idnits 2.17.1 draft-chiappa-lisp-architecture-01.txt: Checking boilerplate required by RFC 5378 and the IETF Trust (see https://trustee.ietf.org/license-info): ---------------------------------------------------------------------------- -- The document has an IETF Trust Provisions (28 Dec 2009) Section 6.c(i) Publication Limitation clause. Checking nits according to https://www.ietf.org/id-info/1id-guidelines.txt: ---------------------------------------------------------------------------- == The page length should not exceed 58 lines per page, but there was 1 longer page, the longest (page 1) being 1147 lines 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 (July 16, 2012) is 4299 days in the past. Is this intentional? Checking references for intended status: Informational ---------------------------------------------------------------------------- == Outdated reference: A later version (-04) exists of draft-fuller-lisp-ddt-01 -- No information found for draft-chiappa-lisp-evolution - is the name correct? == Outdated reference: A later version (-01) exists of draft-chiappa-lisp-introduction-00 == Outdated reference: A later version (-29) exists of draft-ietf-lisp-sec-02 -- Obsolete informational reference (is this intentional?): RFC 1631 (Obsoleted by RFC 3022) -- Obsolete informational reference (is this intentional?): RFC 4423 (Obsoleted by RFC 9063) == Outdated reference: A later version (-06) exists of draft-irtf-rrg-ilnp-arch-05 Summary: 0 errors (**), 0 flaws (~~), 6 warnings (==), 5 comments (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 2 LISP Working Group J. N. Chiappa 3 Internet-Draft Yorktown Museum of Asian Art 4 Intended status: Informational July 16, 2012 5 Expires: January 17, 2013 7 An Architectural Perspective on the LISP 8 Location-Identity Separation System 9 draft-chiappa-lisp-architecture-01 11 Abstract 13 LISP upgrades the architecture of the IPvN internetworking system by 14 separating location and identity, current intermingled in IPvN 15 addresses. This is a change which has been identified by the IRTF as 16 a critically necessary evolutionary architectural step for the 17 Internet. In LISP, nodes have both a 'locator' (a name which says 18 _where_ in the network's connectivity structure the node is) and an 19 'identifier' (a name which serves only to provide a persistent handle 20 for the node). A node may have more than one locator, or its locator 21 may change over time (e.g. if the node is mobile), but it keeps the 22 same identifier. 24 This document gives additional architectural insight into LISP, and 25 considers a number of aspects of LISP from a high-level standpoint. 27 [NOTE: This is still a somewhat rough draft version; a few sections 28 at the end are just rough frameworks, but almost all the key 29 sections, and all the front part of the document, are here, and in 30 something like reasonably complete form.] 32 Status of This Memo 34 This Internet-Draft is submitted in full conformance with the 35 provisions of BCP 78 and BCP 79. This document may not be modified, 36 and derivative works of it may not be created, except to format it 37 for publication as an RFC or to translate it into languages other 38 than English. 40 Internet-Drafts are working documents of the Internet Engineering 41 Task Force (IETF). Note that other groups may also distribute 42 working documents as Internet-Drafts. The list of current Internet- 43 Drafts is at http://datatracker.ietf.org/drafts/current/. 45 Internet-Drafts are draft documents valid for a maximum of six months 46 and may be updated, replaced, or obsoleted by other documents at any 47 time. It is inappropriate to use Internet-Drafts as reference 48 material or to cite them other than as "work in progress." 50 This Internet-Draft will expire on January 17, 2013. 52 Copyright Notice 54 Copyright (c) 2012 IETF Trust and the persons identified as the 55 document authors. All rights reserved. 57 This document is subject to BCP 78 and the IETF Trust's Legal 58 Provisions Relating to IETF Documents 59 (http://trustee.ietf.org/license-info) in effect on the date of 60 publication of this document. Please review these documents 61 carefully, as they describe your rights and restrictions with respect 62 to this document. Code Components extracted from this document must 63 include Simplified BSD License text as described in Section 4.e of 64 the Trust Legal Provisions and are provided without warranty as 65 described in the Simplified BSD License. 67 Table of Contents 69 1. Introduction 70 2. Goals of LISP 71 2.1. Reduce DFZ Routing Table Size 72 2.2. Deployment of New Namespaces 73 2.3. Future Development of LISP 74 3. Architectual Perspectives 75 3.1. Another Packet-Switching Layer 76 3.2. 'Double-Ended' Approach 77 4. Architectual Aspects 78 4.1. Critical State 79 4.2. Need for a Mapping System 80 4.3. Piggybacking of Control on User Data 81 5. Namespaces 82 5.1. LISP EIDs 83 5.1.1. Residual Location Functionality in EIDs 84 5.2. RLOCs 85 5.3. Overlapping Uses of Existing Namespaces 86 5.4. LCAFs 87 6. Scalability 88 6.1. Demand Loading of Mappings 89 6.2. Caching of Mappings 90 6.3. Amount of State 91 6.4. Scalability of The Indexing Subsystem 92 7. Security 93 7.1. Basic Philosophy 94 7.2. Design Guidance 95 7.2.1. Security Mechanism Complexity 96 7.3. Security Overview 97 7.3.1. Securing Lookups 98 7.3.2. Securing The Indexing Subsystem 99 7.3.3. Securing Mappings 100 7.4. Securing the xTRs 101 8. Robustness 102 9. Fault Discovery/Handling 103 10. Optimization 104 11. Open Issues 105 11.1. Local Open Issues 106 11.1.1. Missing Mapping Packet Queueing 107 11.1.2. Mapping Cache Management Algorithm 108 11.2. Systemic Open Issues 109 11.2.1. Mapping Database Provider Lock-in 110 11.2.2. Automated ETR Synchronization 111 11.2.3. EID Reachability 112 11.2.4. Detect and Avoid Broken ETRs 113 12. Acknowledgments 114 13. IANA Considerations 115 14. Security Considerations 116 15. References 117 15.1. Normative References 118 15.2. Informative References 119 Appendix A. Glossary/Definition of Terms 120 Appendix B. Other Appendices 122 1. Introduction 124 This document begins by introducing some high-level architectural 125 perspectives which have proven useful for thinking about the LISP 126 location-identity separation system. It then discusses some 127 architectural aspects of LISP (e.g. its namespaces). The balance 128 (and bulk) of the document contains architectural analysis of the 129 LISP system; that is, it reviews from a high-level standpoint various 130 aspects of that system; e.g. its scalability, security, robustness, 131 etc. 133 NOTE: This document assumes a fair degree of familiarity with LISP; 134 in particular, the reader should have a good 'high-level' 135 understanding of the overall LISP system architecture, such as is 136 provided by [Introduction], "An Introduction to the LISP System". 138 By "system architecture" above, the restricted meaning used there is: 139 'How the system is broken up into subsystems, and how those 140 subsystems interact; when does information flows from one to another, 141 and what that information is.' There is obviously somewhat more to 142 architecture (e.g. the namespaces of a system, in particular their 143 syntax and semantics), and that remaining architectural content is 144 covered here. 146 2. Goals of LISP 148 As previously stated in the abstract, broadly, the goal of LISP is to 149 be a practically deployable architectural upgrade to IPvN which 150 performs separation of location and identity. But what is the value 151 of that? What will it allow us to do? 153 The answer to that obviously starts with the things mentioned in the 154 "Initial Applications" section of [Introduction], but there are 155 other, longer-range (and broader) goals as well. 157 2.1. Reduce DFZ Routing Table Size 159 One of the main design drivers for LISP, as well as other location- 160 identity separation proposals, is to decrease the overhead of running 161 global routing system. In fact, it was this aspect that led the IRTF 162 Routing RG to conclude that separation of location and identity was a 163 key architectural underpinning needed to control the growth of the 164 global routing system. [RFC6115] 166 As noted in [Introduction], many of the practical needs of Internet 167 users are today met with techniques that increase the load on the 168 global routing system (Provider Independent addresses for the 169 provision of provider independence, multihoming, etc; more-specific 170 routes for TE; etc.) Provision of these capabilities by a mechanism 171 which does not involve extra load on the global routing system is 172 therefore very desirable. 174 A number of factors, including the use of these techniques, has led 175 to a great increase in the fragmentation of the address space, at 176 least in terms of routing table entries. In particular, the growth 177 in demand for multi-homing has been forseen as driving a large 178 increase in the size of the global routing tables. 180 In addition, as the IPv4 address space becomes fuller and fuller, 181 there will be an inevitable tendency to find use in smaller and 182 smaller 'chunks' of that space. [RFC6127] This too would tend to 183 increase the size of the global routing table. 185 LISP, if successful and widely deployed, offers an opportunity to use 186 separation of location and identity to control the growth of the size 187 of the global routing table. (A full examination of this topic is 188 beyond the scope of this document - see {{find reference}}.) 190 2.2. Deployment of New Namespaces 192 Once the mapping system is widely deployed and available, it should 193 make deployment of new namespaces (in the sense of new syntax, if not 194 new semantics) easier. E.g. if someone wishes in the future to 195 devise a system which uses native MPLS [RFC3031] for a data carriage 196 system joining together a large number of xTRs, it would easy enough 197 to arrange to have the mappings for destinations attached to those 198 xTRs abe some sort of MPLS-specific name. 200 More broadly, the existence of a binding layer, with support for 201 multiple namespace built into the interface on both sides (see 202 Section 5) is a tremendously powerful evolutionary tool; one can 203 introduce a new namespace (on one side) more easily, if it is mapped 204 to something which is already deployed (on the other). Then, having 205 taken that step, one can invert the process, and deploy yet another 206 new namespace, but this time on the other. 208 2.3. Future Development of LISP 210 Speculation about long-term future developments which are enabled by 211 the deployment of LISP is not really proper for this document. 212 However, interested readers may wish to consult [Future] for one 213 person's thoughts on this topic. 215 3. Architectual Perspectives 217 This section contains some high-level architectural perspectives 218 which have proven useful in a number of ways for thinking about LISP. 219 For one, when trying to think of LISP as a complete system, they 220 provide a conceptual structure which can aid analysis of LISP. For 221 another, they can allow the application of past analysis of, and 222 experience with, similar designs. 224 3.1. Another Packet-Switching Layer 226 When considering the overall structure of the LISP system at a high 227 level, it has proven most useful to think of it as another packet- 228 switching layer, run on top of the original internet layer - much as 229 the Internet first ran on top of the ARPANET. 231 All the functions that a normal packet switch has to undertake - such 232 as ensuring that it can reach its neighbours, and they they are still 233 up - the devices that make up the LISP overlay also have to do, along 234 the 'tunnels' which connect them to other LISP devices. 236 There is, however, one big difference: the fanout of a typical LISP 237 ITR will be much larger than most classic physical packet switches. 238 (ITRs only need to be considered, as the LISP tunnels are all 239 effectively unidirectional, from ITR to ETR - an ETR needs to keep no 240 per-tunnel state, etc.) 242 LISP is, fundamentally, a 'tunnel' based system. Tunnel system 243 designs do have their issues (e.g. the high inter-'switch' fan-out), 244 but it's important to realize that they also can have advantages, 245 some of which are listed below. 247 3.2. 'Double-Ended' Approach 249 LISP may be thought of as a 'double-ended' approach to enhancing the 250 architecture, in that it uses pairs of devices, one at each end of a 251 communication stream. In particular, to interact with the population 252 of 'legacy' hosts (which will be, inevitably, the vast majority, in 253 the early stages of deployment) it requires a LISP device at both 254 ends of the 'tunnel'. 256 This is in distinction to, say, NAT systems ([RFC1631]), which only 257 need a device deployed at one end: the host at the other end doesn't 258 need a matching device at its end to massage the packets, but can 259 simply consume them on its own, as any packets it receives are fully 260 normal packets. This allows any site which deploys such a 'single- 261 ended' device to get the full benefit, whilst acting entirely on its 262 own. [Wasserman] 264 The issue is not that LISP uses tunnels. Designs like HIP 265 ([RFC4423]) and ILNP ([ILNP]), which do not involve tunnels, inhabit 266 a similar space to tunnel-based designs like LISP, in that unless 267 both ends are upgraded - or there is a proxy at the un-upgraded end - 268 one doen't get any benefits. So it's really not the tunnel which is 269 the key aspect, it's the 'all at one end' part which is key. Whether 270 the system is tunnel, versus non-tunnel, is not that important. 272 However, the double-ended approach of LISP does have advantages, as 273 well as costs. To put it simply, the 'feature' of the alternative 274 approach, that there's only a box at one end, has a 'bug': there's 275 only a box at one end. There are things which such a design cannot 276 accomplish, because of that. 278 To put it another way, does the fact that the packet thus necessarily 279 has only a single 'name' in it for the entities at each end (i.e. the 280 IPvN source and destination addresses), because it is a 'normal' 281 packet, present a limitation? Put that way, it would seem natural 282 that it should cause certain limits. 284 To compile a complete list of the things that can be done, when two 285 separate 'names' are in the packet, is beyond the scope of this 286 document. However, one example of the kind of thing that can be done 287 is mobility with open connections, without needing to 'triangle 288 route' the packets through some sort of 'base station' at the 289 original location. Another is that is possible to automatically 290 tunnel IPv6 traffic over IPv4 infrastructure, or vice versa, 291 invisibly to the hosts on both ends. 293 In the longer term, having having tunnel boxes will allow (and is 294 allowing) us to explore other kinds of wrappings. For example, we 295 can transport 'raw' local-network packets (such as Ethernet MAC 296 frames) across an IPvN infrastructure. 298 One could also wrap packets in non-IPvN formats: perhaps to take 299 direct advantage of the capabilities of underlying switching fabrics 300 (e.g. MPLS [RFC3031]); perhaps to deploy new carriage protocols, 301 etc, where non-standard packet formats will allow extended semantics. 303 4. Architectual Aspects 305 LISP does take some novel architectural approaches in a number of 306 ways: e.g. its use of a separate mapping system, etc, etc. This 307 section contains some commentary on some of the high-level 308 architectural aspects of LISP. 310 4.1. Critical State 312 LISP does have 'critical state' in the network (i.e. state which, if 313 if lost, causes the communication to fail). However, because LISP is 314 designed as an overall system, 'designing it in' allows for a 315 'systems' approach to its state issues. In LISP, this state has been 316 designed to be maintained in an 'architected' way, so it does not 317 produce systemic brittleness in the way that the state in NATs does. 319 For instance, throughout the system, provisions have been made to 320 have redundant copies of state, in multiple devices, so that the loss 321 of any one device does not necessarily cause a failure of an ongoing 322 connection. 324 4.2. Need for a Mapping System 326 LISP does need to have a mapping system, which brings design, 327 implementation, configuration and operational costs. Surely all 328 these costs are a bad thing? However, having a mapping system have 329 advantages, especially when there is a mapping layer which has global 330 visibility (i.e. other entities know that it is there, and have an 331 interface designed to be able to interact with it). This is unlike, 332 say, the mappings in NAT, which are 'invisible' to the rest of the 333 network. 335 In fact, one could argue that the mapping layer is LISP's greatest 336 strength. Wheeler's Axiom* ('Any problem in computer science can be 337 solved with another level of indirection') indicates that the binding 338 layer available with the LISP mapping system will be of great value. 339 Again, it is not the job of this document to list them all - and in 340 any event, there is no way to forsee them all. 342 The author of this document has often opined that the hallmark of 343 great architecture is not how well it does the things it was designed 344 to do, but how well it does things it was never expected to have to 345 handle. Providing such a powerful and generic binding layer is one 346 sure way to achieve the sort of lasting flexibility and power that 347 leads to that outcome. 349 [Footnote *: This Axiom is often mis-attributed to Butler Lampson, 350 but Lampson himself indicated that it came from David Wheeler.] 352 4.3. Piggybacking of Control on User Data 354 LISP piggybacks control transactions on top of user data packets. 355 This is a technique that has a long history in data networking, going 356 back to the early ARPANET. [McQuillan] It is now apparently regarded 357 as a somewhat dubious technique, the feeling seemingly being that 358 control and user data should be strictly segregated. 360 It should be noted that _none_ of the piggybacking of control 361 functionality in LISP is _architecturally fundamental_ to LISP. All 362 of the functions in LISP which are performed with piggybacking could 363 be performed almost equally well with separate control packets. 365 The "almost" is solely because it would cause more overhead (i.e. 366 control packets); neither the response time, robustness, etc would 367 necessarily be affected - although for some functions, to match the 368 response time observed using piggybacking on user data would need as 369 much control traffic as user data traffic. 371 This technique is particularly important, however, because of the 372 issue identified at the start of this section - the very large fanout 373 of the typical LISP switch. Unlike a typical router, which will have 374 control interactions with only a few neighbours, a LISP switch could 375 eventually have control interactions with hundreds, or perhaps even 376 thousands (for a large site) of neighbours. 378 Explicit control traffic, especially if good response times are 379 desired, could amount to a very great deal of overhead in such a 380 case. 382 5. Namespaces 384 One of the key elements in any architecture, or architectural 385 analysis, are the namespaces involved: what are their semantics and 386 syntax, what are the kinds of things they name, etc. 388 LISP has two key namespace, EIDs and RLOCs, but it must be emphasized 389 that on an architectural level, neither the syntax, or, to a lesser 390 degree, the semantics, of either are absolutely fixed. There are 391 certain core semantics which are generaly unchanging (such as the 392 notion that EIDs provide only identity, whereas RLOCs provide 393 location), but as we will see, there is a certain amount of 394 flexibility available for the long-term. 396 In particular, all of LISP's key interfaces always include an Address 397 Family Identifier (AFI) [AFI] for all names, so that new forms can be 398 introduced at any time the need is felt. Of course, in practise such 399 an introduction would not be a trivial exercise - but neither is is 400 impossibly painful, as is the case with IPv4's 32-bit addresses, 401 which are effectively impossible to upgrade. 403 5.1. LISP EIDs 405 A 'classic' EID is defined as a subset of the possible namespaces for 406 endpoints. [Chiappa] Like most 'proper' endpoint names, as proposed 407 there, they contain contain no information about the location of the 408 endpoint. EIDs are the subset of possible endpoint names which are: 409 fixed length, 'reasonably' short', binary (i.e. not intended for 410 direct human use), globally unique (in theory), and allocated in a 411 top-down fashion (to achieve the former). 413 LISP EIDs are, in line with the general LISP deployment philosophy, a 414 reuse of something already existing - i.e. IPvN addresses. For 415 those used as in LISP as EIDs, LISP removes much (or, in some cases, 416 all) of the location-naming function of IPvN addresses. 418 In addition, the goal is to have EIDs name hosts (or, more properly, 419 their end-end communication stacks), whereas the other LISP namespace 420 group (RLOCs) names interfaces. The idea is not just to have two 421 namespaces (with different semantics), but also to use them to name 422 _different classes of things_ - classes which currently do not have 423 clearly differentiated names. This should produce even more 424 functionality. 426 5.1.1. Residual Location Functionality in EIDs 428 LISP retains, especially in the early stages of the deployment, in 429 many cases some residual location-naming functionality in EIDs, This 430 is to allow the packet to be correctly routed/forwarded to the 431 destination node, once it has been unwrapped by the ETR - and this is 432 a direct result of LISP's deployment philosophy (see [Introduction], 433 Section "Deployment"). 435 Clearly, if there are one or more unmodified routers between the ETR 436 and the desination node, those routers will have to perform a routing 437 step on the packet, for which it will need _some_ information as to 438 the location of the destination. 440 One can thus view such LISP EIDs, which retain 'stub' location 441 information, as 'addresses' (in the definition of the generic sense 442 of this term, as used here), but with the location information 443 restricted to a limited, local scope. 445 This retention of some location functionality in LISP EIDs, in some 446 cases, has led some people to argue that use of the name 'EID' is 447 improper. In response, it was suggested that LISP use the term 448 'LEID', to distinguish LISP's 'bastardized' EIDs from 'true' EIDs, 449 but this usage has never caught on. 451 It has also been suggested that one usage mode for LISP EIDs, in 452 existing software loads, is to assign them as the address on an 453 internal virtual interface; all the real interfaces would have RLOCs 454 only. [Templin] This would make such LISP EIDs functionally 455 equivalent to 'real' EIDs - they are names which are purely identity, 456 have no location information of any kind in them, and cannot be used 457 to make any routing decisions anywhere outside the host. 459 It is true that even in such cases, the EID is still not a 'pure' 460 EID, as it names an interface, not the end-end stack directly. 461 However, to do a perfect job here (or on separation of location and 462 identity) is impossible without modifying existing hosts (which are, 463 inevitably, almost always one end of an end-end communication) - and 464 that has been ruled out, for reasons of viable deployment. 466 The need for interoperation with existing unmodified hosts limits the 467 semantic changes one can impose, much as one might like to provide a 468 cleaner separation. (Future evolution can bring us toward that 469 state, however: see [Future].) 471 5.2. RLOCs 473 RLOCs are basically pure 'locators' [RFC1992], although their syntax 474 and semantics is restricted at the moment, because in practise the 475 only forms of RLOCs supported are IPv4 and IPv6. 477 5.3. Overlapping Uses of Existing Namespaces 479 It is in theory possible to have a block of IPvN namespace used as 480 both EIDs and RLOCs. In other words, EIDs from that block might map 481 to some other RLOCs, and that block might also appear in the DFZ as 482 the locators of some other ETRs. 484 This is obviously potentially confusing - when a 'bare' IPvN address 485 from one of these blocks, is it the RLOC, or the EID? Sometimes it 486 it obvious from the context, but in general one could not simply have 487 a (hypothetical) table which assigns all of the address space to 488 either 'EID' or 'RLOC'. 490 In addition, such usage will not allow interoperation of the sites 491 named by those EIDs with legacy sites, using the PITR mechanism 492 ([Introduction], Section "Proxy Devices"), since that mechanisms 493 depends on advertizing the EIDs into the DFZ, although the LISP-NAT 494 mechanism should still work ([Introduction], Section "LISP-NAT"). 496 Nevertheless, as the IPv4 namespace becomes increasingly used up, 497 this may be an increasingly attractive way of getting the 'absolute 498 last drop' out of that space. 500 5.4. LCAFs 502 {{To be written.}} 504 --- Key-ID 505 --- Instance-IDs 507 6. Scalability 509 As with robustness, any global communication system must be scalable, 510 and scalable up to almost any size. As previously mentioned (xref 511 target="Perspectives-Packet"/), the large fanouts to be seen with 512 LISP, due to its 'overlay' nature, present a special challenge. 514 One likely saving grace is that as the Internet grows, most sites 515 will likely only interact with a limited subset of the Internet; if 516 nothing else, the separation of the world into language blocks means 517 that content in, say, Chinese, will not be of interest to most of the 518 rest of the world. This tendency will help with a lot of things 519 which could be problematic if constant, full, N^2 connectivity were 520 likely on all nodes; for example the caching of mappings. 522 6.1. Demand Loading of Mappings 524 One question that many will have about LISP's design is 'why demand- 525 load mappings - why not just load them all'? It is certainly true 526 that with the growth of memory sizes, the size of the complete 527 database is such that one could reasonably propose keeping the entire 528 thing in each LISP device. (In fact, one proposed mapping system for 529 LISP, named NERD, did just that. [NERD]) 531 A 'pull'-based system was chosen over 'push' for several reasons; the 532 main one being that the issue is not just the pure _size_ of the 533 mapping database, but its _dynamicity_. Depending on how often 534 mappings change, the update rate of a complete database could be 535 relatively large. 537 It is especially important to realize that, depending on what 538 (probably unforseeable) uses eventually evolve for the 539 identity->location mapping capability LISP provides, the update rate 540 could be very high indeed. E.g. if LISP is used for mobility, that 541 will greatly increase the update rate. Such a powerful and flexible 542 tool is likely be used in unforseen ways (Section 4.2), so it's 543 unwise to make a choice that would preclude any which raise the 544 update rate significantly. 546 Push as a mechanism is also fundamentally less desirable than pull, 547 since the control plane overhead consumed to load and maintain 548 information about unused destinations is entirely wasted. The only 549 potential downside to the pull option is the delay required for the 550 demand-loading of information. 552 (It's also probably worth noting that many issues that some people 553 have with the mapping approach of LISP, such as the total mapping 554 database size, etc are the same - if not worse - for push as they are 555 for pull.) 557 Finally, for IPv4, as the address space becomes more highly used, it 558 will become more fragmented - i.e. there will tend to be more, 559 smaller, entries. For a routing table, which every router has to 560 hold, this is problematic. For a demand-loaded mapping table, it is 561 not bad. Indeed, this was the original motivation for LISP 562 ([RFC4984]) - although many other useful and desirable uses for it 563 have since been enumerated (see [Introduction], Section 564 "Applications"). 566 For all of these reasons, as long as there is locality of reference 567 (i.e. most ITRs will use only a subset of the entire set), it makes 568 much more sense to use the a pull model, than the classic push one 569 heretofore seen widely at the internetwork layer (with a pull 570 approach thus being somewhat novel - and thus unsettling to many - to 571 people who work at that layer). 573 It may well be that some sites (e.g. large content providers) may 574 need non-standard mechanisms - perhaps something more of a 'push' 575 model. This remains to be determined, but it is certainly feasible. 577 6.2. Caching of Mappings 579 It should be noted that the caching spoken of here is likely not 580 classic caching, where there is a fixed/limited size cache, and 581 entries have to be discarded to make room for newly needed entries. 582 The economics of memory being what they are, there is no reason to 583 discard mappings once they have been loaded (although of course 584 implementations are free to chose to do so, if they wish to). 586 This leads to another point about the caching of mappings: the 587 algorithms for management of the cache are purely a local issue. The 588 algorithm in any particular ITR can be changed at will, with no need 589 for any coordination. A change might be for purposes of 590 experimentation, or for upgrade, or even because of environmental 591 variations - different environments might call for different cache 592 management strategies. 594 The local, unsynchronized replacability of the cache management 595 scheme is the architectural aspect of the design; the exact 596 algorithm, which is engineering, is not. 598 6.3. Amount of State 600 {{To be written.}} [Iannone] 602 -- Mapping cache size 603 --- Mention studies 604 -- Delegation cache size (in MRs) 605 --- Mention studies 606 -- Any others? 608 6.4. Scalability of The Indexing Subsystem 610 LISP initially used an indexing subsystem called ALT. [ALT] ALT was 611 relatively easy to construct from existing tools (GRE, BGP, etc), but 612 it had a number of issues that made it unsuitable for large-scale 613 use. ALT is now being superseded by DDT. [DDT] 615 The basic structure and operation of DDT is identical to that of 616 TREE, so the extensive simulation work done for TREE applies equally 617 to DDT, as do the conclusions drawn about TREE's superiority to ALT. 618 [Jakab] 620 From an architectural point of view, the main advantage of DDT is 621 that it enables client side caching of information about intermediate 622 nodes in the resolution hierarchy, and also enables direct 623 communication with them. As a result, DDT has much better scaling 624 properties than ALT. 626 The most important result of this change is that it avoids a 627 concentration of resolution request traffic at the root of the 628 indexing tree, a problem which by itself made ALT unsuitable for a 629 global-scale system. The problem of root concentration (and thus 630 overload) is almost unavoidable in ALT (even if masses of 'bypass' 631 links are created). 633 ALT's scalability also depends on enforcing an intelligent 634 organization that aincreases aggregation. Unfortunately, the current 635 backbone routing BGP system shows that there is a risk of an organic 636 growth of ALT, one which does not achieve aggregation. DDT does not 637 display this weakness, since its organization is inherently 638 hierarchical (and thus inherently aggregable). 640 The hierarchical organization of DDT also reduces the possibility for 641 a configuration error which interferes with the operation of the 642 network (unlike the situation with the current BGP DFZ). DDT 643 security mechanisms can also help produce a high degree of 644 robustness, both against misconfiguration, and deliberate attack. 645 The direct communication with intermediate nodes in DDT also helps to 646 quickly locate problems when they occur, resulting in better 647 operational characteristics. 649 Next, since in ALT mapping requests must be transmitted through an 650 overlay network, a significant share of requests can see 651 substantially increased latencies. Simulation results in the TREE 652 work clearly showed, and quantified, this effect. 654 The simulations also showed that the nodes composing the ALT and DDT 655 networks for a mapping database of full Internet size could have 656 thousands of neighbours. This is not an issue for DDT, but would 657 almost certainly have been problematic for ALT nodes, since handling 658 that number of simultaneous BGP sessions would likely to be 659 difficult. 661 7. Security 663 LISP does not yet have an overarching security architecture. Many 664 parts of the system have been hardened, but more on a case-by case 665 basis, rather than from an overall perspective. (This is in part due 666 to the 'just enough' approach to security initially taken in LISP; 667 see [Introduction], Section "Just Enough Security".) 669 This section represents an attempt to produce a more broadly-based 670 view of security in LISP; it mostly resulted from an attempt to add 671 security to the DDT indexing system ([DDT]), but the analysis is is 672 general enough to apply to LISP broadly. 674 The _good_ thing about the Internet is that it brings the world to 675 your doorstep - masses of information from all around the world are 676 instantly available on your computing device. The _bad_ thing about 677 the Internet is that it brings the world to your doorstep - including 678 legions of crackers, thieves, and general scum and villainy. Thus, 679 any node may be the target of fairly sophisticated attack - often 680 automated (thereby reducing the effort required of the attacker to 681 spread their attack as broadly as possible). 683 Security in LISP faces many of the same challenges as security for 684 other parts of the Internet: good security usually means work for the 685 users, but without good security, things are vulnerable. 687 The Internet has seen many very secure systems devised, only to see 688 them fail to reach wide adoption; the reasons for that are complex, 689 and vary, but being too much work to use is a common thread. It is 690 for this reason that LISP attempts to provide 'just enough' security 691 (see [Introduction], Section "Just Enough Security"). 693 7.1. Basic Philosophy 695 To square this circle, of needing to have very good security, but of 696 it being too difficult to use very good security, the general concept 697 is for LISP to have a series of 'graded' security measures available, 698 with the 'ultimate' security mechanisms being very high-grade indeed. 700 The concept is to devise a plan in which LISP can simultaneously 701 attempt to have not just 'ultimate' security, but also one or more 702 'easier' modes, ones which will be easier to configure and use. This 703 'easier' mode can be both an interim system (with the full powered 704 system available for when it it needed), as well as the system used 705 in sections of the network where security is less critical (following 706 the general rule that the level of any security should generally be 707 matched to what is being protected). 709 The challenge is to do this in a way that does not make the design 710 more complex, since it has to include both the 'full strength' 711 mechanism(s), and the 'easier to configure' mechanism(s). This is 712 one of the fundamental tradeoffs to struggle with: it is easy to 713 provide 'easier to configure' options, but that may make the overall 714 design more complex. 716 As far as making it hard to implement to begin with (also something 717 of a concern initially, although obviously not for the long term): we 718 can make it 'easy' to deploy initially by simply not implementing/ 719 configuring the heavy-duty security early on. (Provided, of course, 720 that the packet formats, etc, needed to support such security are all 721 included in the design to begin with.) 723 7.2. Design Guidance 725 In designing the security, there are a small number of key points 726 that will guide the design: 728 - Design lifetime 729 - Threat level 731 How long is the design intended to last? If LISP is successful, a 732 minimum of a 50-year lifetime is quite possible. (For comparison, 733 IPv4 is now 34 at the time of writing this, and will be around for at 734 least several decades yet, if not longer; DNS is 28, and will 735 probably last indefinitely.) 737 How serious are the threats it needs to meet? As mentioned above, 738 the Internet can bring the worst crackers from anywhere to any 739 location, in a flash. Their sophistication level is rising all the 740 time: as the easier holes are plugged, they go after others. This 741 will inevitably eventually require the most powerful security 742 mechanisms available to counteract their attacks. 744 Which is not to say that LISP needs to be that secure _right away_. 745 The threat will develop and grow over a long time period. However, 746 the basic design has to be capable of being _securable_ to the 747 expanded degree that will eventually be necessary. However, 748 _eventually_ it will need to be as securable as, say, DNS - i.e. it 749 _can_ be secured to the same level, although people may chose not to 750 secure their LISP infrastructure as well as DNSSEC potentially does. 751 [RFC4033] 753 In particular, it should be noted that historically many systems have 754 been broken into, not through a weakness in the algorithms, etc, but 755 because of poor operational mechanics. (The well-known 'Ultra' 756 breakins of the Allies were mostly due to failures in operational 757 procedure. [Welchman]) So operational capabilities intended to 758 reduce the chance of human operational failure are just as important 759 as strong algorithms; making things operationally robust is a key 760 part of 'real' security. 762 7.2.1. Security Mechanism Complexity 764 Complexity is bad for several reasons, and should always be reduced 765 to a minimum. There are three kinds of complexity cost: protocol 766 complexity, implementation complexity, and configuration complexity. 767 We can further subdivide protocol complexity into packet format 768 complexity, and algorithm complexity. (There is some overlap of 769 algorithm complexity, and implementation complexity.) 771 We can, within some limits, trade off one kind of complexity for 772 others: e.g. we can provide configuration _options_ which are simpler 773 for the users to operate, at the cost of making the protocol and 774 implementation complexity greater. And we can make initial (less 775 capable) implementations simpler if we make the protocols slightly 776 more complex (so that early implementations don't have to implement 777 all the features of the full-blown protocol). 779 It's more of a question of some operational convenience/etc issues - 780 e.g. 'How easy will it be to recover from a cryptosystem 781 compromise'. If we have two ways to recover from a security 782 compromise, one which is mostly manual and a lot of work, and another 783 which is more automated but makes the protocol more complicated, if 784 compromises really are very rare, maybe the smart call _is_ to go 785 with the manual thing - as long as we have looked carefully at both 786 options, and understood in some detail the costs and benefits of 787 each. 789 7.3. Security Overview 791 First, there are two different classes of attack to be considered: 792 denial of service (DoS, i.e. the ability of an intruder to simply 793 cause traffic not to successfully flow) versus exploitation (i.e. the 794 ability to cause traffic to be 'highjacked', i.e. traffic to be sent 795 to the wrong location). 797 Second, one needs to look at all the places that may be attacked. 798 Again, LISP is a relatively simple system, so there are not that many 799 parts to examine. The following are the things we need to secure: 801 - Lookups 802 - Indexing 803 - Mappings 805 7.3.1. Securing Lookups 807 {{To be written.}} Nonces, [SecurityReq] 809 7.3.2. Securing The Indexing Subsystem 811 It is envisioned that DDT will be highly securable, with all the 812 delegations cryptographiclly secured via public-private signatures, 813 very similar to the way DNS is ([RFC4033]). 815 The detailed mechanisms will be based on DNS's; this has the obvious 816 benefit that all the lessons of DNS's years of practical experience 817 with deployment, operations, etc, as well as the improvements to the 818 basic design of DNS Security to provide a secure but usable system 819 can be taken into account. However, DDT's security will also apply 820 the thinking above, about making a 'versio' which is easier to use 821 available. 823 {{To be written.}} 825 7.3.3. Securing Mappings 827 There are two approaches to securing the provision of mappings. The 828 first, which is of course not completely satisfactory, is to only 829 secure the channel between the ITR and the entities involved in 830 providing mappings for it. (See above, Section 7.3.1) 832 The second is to secure the mappings themselves, by signing them 'at 833 birth' (much the same way in which DNS Security operates). 834 [RFC4033]. There was an attempt early on to suggest such a system 835 for LISP ([SecurityAuth]), but it was not adopted (although the 836 particular proposal was rather complex). 838 In the long run, the latter approach would obviously be superior, 839 since it would be almost immune to any compromises of the mapping 840 distribution system. {{Tie-in to space allocation security}} 842 7.4. Securing the xTRs 844 --- Cache management 845 --- Unsoliticed Map-Replies are _very bad_ - must go through 846 mapping system to verify that the sender is authoritative for 847 that range of EIDs 849 8. Robustness 851 -- Depends on deployment as well as design 852 -- Architected, visible replication of state/data 853 -- Overlapping mechanisms (ref redundancy as key for robustness) 855 9. Fault Discovery/Handling 857 Any global communication system must be robust, and to be robust, it 858 must be able to discover and handle problems. LISP's general 859 philosophy of robustness is usually to have overlapping, simple 860 mechanisms to discover and repair problems. 862 10. Optimization 864 -- Philosophy 865 -- Piggybacking 866 -- 'Wiretapping' return mappings 867 --- Security is an issue on that 869 11. Open Issues 871 Although much work has been done on LISP, and it operates 872 satisfactorily in a reasonably large initial deployment, there are a 873 few potentially problematic issues which remain. It is not clear if 874 they will be issues which need to be dealt, since they have not 875 proven to be obstacles so far, but it is worth listing them. 877 We can divide them in _local_ issues, i.e. ones which can be solved 878 on a node-by-node basis, without requiring co-ordinated change, and 879 systemic issues, which are obviously more problematic, since they 880 could require co-ordinated changes to the protocols. 882 11.1. Local Open Issues 884 11.1.1. Missing Mapping Packet Queueing 886 Currently, some (all?) ITRs discard packets when they need a 887 mapping, but have not loaded one yet, thereby causing the applicaton 888 to have to retransmit their opening packet. True, many ARP 889 implementations use the same strategy, but the average APR cache will 890 only ever contain a few mappings, so it will not be so noticeable as 891 with the mapping cache in an ITR, which will likely contain 892 thousands. 894 Obviously, they could queue the packets while waiting to load the 895 mapping, but this presents a number of subtle implementation issues: 896 the ITR must make sure that it does not queue too many packets, etc. 898 In particular, if such packets are queued, this presents a potential 899 DoS attack vector, unless the code is carefully written with that 900 possibility in mind. 902 11.1.2. Mapping Cache Management Algorithm 904 Relatively little work has been done on sophisticated mapping cache 905 management algorithms; in particular, the issue of which mapping(s) 906 to drop if the cache reaches some maximum allowed size. 908 This particular issue has also been identified as another potential 909 DoS attack vector. 911 11.2. Systemic Open Issues 913 11.2.1. Mapping Database Provider Lock-in 915 This refers to the fact that if one does not like the entity which is 916 providing the indexing for the part of the address space which one's 917 EIDs are allocated out of, there isn't probably isn't any way to 918 switch to an alternative provider. 920 It is not clear that this is a real probem, though - the fact that 921 all DNS top-level zones only have a single registry has not been a 922 problem, nor has the fact that if one doesn't like the service the 923 registry offers, one can't take one's DNS name to another registry. 925 Doing anything about it would also be difficult. Although it is 926 _technically_ possible to duplicate any node in the delegation tree, 927 and in theory such duplicates could be provided by different 928 providers, it is not clear that such an arrangement would make 929 _business_ sense. 931 For instance, if the holder of 10.1.1/24 decides they do not like the 932 entity providing indexing for 10.1/16 (call them E1), and ask another 933 entity (E2) to provide alternative service for 10.1/16, two problems 934 arise. First, E1 is _still_ going to have to maintain the correct 935 data for 10.1.1/24, and response to queries asking about them. 936 Second, E2 will similarly have to maintain data for, and reply to 937 queries about, all the other space-holders in 10.1/16 - even though 938 they will likely not have any business relationship with them. 940 11.2.2. Automated ETR Synchronization 942 LISP requires that all the ETRs which are authoritative for the 943 mappings for a particular address block return the same mapping data. 944 In particular, their idea of the 'liveness' of all the ETRs should be 945 identical, and correct. 947 At the moment, this is mostly a manual process, although liveness 948 information can be currently be gathered from some IGPs. 950 11.2.3. EID Reachability 952 At the moment, LISP assumes that if an ETR is reachable from a given 953 ITR, all destination EIDs behind that ETR are reachable from that 954 ETR. There is no way to detect if any are not, nor to switch to an 955 alternate ETR. 957 It is not clear that this is a problem that needs attention. The 958 same has been true for all border routers for many years now, and 959 there does not seem to be any general mechanism to deal with it 960 (Although some BGP implementations may advertize changes in 961 reachability status if what they are seeing from their IGP changes.) 963 11.2.4. Detect and Avoid Broken ETRs 965 {{To be written}} 967 12. Acknowledgments 969 The author would like thank all the members of the core LISP group 970 for their willingness to allow him to add himself to their effort, 971 and for their enthusiasm for whatever assistance he has been able to 972 provide. He would also like to thank (in alphabetical order) Vina 973 Ermagan, Vince Fuller, and Joel Halpern for their careful review of, 974 and helpful suggestions for, this document. Grateful thanks also to 975 Vince Fuller for help with XML. 977 A final thanks is due to John Wrocklawski for the author's 978 organizational affiliation. This memo was created using the xml2rfc 979 tool 981 13. IANA Considerations 983 This document makes no request of the IANA. 985 14. Security Considerations 987 This memo does not define any protocol and therefore creates no new 988 security issues. 990 15. References 992 15.1. Normative References 994 [DDT] V. Fuller, D. Lewis, and D. Farinacci, "LISP 995 Delegated Database Tree", draft-fuller-lisp-ddt-01 996 (work in progress), March 2012. 998 [Future] J. N. Chiappa, "Potential Long-Term Developments With 999 the LISP System", draft-chiappa-lisp-evolution-00 1000 (work in progress), July 2012. 1002 [Introduction] J. N. Chiappa, "An Introduction to the LISP Location- 1003 Identity Separation System", 1004 draft-chiappa-lisp-introduction-00 (work in 1005 progress), July 2012. 1007 [SecurityAuth] R. Gagliano, "A Profile for Endpoint Identifier 1008 Origin Authorizations (IOA)", 1009 draft-rgaglian-lisp-iao-00 (work in progress), 1010 March 2009. 1012 [SecurityReq] F. Maino, V. Ermagan, A. Cabellos, D. Saucez, and 1013 O. Bonaventure, "LISP-Security (LISP-SEC)", 1014 draft-ietf-lisp-sec-02 (work in progress), 1015 March 2012. 1017 [AFI] IANA, "Address Family Indicators (AFIs)", Address 1018 Family Numbers, January 2011, . 1021 15.2. Informative References 1023 [RFC1631] K. Egevang and P. Francis, "The IP Network Address 1024 Translator (NAT)", RFC 1631, May 1994. 1026 [RFC1992] I. Castineyra, J. N. Chiappa, and M. Steenstrup, "The 1027 Nimrod Routing Architecture", RFC 1992, August 1996. 1029 [RFC3031] E. Rosen, A. Viswanathan, and R. Callon, 1030 "Multiprotocol Label Switching Architecture", 1031 RFC 3031, January 2001. 1033 [RFC4033] R. Arends, R. Austein, M. Larson, D. Massey, and 1034 S. Rose, "DNS Security: Introduction and 1035 Requirements", RFC 4033, March 2005. 1037 [RFC4423] R. Moskowitz and P. Nikander, "Host Identity Protocol 1038 (HIP) Architecture", RFC 4423, May 2006. 1040 [RFC4984] D. Meyer, L. Zhang, and K. Fall, "Report from the IAB 1041 Workshop on Routing and Addressing", RFC 4984, 1042 September 2007. 1044 [RFC6115] T. Li, Ed., "Recommendation for a Routing 1045 Architecture", RFC 6115, February 2011. 1047 Perhaps the most ill-named RFC of all time; it 1048 contains nothing that could truly be called a 1049 'routing architecture'. 1051 [RFC6127] J. Arkko and M. Townsley, "IPv4 Run-Out and IPv4-IPv6 1052 Co-Existence Scenarios", RFC 6127, May 2011. 1054 [ALT] D. Farinacci, V. Fuller, D. Meyer, and D. Lewis, 1055 "LISP Alternative Topology (LISP-ALT)", 1056 draft-ietf-lisp-alt-10 (work in progress), 1057 December 2011. 1059 [NERD] E. Lear, "NERD: A Not-so-novel EID to RLOC Database", 1060 draft-lear-lisp-nerd-09 (work in progress), 1061 April 2012. 1063 [ILNP] R.J. Atkinson and S.N. Bhatti, "ILNP Architectural 1064 Description", draft-irtf-rrg-ilnp-arch-05 (work in 1065 progress), May 2012. 1067 [Chiappa] J. N. Chiappa, "Endpoints and Endpoint Names: A 1068 Proposed Enhancement to the Internet Architecture", 1069 Personal draft (work in progress), 1999, 1070 . 1072 [Jakab] L. Jakab, A. Cabellos-Aparicio, F. Coras, D. Saucez, 1073 and O. Bonaventure, "LISP-TREE: A DNS Hierarchy to 1074 Support the LISP Mapping System", in 'IEEE Journal on 1075 Selected Areas in Communications', Vol. 28, No. 8, 1076 pp. 1332-1343, October 2010. 1078 [Iannone] L. Iannone and O. Bonaventure, "On the Cost of 1079 Caching Locator/ID Mappings", in 'Proceedings of the 1080 3rd International Conference on emerging Networking 1081 EXperiments and Technologies (CoNEXT'07)', ACM, pp. 1082 1-12, December 2007. 1084 [McQuillan] J. M. McQuillan, W. R. Crowther, B. P. Cosell, 1085 D. C. Walden, and F. E. Heart, "Improvements in the 1086 Design and Performance of the ARPA Network", 1087 Proceedings AFIPS 1972 FJCC, Vol. 40, pp. 741-754. 1089 [Templin] F. Templin, "LISP WG", LISP WG list 1090 message, Message-ID: 39C363776A4E8C4A94691D2BD9D1C9A1 1091 05B0AC71@XCH-NW-7V2.nw.nos.boeing.com, 13 1092 March 2009,, . 1095 [Wasserman] M. Wasserman, "IPv6 networking: Bad news for small 1096 biz", IETF list message, Message-Id: 1097 D11C4A34-7362-423E-A60E-476FC5D61D37@lilacglade.org, 1098 5 April 2012, . 1102 [Welchman] G. Welchman, "The Hut Six Story", Allen Lane, 1103 London, pg. 3, 1982. 1105 A truly monumental book; the ground it covers ranges 1106 from his work helping break German codes in World War 1107 II to his experience with securing data packet 1108 networks! 1110 Appendix A. Glossary/Definition of Terms 1112 - Address 1113 - Locator 1114 - EID 1115 - RLOC 1116 - ITR 1117 - ETR 1118 - xTR 1119 - PITR 1120 - PETR 1121 - MR 1122 - MS 1123 - DFZ 1125 Appendix B. Other Appendices 1127 -- Location/Identity Separation Brief History 1128 -- LISP History 1129 -- Old models (LISP 1, LISP 1.5, etc) 1130 -- Different mapping distribution models (e.g. LISP-NERD) 1131 -- Different mapping indexing models (LISP-ALT 1132 forwarding/overlay model), 1133 LISP-TREE DNS-based, LISP-CONS) 1135 Author's Address 1137 J. Noel Chiappa 1138 Yorktown Museum of Asian Art 1139 Yorktown, Virginia 1140 USA 1142 EMail: jnc@mit.edu