead. 648 7.2.2. Segment type 650 Depending of the type of segment a particular label is bound to, an 651 implementation may deduce that this particular label will be subject 652 to load-balancing on the path. 654 7.2.2.1. Node-SID 656 An MPLS label bound to a Node-SID represents a path that may cross 657 multiple hops. Load-balancing may be needed on the node starting 658 this path but also on any node along the path. 660 In the figure 6, let's consider a path from PE1 to PE2 using the 661 following stack pushed by PE1: . 664 If, for example, PE1 is limited to push 6 labels, it can add a single 665 ELI/EL within the label stack. An operator may want to favor a 666 placement that would allow load-balancing along the Node-SID path. 667 In the figure above, P3 which is along the Node-SID path requires 668 load-balancing on two equal-cost paths. 670 An implementation may try to evaluate if load-balancing is really 671 required within a node segment path. This could be done by running 672 an additional SPT computation and analysis of the node segment path 673 to prevent a node segment that does not really require load-balancing 674 from being preferred when placing EL/ELIs. Such inspection may be 675 time consuming for implementations and without a 100% guarantee, as a 676 node segment path may use LAG that could be invisible from the IP 677 topology. A simpler approach would be to consider that a label bound 678 to a Node-SID will be subject to load-balancing and requires an EL/ 679 ELI. 681 7.2.2.2. Adjacency-set SID 683 An adjacency-set is an adjacency SID that refers to a set of 684 adjacencies. When an adjacency-set segment is used within a label 685 stack, an implementation can deduce that load-balancing is expected 686 at the node that advertised this adjacency segment. An 687 implementation could then favor this particular label value when 688 placing ELI/ELs. 690 7.2.2.3. Adjacency-SID representing a single IP link 692 When an adjacency segment representing a single IP link is used 693 within a label stack, an implementation can deduce that load- 694 balancing may not be expected at the node that advertised this 695 adjacency segment. 697 The implementation could then decide to place ELI/ELs to favor other 698 LSRs than the one advertising this adjacency segment. 700 Readers should note that an adjacency segment representing a single 701 IP link may require load-balancing. This is the case when a LAG (L2 702 bundle) is implemented between two IP nodes and the L2 bundle SR 703 extensions [I-D.ietf-isis-l2bundles] are not implemented. In such a 704 case, it may be useful to insert an EL/ELI in a readable position for 705 the LSR advertising the label associated with the adjacency segment. 707 7.2.2.4. Adjacency-SID representing a single link within a L2 bundle 709 When L2 bundle SR extensions [I-D.ietf-isis-l2bundles] are used, 710 adjacency segments may be advertised for each member of the bundle. 711 In this case, an implementation can deduce that load-balancing is not 712 expected on the LSR advertising this segment and could then decide to 713 place ELI/ELs to favor other LSRs than the one advertising this 714 adjacency segment. 716 7.2.2.5. Adjacency-SID representing a L2 bundle 718 When L2 bundle SR extensions [I-D.ietf-isis-l2bundles] are used, an 719 adjacency segment may be advertised to represent the bundle. In this 720 case, an implementation can deduce that load-balancing is expected on 721 the LSR advertising this segment and could then decide to place ELI/ 722 ELs to favor this LSR. 724 7.2.3. Maximizing number of LSRs that will load-balance 726 When placing ELI/ELs, an implementation may try to maximize the 727 number of LSRs that both need to load-balance (i.e., have ECMP paths) 728 and that will be able to perform load-balancing (i.e., the EL label 729 is within their ERLD). 731 Let's consider a path from PE1 to PE2 using the following stack 732 pushed by PE1: . All 733 routers have an ERLD of 10, expect P1 and P2 which have an ERLD of 4. 734 PE1 is able to push 6 labels, so only a single ELI/EL can be added. 736 In the example above, adding ELI/EL next to Adj_P1P2 will only allow 737 load-balancing at P1 while inserting it next to Adj_PE2P9, will allow 738 load-balancing at P2,P3 ... P9 and maximizing the number of LSRs that 739 could perform load-balancing. 741 7.2.4. Preference for a part of the path 743 An implementation may propose to favor a part of the end-to-end path 744 when the number of EL/ELI that can be pushed is not enough to cover 745 the entire path. As example, a service provider may want to favor 746 load-balancing at the beginning of the path or at the end of path, so 747 the implementation should prefer putting the ELI/ELs near the top or 748 near of the bottom of the stack. 750 7.2.5. Combining criteria 752 An implementation can combine multiple criteria to determine the best 753 EL/ELIs placement. However, combining too many criteria may lead to 754 implementation complexity and high resource consumption. Each time 755 the network topology changes, a new evaluation of the EL/ELI 756 placement will be necessary for each impacted LSPs. 758 8. A simple example algorithm 760 A simple implementation might take into account ERLD when placing 761 ELI/EL while trying to minimize the number of EL/ELIs inserted and 762 trying to maximize the number of LSRs that can load-balance. 764 The example algorithm is based on the following considerations: 766 o An LSR that is limited in the number of pairs that it 767 can insert SHOULD insert such pairs deeper in the stack. 769 o An LSR should try to insert pairs at positions so that 770 for the maximum number of transit LSRs, the EL occurs within the 771 ERLD of those LSRs. 773 o An LSR should try to insert the minimum number of such pairs while 774 trying to satisfy the above criteria. 776 The pseudocode of the example algorithm is shown below. 778 Initialize the current EL insertion point to the 779 bottommost label in the stack that is EL-capable 780 while (local-node can push more pairs OR 781 insertion point is not above label stack) { 782 insert an pair below current insertion point 783 move new insertion point up from current insertion point until 784 ((last inserted EL is below the ERLD) AND (ERLD > 2) 785 AND 786 (new insertion point is EL-capable)) 787 set current insertion point to new insertion point 788 } 790 Figure 7: Example algorithm to insert pairs in a label 791 stack 793 When this algorithm is applied to the example described in Section 3, 794 it will result in ELs being inserted in two positions, one below the 795 label L_N-D and another below L_N-P3. Thus the resulting label stack 796 would be 798 9. Deployment Considerations 800 As long as LSR node dataplane capabilities are limited (number of 801 labels that can be pushed, or number of labels that can be 802 inspected), hop-by-hop load-balancing of SPRING encapsulated flows 803 will require trade-offs. 805 Entropy label is still a good and usable solution as it allows load- 806 balancing without having to perform a deep packet inspection on each 807 LSR: it does not seem reasonable to have an LSR inspecting UDP ports 808 within a GRE tunnel carried over a 15 label SPRING tunnel. 810 Due to the limited capacity of reading a deep stack of MPLS labels, 811 multiple EL/ELIs may be required within the stack which directly 812 impacts the capacity of the head-end to push a deep stack: each EL/ 813 ELI inserted requires two additional labels to be pushed. 815 Placement strategies of EL/ELIs are required to find the best trade- 816 off. Multiple criteria may be taken into account and some level of 817 customization (by the user) may be required to accommodate the 818 different deployments. Analyzing the path of each destination to 819 determine the best EL/ELI placement may be time consuming for the 820 control plane, we encourage implementations to find the best trade- 821 off between simplicity, resource consumption, and load-balancing 822 efficiency. 824 In future, hardware and software capacity may increase dataplane 825 capabilities and may be remove some of these limitations, increasing 826 load-balancing capability using entropy labels. 828 10. Options considered 830 Different options that were considered to arrive at the recommended 831 solution are documented in this section. 833 These options are detailed here only for historical purposes. 835 10.1. Single EL at the bottom of the stack 837 In this option, a single EL is used for the entire label stack. The 838 source LSR S encodes the entropy label at the bottom of the label 839 stack. In the example described in Section 3, it will result in the 840 label stack at LSR S to look like 841 . Note that the notation in [RFC6790] is 842 used to describe the label stack. An issue with this approach is 843 that as the label stack grows due an increase in the number of SIDs, 844 the EL goes correspondingly deeper in the label stack. Hence, 845 transit LSRs have to access a larger number of bytes in the packet 846 header when making forwarding decisions. In the example described in 847 Section 3, if we consider that the LSR P1 has an ERLD of 3, P1 would 848 load-balance traffic poorly on the parallel links L3 and L4 since the 849 EL is below the ERLD of P1. A load-balanced network design using 850 this approach must ensure that all intermediate LSRs have the 851 capability to read the maximum label stack depth as required for the 852 application that uses source routed stacking. 854 This option was rejected since there exist a number of hardware 855 implementations which have a low maximum readable label depth. 856 Choosing this option can lead to a loss of load-balancing using EL in 857 a significant part of the network when that is a critical requirement 858 in a service-provider network. 860 10.2. An EL per segment in the stack 862 In this option, each segment/label in the stack can be given its own 863 EL. When load-balancing is required to direct traffic on a segment, 864 the source LSR pushes an before pushing the label 865 associated to this segment . In the example described in Section 3, 866 the source LSR S encoded label stack would be where all the ELs can be the same. Accessing the 868 EL at an intermediate LSR is independent of the depth of the label 869 stack and hence independent of the specific application that uses 870 source routed tunnels with label stacking. A drawback is that the 871 depth of the label stack grows significantly, almost 3 times as the 872 number of labels in the label stack. The network design should 873 ensure that source LSRs have the capability to push such a deep label 874 stack. Also, the bandwidth overhead and potential MTU issues of deep 875 label stacks should be considered in the network design. 877 This option was rejected due to the existence of hardware 878 implementations that can push a limited number of labels on the label 879 stack. Choosing this option would result in a hardware requirement 880 to push two additional labels per tunnel label. Hence it would 881 restrict the number of tunnels that can be stacked in a LSP and hence 882 constrain the types of LSPs that can be created. This was considered 883 unacceptable. 885 10.3. A re-usable EL for a stack of tunnels 887 In this option an LSR that terminates a tunnel re-uses the EL of the 888 terminated tunnel for the next inner tunnel. It does this by storing 889 the EL from the outer tunnel when that tunnel is terminated and re- 890 inserting it below the next inner tunnel label during the label swap 891 operation. The LSR that stacks tunnels should insert an EL below the 892 outermost tunnel. It should not insert ELs for any inner tunnels. 893 Also, the penultimate hop LSR of a segment must not pop the ELI and 894 EL even though they are exposed as the top labels since the 895 terminating LSR of that segment would re-use the EL for the next 896 segment. 898 In Section 3 above, the source LSR S encoded label stack would be 899 . At P1, the outgoing label stack 900 would be after it has load-balanced 901 to one of the links L3 or L4. At P3 the outgoing label stack would 902 be . At P2, the outgoing label stack would be and it would load-balance to one of the nexthop LSRs P4 904 or P5. Accessing the EL at an intermediate LSR (e.g., P1) is 905 independent of the depth of the label stack and hence independent of 906 the specific use-case to which the label stack is applied. 908 This option was rejected due to the significant change in label swap 909 operations that would be required for existing hardware. 911 10.4. EL at top of stack 913 A slight variant of the re-usable EL option is to keep the EL at the 914 top of the stack rather than below the tunnel label. In this case, 915 each LSR that is not terminating a segment should continue to keep 916 the received EL at the top of the stack when forwarding the packet 917 along the segment. An LSR that terminates a segment should use the 918 EL from the terminated segment at the top of the stack when 919 forwarding onto the next segment. 921 This option was rejected due to the significant change in label swap 922 operations that would be required for existing hardware. 924 10.5. ELs at readable label stack depths 926 In this option the source LSR inserts ELs for tunnels in the label 927 stack at depths such that each LSR along the path that must load 928 balance is able to access at least one EL. Note that the source LSR 929 may have to insert multiple ELs in the label stack at different 930 depths for this to work since intermediate LSRs may have differing 931 capabilities in accessing the depth of a label stack. The label 932 stack depth access value of intermediate LSRs must be known to create 933 such a label stack. How this value is determined is outside the 934 scope of this document. This value can be advertised using a 935 protocol such as an IGP. 937 Applying this method to the example in Section 3 above, if LSR P1 938 needs to have the EL within a depth of 4, then the source LSR S 939 encoded label stack would be where all the ELs would typically have the same value. 942 In the case where the ERLD has different values along the path and 943 the LSR that is inserting pairs has no limit on how many 944 pairs it can insert, and it knows the appropriate positions in the 945 stack where they should be inserted, this option is the same as the 946 recommended solution in Section 7. 948 Note that a refinement of this solution which balances the number of 949 pushed labels against the desired entropy is the solution described 950 in Section 7. 952 11. Acknowledgements 954 The authors would like to thank John Drake, Loa Andersson, Curtis 955 Villamizar, Greg Mirsky, Markus Jork, Kamran Raza, Carlos Pignataro, 956 Bruno Decraene, Chris Bowers and Nobo Akiya for their review comments 957 and suggestions. 959 12. Contributors 961 Xiaohu Xu 962 Huawei 964 Email: xuxiaohu@huawei.com 966 Wim Hendrickx 967 Nokia 969 Email: wim.henderickx@nokia.com 971 Gunter Van De Velde 972 Nokia 974 Email: gunter.van_de_velde@nokia.com 976 Acee Lindem 977 Cisco 979 Email: acee@cisco.com 981 13. IANA Considerations 983 This memo includes no request to IANA. Note to RFC Editor: Remove 984 this section before publication. 986 14. Security Considerations 988 This document does not introduce any new security considerations 989 beyond those already listed in [RFC6790]. 991 15. References 993 15.1. Normative References 995 [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate 996 Requirement Levels", BCP 14, RFC 2119, 997 DOI 10.17487/RFC2119, March 1997, 998 . 1000 [RFC6790] Kompella, K., Drake, J., Amante, S., Henderickx, W., and 1001 L. Yong, "The Use of Entropy Labels in MPLS Forwarding", 1002 RFC 6790, DOI 10.17487/RFC6790, November 2012, 1003 . 1005 [RFC7855] Previdi, S., Ed., Filsfils, C., Ed., Decraene, B., 1006 Litkowski, S., Horneffer, M., and R. Shakir, "Source 1007 Packet Routing in Networking (SPRING) Problem Statement 1008 and Requirements", RFC 7855, DOI 10.17487/RFC7855, May 1009 2016, . 1011 [RFC8174] Leiba, B., "Ambiguity of Uppercase vs Lowercase in RFC 1012 2119 Key Words", BCP 14, RFC 8174, DOI 10.17487/RFC8174, 1013 May 2017, . 1015 [I-D.ietf-spring-segment-routing] 1016 Filsfils, C., Previdi, S., Ginsberg, L., Decraene, B., 1017 Litkowski, S., and R. Shakir, "Segment Routing 1018 Architecture", draft-ietf-spring-segment-routing-15 (work 1019 in progress), January 2018. 1021 [I-D.ietf-spring-segment-routing-mpls] 1022 Bashandy, A., Filsfils, C., Previdi, S., Decraene, B., 1023 Litkowski, S., and R. Shakir, "Segment Routing with MPLS 1024 data plane", draft-ietf-spring-segment-routing-mpls-13 1025 (work in progress), April 2018. 1027 15.2. Informative References 1029 [I-D.ietf-isis-mpls-elc] 1030 Xu, X., Kini, S., Sivabalan, S., Filsfils, C., and S. 1031 Litkowski, "Signaling Entropy Label Capability and 1032 Readable Label-stack Depth Using IS-IS", draft-ietf-isis- 1033 mpls-elc-03 (work in progress), January 2018. 1035 [I-D.ietf-ospf-mpls-elc] 1036 Xu, X., Kini, S., Sivabalan, S., Filsfils, C., and S. 1037 Litkowski, "Signaling Entropy Label Capability and 1038 Readable Label-stack Depth Using OSPF", draft-ietf-ospf- 1039 mpls-elc-05 (work in progress), January 2018. 1041 [I-D.ietf-isis-l2bundles] 1042 Ginsberg, L., Bashandy, A., Filsfils, C., Nanduri, M., and 1043 E. Aries, "Advertising L2 Bundle Member Link Attributes in 1044 IS-IS", draft-ietf-isis-l2bundles-07 (work in progress), 1045 May 2017. 1047 Authors' Addresses 1049 Sriganesh Kini 1051 EMail: sriganeshkini@gmail.com 1053 Kireeti Kompella 1054 Juniper 1056 EMail: kireeti@juniper.net 1058 Siva Sivabalan 1059 Cisco 1061 EMail: msiva@cisco.com 1063 Stephane Litkowski 1064 Orange 1066 EMail: stephane.litkowski@orange.com 1068 Rob Shakir 1069 Google 1071 EMail: rjs@rob.sh 1073 Jeff Tantsura 1075 EMail: jefftant@gmail.com