idnits 2.17.1 draft-song-yeti-testbed-experience-06.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 : ---------------------------------------------------------------------------- ** The document seems to lack a Security Considerations section. ** The document seems to lack separate sections for Informative/Normative References. All references will be assumed normative when checking for downward references. == There are 26 instances of lines with non-RFC2606-compliant FQDNs in the document. == There are 4 instances of lines with non-RFC3849-compliant IPv6 addresses in the document. If these are example addresses, they should be changed. Miscellaneous warnings: ---------------------------------------------------------------------------- == The copyright year in the IETF Trust and authors Copyright Line does not match the current year -- The document date (December 12, 2017) is 2325 days in the past. Is this intentional? Checking references for intended status: Informational ---------------------------------------------------------------------------- == Outdated reference: A later version (-03) exists of draft-song-atr-large-resp-00 -- No information found for draft-icann-dnssec-keymgmt - is the name correct? ** Obsolete normative reference: RFC 2845 (Obsoleted by RFC 8945) Summary: 3 errors (**), 0 flaws (~~), 4 warnings (==), 2 comments (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 2 Internet Engineering Task Force (IETF) L. Song, Ed. 3 Internet-Draft D. Liu 4 Intended status: Informational Beijing Internet Institute 5 Expires: June 15, 2018 P. Vixie 6 TISF 7 Kato 8 Keio University/WIDE Project 9 S. Kerr 10 December 12, 2017 12 Yeti DNS Testbed 13 draft-song-yeti-testbed-experience-06 15 Abstract 17 The Internet's Domain Name System (DNS) is built upon the foundation 18 provided by the Root Server System -- that is, the critical 19 infrastructure that serves the DNS root zone. 21 Yeti DNS is an experimental, non-production testbed that provides an 22 environment where technical and operational experiments can safely be 23 performed without risk to production infrastructure. This testbed 24 has been used by a broad community of participants to perform 25 experiments that aim to inform operations and future development of 26 the production DNS. Yeti DNS is an independently-coordinated project 27 and is not affiliated with ICANN, IANA or any Root Server Operator. 29 The Yeti DNS testbed implementation includes various novel and 30 experimental components including IPv6-only transport, independent, 31 autonomous Zone Signing Key management, large cryptographic keys and 32 a large number of Yeti-Root Servers. These differences from the Root 33 Server System have operational consequences such as large responses 34 to priming queries and the coordination of a large pool of 35 independent operators; by deploying such a system globally but 36 outside the production DNS system, the Yeti DNS project provides an 37 opportunity to gain insight into those consequences without 38 threatening the stability of the DNS. 40 This document neither addresses the relevant policies under which the 41 Root Server System is operated nor makes any proposal for changing 42 any aspect of its implementation or operation. This document aims 43 solely to document the technical and operational experience of 44 deploying a system which is similar to but different from the Root 45 Server System. 47 Status of This Memo 49 This Internet-Draft is submitted in full conformance with the 50 provisions of BCP 78 and BCP 79. 52 Internet-Drafts are working documents of the Internet Engineering 53 Task Force (IETF). Note that other groups may also distribute 54 working documents as Internet-Drafts. The list of current Internet- 55 Drafts is at https://datatracker.ietf.org/drafts/current/. 57 Internet-Drafts are draft documents valid for a maximum of six months 58 and may be updated, replaced, or obsoleted by other documents at any 59 time. It is inappropriate to use Internet-Drafts as reference 60 material or to cite them other than as "work in progress." 62 This Internet-Draft will expire on June 15, 2018. 64 Copyright Notice 66 Copyright (c) 2017 IETF Trust and the persons identified as the 67 document authors. All rights reserved. 69 This document is subject to BCP 78 and the IETF Trust's Legal 70 Provisions Relating to IETF Documents 71 (https://trustee.ietf.org/license-info) in effect on the date of 72 publication of this document. Please review these documents 73 carefully, as they describe your rights and restrictions with respect 74 to this document. Code Components extracted from this document must 75 include Simplified BSD License text as described in Section 4.e of 76 the Trust Legal Provisions and are provided without warranty as 77 described in the Simplified BSD License. 79 Table of Contents 81 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 3 82 2. Areas of Study . . . . . . . . . . . . . . . . . . . . . . . 5 83 2.1. Implementation of a Root Server System-like Testbed . . . 5 84 2.2. Yeti-Root Zone Distribution . . . . . . . . . . . . . . . 5 85 2.3. Yeti-Root Server Names and Addressing . . . . . . . . . . 5 86 2.4. IPv6-Only Yeti-Root Servers . . . . . . . . . . . . . . . 5 87 2.5. DNSSEC in the Yeti-Root Zone . . . . . . . . . . . . . . 6 88 3. Yeti DNS Testbed Infrastructure . . . . . . . . . . . . . . . 6 89 3.1. Root Zone Retrieval . . . . . . . . . . . . . . . . . . . 8 90 3.2. Transformation of Root Zone to Yeti-Root Zone . . . . . . 8 91 3.2.1. ZSK and KSK Key Sets Shared Between DMs . . . . . . . 9 92 3.2.2. Unique ZSK per DM; No Shared KSK . . . . . . . . . . 10 93 3.2.3. Preserving Root Zone NSEC Chain and ZSK RRSIGs . . . 11 94 3.3. Yeti-Root Zone Distribution . . . . . . . . . . . . . . . 11 95 3.4. Synchronisation of Service Metadata . . . . . . . . . . . 11 96 3.5. Yeti-Root Server Naming Scheme . . . . . . . . . . . . . 12 97 3.6. Yeti-Root Servers . . . . . . . . . . . . . . . . . . . . 13 98 3.7. Experimental Traffic . . . . . . . . . . . . . . . . . . 15 99 3.8. Traffic Capture and Analysis . . . . . . . . . . . . . . 15 100 4. Operational Experience with the Yeti DNS Testbed . . . . . . 16 101 4.1. Viability of IPv6-Only Operation . . . . . . . . . . . . 16 102 4.1.1. IPv6 Fragmentation . . . . . . . . . . . . . . . . . 16 103 4.1.2. Serving IPv4-Only End-Users . . . . . . . . . . . . . 18 104 4.2. Zone Distribution . . . . . . . . . . . . . . . . . . . . 18 105 4.2.1. Zone Transfers . . . . . . . . . . . . . . . . . . . 18 106 4.2.2. Delays in Yeti-Root Zone Distribution . . . . . . . . 19 107 4.3. DNSSEC KSK Rollover . . . . . . . . . . . . . . . . . . . 20 108 4.3.1. Failure-Case KSK Rollover . . . . . . . . . . . . . . 20 109 4.3.2. KSK Rollover vs. BIND9 Views . . . . . . . . . . . . 20 110 4.3.3. Large Responses during KSK Rollover . . . . . . . . . 21 111 4.4. Capture of Large DNS Response . . . . . . . . . . . . . . 22 112 4.5. Automated Hints File Maintenance . . . . . . . . . . . . 22 113 4.6. Root Label Compression in Knot . . . . . . . . . . . . . 23 114 5. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . 24 115 6. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 26 116 7. Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . 26 117 8. References . . . . . . . . . . . . . . . . . . . . . . . . . 26 118 Appendix A. Yeti-Root Hints File . . . . . . . . . . . . . . . . 29 119 Appendix B. Yeti-Root Server Priming Response . . . . . . . . . 31 120 Appendix C. Active IPv6 Prefixes in Yeti DNS testbed . . . . . . 32 121 Appendix D. Tools developed for Yeti DNS testbed . . . . . . . . 33 122 Appendix E. Controversy . . . . . . . . . . . . . . . . . . . . 34 123 Appendix F. About This Document . . . . . . . . . . . . . . . . 34 124 F.1. Venue . . . . . . . . . . . . . . . . . . . . . . . . . . 35 125 F.2. Revision History . . . . . . . . . . . . . . . . . . . . 35 126 F.2.1. draft-song-yeti-testbed-experience-00 through -03 . . 35 127 F.2.2. draft-song-yeti-testbed-experience-04 . . . . . . . . 35 128 F.2.3. draft-song-yeti-testbed-experience-05 . . . . . . . . 36 129 F.2.4. draft-song-yeti-testbed-experience-06 . . . . . . . . 36 130 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 36 132 1. Introduction 134 The Domain Name System (DNS), as originally specified in [RFC1034] 135 and [RFC1035], has proved to be an enduring and important platform 136 upon which almost every end-user of the Internet relies. Despite its 137 longevity, extensions to the protocol, new implementations and 138 refinements to DNS operations continue to emerge both inside and 139 outside the IETF. 141 The Root Server System in particular has seen technical innovation 142 and development, for example in the form of wide-scale anycast 143 deployment, the mitigation of unwanted traffic on a global scale, the 144 widespread deployment of Response Rate Limiting [RRL], the 145 introduction of IPv6 transport, the deployment of DNSSEC, changes in 146 DNSSEC key sizes and preparations to roll the root zone's Key Signing 147 Key (KSK) and corresponding trust anchor. Together, even the 148 projects listed in this brief summary imply tremendous operational 149 change, all the more impressive when considered the necessary caution 150 when managing Internet critical infrastructure, and the context of 151 the adjacent administrative changes involved in root zone management 152 and the (relatively speaking) massive increase in the the number of 153 delegations in the root zone itself . 155 Aspects of the operational structure of the Root Server System have 156 been described in such documents as [TNO2009], [ISC-TN-2003-1], 157 [RSSAC001] and [RFC7720]. Such references, considered together, 158 provide sufficient insight into the operations of the system as a 159 whole that it is straightforward to imagine structural changes to the 160 root server system's infrastructure and to wonder what the 161 operational implications of such changes might be. 163 The Yeti DNS Project was conceived in May 2015 to provide a non- 164 production testbed upon which the technical community could propose 165 and run experiments designed to answer these kinds of questions. 166 Coordination for the project was provided by TISF, the WIDE Project 167 and the Beijing Internet Institute. Many volunteers collaborated to 168 build a distributed testbed that at the time of writing includes 25 169 Yeti root servers with 16 operators and handles experimental traffic 170 from individual volunteers, universities, DNS vendors and distributed 171 measurement networks. 173 By design, the Yeti testbed system serves the root zone published by 174 the IANA with only those structural modifications necessary to ensure 175 that it is able to function usefully in the Yeti testbed system 176 instead of the production Root Server system. In particular, no 177 delegation for any top-level zone is changed, added or removed from 178 the IANA-published root zone to construct the root zone served by the 179 Yeti testbed system and changes in the root zone are reflected in the 180 testbed in near real-time. In this document, for clarity, we refer 181 to the zone derived from the IANA-published root zone as the Yeti- 182 Root zone. 184 The Yeti DNS testbed serves a similar function to the Root Server 185 System in the sense that they both serve similar zones: the Yeti-Root 186 zone and the IANA-published root zone. However, the Yeti DNS testbed 187 only serves clients that are explicitly configured to participate in 188 the experiment, whereas the Root Server System serves the whole 189 Internet. Since the dependent end-users and systems of the Yeti DNS 190 testbed are known and their operations well-coordinated with those of 191 the Yeti project, it has been possible to deploy structural changes 192 in the Yeti DNS testbed with effective measurement and analysis, 193 something that is difficult or simply impractical in the production 194 Root Server System. 196 2. Areas of Study 198 Examples of topics that the Yeti DNS Testbed was built to address are 199 included below, each illustrated with indicative questions. 201 2.1. Implementation of a Root Server System-like Testbed 203 o How can a testbed be constructed and deployed on the Internet, 204 allowing useful public participation without any risk of 205 disruption of the Root Server System? 207 o How can representative traffic be introduced into such a testbed 208 such that insights into the impact of specific differences between 209 the testbed and the Root Server System can be observed? 211 2.2. Yeti-Root Zone Distribution 213 o What are the scaling properties of Yeti-Root zone distribution as 214 the number of Yeti-Root servers, Yeti-Root server instances or 215 intermediate distribution points increase? 217 2.3. Yeti-Root Server Names and Addressing 219 o What naming schemes other than those closely analogous to the use 220 of ROOT-SERVERS.NET in the production root zone are practical, and 221 what are their respective advantages and disadvantages? 223 o What are the risks and benefits of signing the zone that contains 224 the names of the Yeti-Root servers? 226 o What automatic mechanisms might be useful to improve the rate at 227 which clients of Yeti-Root servers are able to react to a Yeti- 228 Root server renumbering event? 230 2.4. IPv6-Only Yeti-Root Servers 232 o Are there negative operational effects in the use of IPv6-only 233 Yeti-Root servers, compared to the use of servers that are dual- 234 stack? 236 o What effect does the IPv6 fragmentation model have on the 237 operation of Yeti-Root servers, compared with that of IPv4? 239 2.5. DNSSEC in the Yeti-Root Zone 241 o Is it practical to sign the Yeti-Root zone using multiple, 242 independently-operated DNSSEC signers and multiple corresponding 243 ZSKs? 245 o To what extent is [RFC5011] supported by resolvers? 247 o Does the KSK Rollover plan designed and in the process of being 248 implemented by ICANN work as expected on the Yeti testbed? 250 o What is the operational impact of using much larger RSA key sizes 251 in the ZSKs used in the Yeti-Root? 253 o What are the operational consequences of choosing DNSSEC 254 algorithms other than RSA to sign the Yeti-Root zone? 256 3. Yeti DNS Testbed Infrastructure 258 The purpose of the testbed is to allow DNS queries from stub 259 resolvers, mediated by recursive resolvers, to be delivered to Yeti- 260 Root servers, and for corresponding responses generated on the Yeti- 261 Root servers to be returned, as illustrated in Figure 1. 263 ,----------. ,-----------. ,------------. 264 | stub +------> | recursive +------> | Yeti-Root | 265 | resolver | <------+ resolver | <------+ nameserver | 266 `----------' `-----------' `------------' 267 ^ ^ ^ 268 | appropriate | Yeti-Root hints; | Yeti-Root zone 269 `- resolver `- Yeti-Root trust `- with DNSKEY RRSet 270 configured anchor signed by Yeti-KSK 272 Figure 1: High-Level Testbed Components 274 To use the Yeti DNS testbed, a recursive resolver must be configured 275 to use the Yeti-Root servers. That configuration consists of a list 276 of names and addresses for the Yeti-Root servers (often referred to 277 as a "hints file") that replaces the corresponding hints used for the 278 production Root Server System Appendix A. Resolvers also need to be 279 configured with a DNSSEC trust anchor that corresponds to a KSK used 280 in the Yeti DNS Project, in place of the normal trust anchor set used 281 for the root zone. 283 The need for a Yeti-specific trust anchor in the resolver stems from 284 the need to make minimal changes to the root zone, as retrieved from 285 the IANA, to transform it into the Yeti-Root zone that can be used in 286 the testbed. Corresponding changes are required in the Yeti-Root 287 hints file Appendix A. Those changes would be properly rejected by 288 any validator using the production Root Server System's root zone 289 trust anchor set as bogus. 291 Stub resolvers become part of the Yeti DNS Testbed by their use of 292 recursive resolvers that are configured as described above. 294 The data flow from IANA to stub resolvers through the Yeti testbed is 295 illustrated in Figure 2 and are described in more detail in the 296 sections that follow. 298 ,----------------. 299 ,-- / IANA Root Zone / ---. 300 | `----------------' | 301 | | | 302 | | | Root Zone 303 ,--------------. ,---V---. ,---V---. ,---V---. 304 | Yeti Traffic | | BII | | WIDE | | TISF | 305 | Collection | | DM | | DM | | DM | 306 `----+----+----' `---+---' `---+---' `---+---' 307 | | ,-----' ,-------' `----. 308 | | | | | Yeti-Root 309 ^ ^ | | | Zone 310 | | ,---V---. ,---V---. ,---V---. 311 | `---+ Yeti | | Yeti | . . . . . . . | Yeti | 312 | | Root | | Root | | Root | 313 | `---+---' `---+---' `---+---' 314 | | | | DNS 315 | | | | Response 316 | ,--V----------V-------------------------V--. 317 `---------+ Yeti Resolvers | 318 `--------------------+---------------------' 319 | DNS 320 | Response 321 ,--------------------V---------------------. 322 | Yeti Stub Resolvers | 323 `------------------------------------------' 325 Figure 2: Testbed Data Flow 327 3.1. Root Zone Retrieval 329 The Yeti-Root Zone is distributed within the Yeti DNS testbed through 330 a set of internal master servers that are referred to as Distribution 331 Masters (DMs). These server elements distribute the Yeti-Root zone 332 to all Yeti-Root servers. The means by which the Yeti DMs construct 333 the Yeti-Root zone for distribution is described below. 335 Since Yeti DNS DMs do not receive DNS NOTIFY [RFC1996] messages from 336 the Root Server System, a polling approach is used to determine when 337 new revisions of the root zone are available from the production Root 338 Server System. Each Yeti DM requests the root zone SOA record from a 339 root server that permits unauthenticated zone transfers of the root 340 zone, and performs a zone transfer from that server if the retrieved 341 value of SOA.SERIAL is greater than that of the last retrieved zone. 343 At the time of writing, unauthenticated zone transfers of the root 344 zone are available directly from B-Root, C-Root, F-Root, G-Root and 345 K-Root, and from L-Root via the two servers XFR.CJR.DNS.ICANN.ORG and 346 XFR.LAX.DNS.ICANN.ORG, as well as via FTP from sites maintained by 347 the Root Zone Maintainer and the IANA Functions Operator. The Yeti 348 DNS Testbed retrieves the root zone from using zone transfers from 349 F-Root. The schedule on which F-Root is polled by each Yeti DM is as 350 follows: 352 +-------------+-----------------------+ 353 | DM Operator | Time | 354 +-------------+-----------------------+ 355 | BII | UTC hour + 00 minutes | 356 | WIDE | UTC hour + 20 minutes | 357 | TISF | UTC hour + 40 minutes | 358 +-------------+-----------------------+ 360 The Yeti DNS testbed uses multiple DMs, each of which acts 361 autonomously and equivalently to its siblings. Any single DM can act 362 to distribute new revisions of the Yeti-Root zone, and is also 363 responsible for signing the RRSets that are changed as part of the 364 transformation of the Root Zone into the Yeti-Root zone described in 365 Section 3.2. This shared control over the processing and 366 distribution of the Yeti-Root zone approximates some of the ideas 367 around shared zone control explored in [ITI2014]. 369 3.2. Transformation of Root Zone to Yeti-Root Zone 371 Two distinct approaches have been deployed in the Yeti-DNS Testbed, 372 separately, to transform the Root Zone into the Yeti-Root Zone. At a 373 high level both approaches are equivalent in the sense that they 374 replace a minimal set of information in the root zone with 375 corresponding data for the Yeti DNS Testbed; the mechanisms by which 376 the transforms are executed are different, however. Each is 377 discussed in turn in Section 3.2.1 and Section 3.2.2, respectively. 379 A third approach has also been proposed, but not yet implemented. 380 The motivations and changes implied by that approach are described in 381 Section 3.2.3. 383 3.2.1. ZSK and KSK Key Sets Shared Between DMs 385 The approach described here was the first to be implemented. It 386 features entirely autonomous operation of each DM, but also requires 387 secret key material (the private key in each of the Yeti-Root KSK and 388 ZSK key-pairs) to be distributed and maintained on each DM in a 389 coordinated way. 391 The Root Zone is transformed as follows to produce the Yeti-Root 392 Zone. This transformation is carried out autonomously on each Yeti 393 DNS Project DM. Each DM carries an authentic copy of the current set 394 of Yeti KSK and ZSK key pairs, synchronized between all DMs (see 395 Section 3.4). 397 1. SOA.MNAME is set to www.yeti-dns.org. 399 2. SOA.RNAME is set to .yeti-dns.org. where is currently one of "wide", "bii" or "tisf". 402 3. All DNSKEY, RRSIG and NSEC records are removed. 404 4. The apex NS RRSet is removed, with the corresponding root server 405 glue (A and AAAA) RRSets. 407 5. A Yeti DNSKEY RRSet is added to the apex, comprising the public 408 parts of all Yeti KSK and ZSKs. 410 6. A Yeti NS RRSet is added to the apex that includes all Yeti-Root 411 servers. 413 7. Glue records (AAAA only, since Yeti-Root servers are v6-only) for 414 all Yeti-Root servers are added. 416 8. The Yeti-Root Zone is signed: the NSEC chain is regenerated; the 417 Yeti KSK is used to sign the DNSKEY RRSet, and the DM's local ZSK 418 is used to sign every other RRSet. 420 Note that the SOA.SERIAL value published in the Yeti-Root Zone is 421 identical to that found in the root zone. 423 3.2.2. Unique ZSK per DM; No Shared KSK 425 The approach described here was the second to be implemented. Each 426 DM is provisioned with its own, dedicated ZSK key pairs that are not 427 shared with other DMs. A Yeti-Root DNSKEY RRSet is constructed and 428 signed upstream of all DMs as the union of the set of active KSKs and 429 the set of active ZSKs for every individual DM. Each DM now only 430 requires the secret part of its own dedicated ZSK key pairs to be 431 available locally, and no other secret key material is shared. The 432 high-level approach is illustrated in Figure 3. 434 ,----------. ,-----------. 435 .--------> BII ZSK +---------> Yeti-Root | 436 | signs `----------' signs `-----------' 437 | 438 ,-----------. | ,----------. ,-----------. 439 | Yeti KSK +-+--------> TISF ZSK +---------> Yeti-Root | 440 `-----------' | signs `----------' signs `-----------' 441 | 442 | ,----------. ,-----------. 443 `--------> WIDE ZSK +---------> Yeti-Root | 444 signs `----------' signs `-----------' 446 Figure 3: Unique ZSK per DM 448 The process of retrieving the Root Zone from the Root Server System 449 and replacing and signing the apex DNSKEY RRSet no longer takes place 450 on the DMs, and instead takes place on a central Hidden Master. The 451 production of signed DNSKEY RRSets is analogous to the use of Signed 452 Key Responses (SKR) produced during ICANN KSK key ceremonies 453 [ICANN2010]. 455 Each DM now retrieves source data (with pre-modified and Yeti-signed 456 DNSKEY RRset, but otherwise unchanged) from the Yeti DNS Hidden 457 Master instead of from the Root Server System. 459 Each DM carries out a similar transformation to that described in 460 Section 3.2.1, except that DMs no longer need to modify or sign the 461 DNSKEY RRSet. 463 The Yeti-Root Zone served by any particular Yeti-Root Server will 464 include signatures generated using the ZSK from the DM that served 465 the Yeti-Root Zone to that Yeti-Root Server. Signatures cached at 466 resolvers might be retrieved from any Yeti-Root Server, and hence are 467 expected to be a mixture of signatures generated by different ZSKs. 468 Since all ZSKs can be trusted through the signature by the Yeti KSK 469 over the DNSKEY RRSet, which includes all ZSKs, the mixture of 470 signatures was predicted not to be a threat to reliable validation. 471 Deployment and experimentation confirms this to be the case, even 472 when individual ZSKs are rolled on different schedules. 474 A consequence of this approach is that the apex DNSKEY RRSet in the 475 Yeti-Root zone is much larger than the corresponding DNSKEY RRSet in 476 the Root Zone. 478 3.2.3. Preserving Root Zone NSEC Chain and ZSK RRSIGs 480 A change to the transformation described in Section 3.2.2 has been 481 proposed that would preserve the NSEC chain from the Root Zone and 482 all RRSIG RRs generated using the Root Zone's ZSKs. The DNSKEY RRSet 483 would continue to be modified to replace the Root Zone KSKs, and the 484 Yeti KSK would be used to generate replacement signatures over the 485 apex DNSKEY and NS RRSets. Source data would continue to flow from 486 the Root Server System through the Hidden Master to the set of DMs, 487 but no DNSSEC operations would be required on the DMs and the source 488 NSEC and most RRSIGs would remain intact. 490 This approach has been suggested in order to provide 491 cryptographically-verifiable confidence that no owner name in the 492 root zone had been changed in the process of producing the Yeti-Root 493 zone from the Root Zone, addressing one of the concerns described in 494 Appendix E in a way that can be verified automatically. 496 3.3. Yeti-Root Zone Distribution 498 Each Yeti DM is configured with a full list of Yeti-Root Server 499 addresses to send NOTIFY [RFC1996] messages to, which also forms the 500 basis for an address-based access-control list for zone transfers. 501 Authentication by address could be replaced with more rigourous 502 mechanisms (e.g. using Transaction Signatures (TSIG) [RFC2845]); this 503 has not been done at the time of writing since the use of address- 504 based controls avoids the need for the distribution of shared secrets 505 amongst the Yeti-Root Server Operators. 507 Individual Yeti-Root Servers are configured with a full set of Yeti 508 DM addresses to which SOA and AXFR queries may be sent in the 509 conventional manner. 511 3.4. Synchronisation of Service Metadata 513 Changes in the Yeti-DNS Testbed infrastructure such as the addition 514 or removal of Yeti-Root servers, renumbering Yeti-Root Servers or 515 DNSSEC key rollovers require coordinated changes to take place on all 516 DMs. The Yeti-DNS Testbed is subject to more frequent changes than 517 are observed in the Root Server System and includes substantially 518 more Yeti-Root Servers than there are IANA Root Servers, and hence a 519 manual change process in the Yeti Testbed would be more likely to 520 suffer from human error. An automated process was consequently 521 implemented. 523 A repository of all service metadata involved in the operation of 524 each DM was implemented as a dedicated git repository hosted at 525 github.com, a mechanism chosen since it was simple, transparent and 526 familiar to participants. Requests to change the service metadata 527 for a DM were submitted as pull requests from a fork of the 528 corresponding repository; each DM operator reviewed pull requests and 529 merged them to indicate approval. Once merged, changes were pulled 530 automatically to individual DMs and promoted to production. 532 3.5. Yeti-Root Server Naming Scheme 534 The current naming scheme for Root Servers was normalized to use 535 single-character host names (A through M) under the domain ROOT- 536 SERVERS.NET, as described in [RSSAC023]). The principal benefit of 537 this naming scheme was that DNS label compression could be used to 538 produce a priming response that would fit within 512 bytes at the 539 time it was introduced, 512 bytes being the maximum DNS message size 540 using UDP transport without EDNS(0) [RFC6891]. 542 Yeti-Root Servers do not use this optimization, but rather use free- 543 form nameserver names chosen by their respective operators -- in 544 other words, no attempt is made to minimize the size of the priming 545 response through the use of label compression. This approach aims to 546 challenge the need for a minimally-sized priming response in a modern 547 DNS ecosystem where EDNS(0) is prevalent. 549 Priming responses from Yeti-Root Servers do not always include server 550 addresses in the additional section, as is the case with priming 551 responses from Root Servers. In particular, Yeti-Root Servers 552 running BIND9 return an empty additional section if the configuration 553 parameter minimum-responses is set, forcing resolvers to complete the 554 priming process with a set of conventional recursive lookups in order 555 to resolve addresses for each Yeti-Root server. The Yeti-Root 556 Servers running NSD were observed to return a fully-populated 557 additional section (depending of course of the EDNS buffer size in 558 use). 560 Various approaches to normalize the composition of the priming 561 response were considered, including: 563 o Require use of DNS implementations that exhibit a desired 564 behaviour in the priming response; 566 o Modify nameserver software or configuration as used by Yeti-Root 567 Servers; 569 o Isolate the names of Yeti-Root Servers in one or more zones that 570 could be slaved on each Yeti-Root Server, renaming servers as 571 necessary, giving each a source of authoritative data with which 572 the authority section of a priming response could be fully 573 populated. This is the approach used in the Root Server System 574 with the ROOT-SERVERS.NET zone. 576 The potential mitigation of renaming all Yeti-Root Servers using a 577 scheme that would allow their names to exist directly in the root 578 zone was not considered, since that approach implies the invention of 579 new top-level labels not present in the Root Zone. 581 Given the relative infrequency of priming queries by individual 582 resolvers and the additional complexity or other compromises implied 583 by each of those mitigations, the decision was made to make no effort 584 to ensure that the composition of priming responses was identical 585 across servers. Even the empty additional sections generated by 586 Yeti-Root Servers running BIND9 seem to be sufficient for all 587 resolver software tested; resolvers simply perform a new recursive 588 lookup for each authoritative server name they need to resolve. 590 3.6. Yeti-Root Servers 592 Various volunteers have donated authoritative servers to act as Yeti- 593 Root servers. At the time of writing there are 25 Yeti-Root servers 594 distributed globally, one of which is named using an IDNA2008 595 [RFC5890] label, shown in the following list in punycode. 597 +-------------------------------------+---------------+-------------+ 598 | Name | Operator | Location | 599 +-------------------------------------+---------------+-------------+ 600 | bii.dns-lab.net | BII | CHINA | 601 | yeti-ns.tsif.net | TSIF | USA | 602 | yeti-ns.wide.ad.jp | WIDE Project | Japan | 603 | yeti-ns.as59715.net | as59715 | Italy | 604 | dahu1.yeti.eu.org | Dahu Group | France | 605 | ns-yeti.bondis.org | Bond Internet | Spain | 606 | | Systems | | 607 | yeti-ns.ix.ru | Russia | MSK-IX | 608 | yeti.bofh.priv.at | CERT Austria | Austria | 609 | yeti.ipv6.ernet.in | ERNET India | India | 610 | yeti-dns01.dnsworkshop.org | dnsworkshop | Germany | 611 | | /informnis | | 612 | dahu2.yeti.eu.org | Dahu Group | France | 613 | yeti.aquaray.com | Aqua Ray SAS | France | 614 | yeti-ns.switch.ch | SWITCH | Switzerland | 615 | yeti-ns.lab.nic.cl | CHILE NIC | Chile | 616 | yeti-ns1.dns-lab.net | BII | China | 617 | yeti-ns2.dns-lab.net | BII | China | 618 | yeti-ns3.dns-lab.net | BII | China | 619 | ca...a23dc.yeti-dns.net | Yeti-ZA | South | 620 | | | Africa | 621 | 3f...374cd.yeti-dns.net | Yeti-AU | Australia | 622 | yeti1.ipv6.ernet.in | ERNET India | India | 623 | xn--r2bi1c.xn--h2bv6c0a.xn--h2brj9c | ERNET India | India | 624 | yeti-dns02.dnsworkshop.org | dnsworkshop | USA | 625 | | /informnis | | 626 | yeti.mind-dns.nl | Monshouwer | Netherlands | 627 | | Internet | | 628 | | Diensten | | 629 | yeti-ns.datev.net | DATEV | Germany | 630 | yeti.jhcloos.net. | jhcloos | USA | 631 +-------------------------------------+---------------+-------------+ 633 The current list of Yeti-Root server is made available to a 634 participating resolver first using a substitute hints file Appendix A 635 and subsequently by the usual resolver priming process [RFC8109]. 636 All Yeti-Root servers are IPv6-only, foreshadowing a future IPv6-only 637 Internet, and hence the Yeti-Root hints file contains no IPv4 638 addresses and the Yeti-Root zone contains no IPv4 glue. 640 At the time of writing, all root servers within the Root Server 641 System serve the ROOT-SERVERS.NET zone in addition to the root zone, 642 and all but one also serve the ARPA zone. Yeti-Root servers serve 643 the Yeti-Root zone only. 645 Significant software diversity exists across the set of Yeti-Root 646 servers, as reported by their volunteer operators at the time of 647 writing: 649 o Platform: 18 of 25 Yeti-Root servers are implemented on a VPS 650 rather than bare metal. 652 o Operating System: 15 Yeti-Root servers run on Linux (Ubuntu, 653 Debian, CentOS, Red Hat and ArchLinux); 4 run on FreeBSD, 1 on 654 NetBSD and 1 in Windows server 2016. 656 o DNS software: 18 of 25 Yeti-Root servers use BIND9 (versions 657 varying between 9.9.7 and 9.10.3); 4 use NSD (4.10 and 4.15); 2 658 use Knot (2.0.1 and 2.1.0), 1 uses Bundy (1.2.0) and 1 uses MS DNS 659 (10.0.14300.1000). 661 3.7. Experimental Traffic 663 For the Yeti DNS Testbed to be useful as a platform for 664 experimentation, it needs to carry statistically representative 665 traffic. Several approaches have been taken to load the system with 666 traffic, including both real-world traffic triggered by end-users and 667 synthetic traffic. 669 Resolvers that have been explicitly configured to participate in the 670 testbed, as described in Section 3, are a source of real-world, end- 671 user traffic. Sustained levels of traffic have been observed from a 672 variety of sources, as summarised in Appendix C. 674 Synthetic traffic has been introduced to the system from time to time 675 in order to increase traffic loads. Approaches include the use of 676 distributed measurement platforms such as RIPE ATLAS to send DNS 677 queries to Yeti-Root servers, and the capture of traffic sent from 678 non-Yeti resolvers to the Root Server System which was subsequently 679 modified and replayed towards Yeti-Root servers. 681 3.8. Traffic Capture and Analysis 683 Query and response traffic capture is available in the testbed in 684 both Yeti resolvers and Yeti-Root servers in anticipation of 685 experiments that require packet-level visibility into DNS traffic. 687 Traffic capture is performed on Yeti-Root servers using either dnscap 688 or pcapdump (part of the 689 pcaputils Debian package , 690 with a patch to facilitate triggered file upload 691 . PCAP- 692 format files containing packet captures are uploaded using rsync to 693 central storage. 695 4. Operational Experience with the Yeti DNS Testbed 697 The following sections provide commentary on the operation and impact 698 analyses of the Yeti-DNS Testbed described in Section 3. More 699 detailed descriptions of observed phenomena are available in Yeti DNS 700 mailing list archives 701 and on the Yeti DNS blog . 703 4.1. Viability of IPv6-Only Operation 705 All Yeti-Root servers were deployed with IPv6 connectivity, and no 706 IPv4 addresses for any Yeti-Root server were made available (e.g. in 707 the Yeti hints file, or in the DNS itself). This implementation 708 decision constrained the Yeti-Root system to be v6-only. 710 DNS implementations are generally adept at using both IPv4 and IPv6 711 when both are available. Servers that cannot be reliably reached 712 over one protocol might be better queried over the other, to the 713 benefit of end-users in the common case where DNS resolution is on 714 the critical path for end-users' perception of performance. However, 715 this optimisation also means that systemic problems with one protocol 716 can be masked by the other. By forcing all traffic to be carried 717 over IPv6, the Yeti DNS testbed aimed to expose any such problems and 718 make them easier to identify and understand. Several examples of 719 IPv6-specific phenomena observed during the operation of the testbed 720 are described in the sections that follow. 722 Although the Yeti-Root servers themselves were only reachable using 723 IPv6, real-world end-users often have no IPv6 connectivity. The 724 testbed was also able to explore the degree to which IPv6-only Yeti- 725 Root servers were able to serve single-stack, IPv4-only end-user 726 populations through the use of dual-stack Yeti resolvers. 728 4.1.1. IPv6 Fragmentation 730 In the Root Server System, structural changes with the potential to 731 increase response sizes (and hence fragmentation, fallback to TCP 732 transport or both) have been exercised with great care, since the 733 impact on clients has been difficult to predict or measure. The Yeti 734 DNS Testbed is experimental and has the luxury of a known client 735 base, making it far easier to make such changes and measure their 736 impact. 738 Many of the experimental design choices described in this document 739 were expected to trigger larger responses. For example, the choice 740 of naming scheme for Yeti-Root Servers described in Section 3.5 741 defeats label compression. It makes a large priming response (up to 742 1754 octets with 25 NS server and their glue) ; the Yeti-Root zone 743 transformation approach described in Section 3.2.2 greatly enlarges 744 the apex DNSKEY RRSet especially during the KSK rollover (up to 1975 745 octets with 3 ZSK and 2 KSK). An increased incidence of 746 fragmentation was therefore expected. 748 The Yeti-DNS Testbed provides service on IPv6 only. IPv6 has a 749 fragmentation model that is different from IPv4 -- in particular, 750 fragmentation always takes place on the sending host, and not on an 751 intermediate router. 753 Fragmentation may cause serious issues; if a single fragment is lost, 754 it results in the loss of the entire datagram of which the fragment 755 was a part, and in the DNS frequently triggers a timeout. It is 756 known at this moment that only a limited number of security middle- 757 box implementations support IPv6 fragments. Some public measurements 758 and reports [I-D.taylor-v6ops-fragdrop] [RFC7872] shows that there is 759 notable packets drop rate due to the mistreatment of middle-box on 760 IPv6 fragment. One APNIC study [IPv6-frag-DNS] reported that 37% of 761 endpoints using IPv6-capable DNS resolver cannot receive a fragmented 762 IPv6 response over UDP. 764 To study the impact, RIPE Atlas probes were used. For each Yeti-Root 765 server, an Atlas measurement was setup using 100 IPv6-enabled probes 766 from five regions, sending a DNS query for ./IN/DNSKEY using UDP 767 transport with DO=1. This measurement, when carried out concurrently 768 with a Yeti KSK rollover, further exacerbating the potential for 769 fragmentation, identified a 7% failure rate compared with a non- 770 fragmented control. A failure rate of 2% was observed with response 771 sizes of 1414 octets, which was surprising given the expected 772 prevalence of 1500-octet (Ethernet-framed) MTUs. 774 The consequences of fragmentation were not limited to failures in 775 delivering DNS responses over UDP transport. There were two cases 776 where a Yeti-Root server failed to transfer the Yeti-Root zone from a 777 DM. DM log files revealed "socket is not connected" errors 778 corresponding to zone transfer requests. Further experimentation 779 revealed that combinations of NetBSD 6.1, NetBSD 7.0RC1, FreeBSD 780 10.0, Debian 3.2 and VMWare ESXI 5.5 resulted in a high TCP MSS value 781 of 1440 octets being negotiated between client and server despite the 782 presence of the IPV6_USE_MIN_MTU socket option, as described in 783 [I-D.andrews-tcp-and-ipv6-use-minmtu]. The mismatch appears to cause 784 outbound segments greater in size than 1280 octets to be dropped 785 before sending. Setting the local TCP MSS to 1220 octets (chosen as 786 1280-60, the size of the IPv6/TCP header with no other extension 787 headers) was observed to be a pragmatic mitigation. 789 4.1.2. Serving IPv4-Only End-Users 791 Yeti resolvers have been successfully used by real-world end-users 792 for general name resolution within a number of participant 793 organisations, including resolution of names to IPv4 addresses and 794 resolution by IPv4-only end-user devices. 796 Some participants, recognising the operational importance of 797 reliability in resolver infrastructure and concerned about the 798 stability of their IPv6 connectivity, chose to deploy Yeti resolvers 799 in parallel to conventional resolvers, making both available to end- 800 users. While the viability of this approach provides a useful data 801 point, end-users using Yeti resolvers exclusively provided a better 802 opportunity to identify and understand any failures in the Yeti DNS 803 testbed infrastructure. 805 Resolvers deployed in IPv4-only environments were able to join the 806 Yeti DNS testbed by way of upstream, dual-stack Yeti resolvers, or in 807 one case, in CERNET2, by assigning IPv4 addresses to Yeti-Root 808 servers and mapping them in dual-stack IVI translation devices 809 [RFC6219]. 811 4.2. Zone Distribution 813 The Yeti DNS testbed makes use of multiple DMs to distribute the 814 Yeti-Root zone, an approach that would allow the number of Yeti-Root 815 servers to scale to a higher number than could be supported by a 816 single distribution source and which provided redundancy. The use of 817 multiple DMs introduced some operational challenges, however, which 818 are described in the following sections. 820 4.2.1. Zone Transfers 822 Yeti-Root Servers were configured to serve the Yeti-Root zone as 823 slaves. Each slave had all DMs configured as masters, providing 824 redundancy in zone synchronisation. 826 Each DM in the Yeti testbed served a Yeti-Root zone which is 827 functionally equivalent but not congruent to that served by every 828 other DM (see Section 3.3). The differences included variations in 829 the SOA.MNAME field and, more critically, in the RRSIGs for 830 everything other than the apex DNSKEY RRSet, since signatures for all 831 other RRSets are generated using a private key that is only available 832 to the DM serving its particular variant of the zone (see 833 Section 3.2, Section 3.2.2). 835 Incremental Zone Transfer (IXFR), as described in [RFC1995], is a 836 viable mechanism to use for zone synchronization between any Yeti- 837 Root server and a consistent, single DM. However, if that Yeti-Root 838 server ever selected a different DM, IXFR would no longer be a safe 839 mechanism; structural changes between the incongruent zones on 840 different DMs would not be included in any transferred delta and the 841 result would be a zone that was not internally self-consistent. For 842 this reason the first transfer after a change of DM would require 843 AXFR, not IXFR. 845 None of the DNS software in use on Yeti-Root Servers supports this 846 mixture of IXFR/AXFR according to the master server in use. This is 847 unsurprising, given that the environment described above in the Yeti- 848 Root system is idiosyncratic; conventional zone transfer graphs 849 involve zones that are congruent between all nodes. For this reason, 850 all Yeti-Root servers are configured to use AXFR at all times, and 851 never IXFR, to ensure that zones being served are internally self- 852 consistent. 854 4.2.2. Delays in Yeti-Root Zone Distribution 856 Each Yeti DM polled the Root Server System for a new revision of the 857 root zone on an interleaved schedule, as described in Section 3.1. 858 Consequently, different DMs were expected to retrieve each revision 859 of the root zone, and make a corresponding revision of the Yeti-Root 860 zone available, at different times. The availability of a new 861 revision of the Yeti-Root zone on the first DM would typically 862 precede that of the last by 40 minutes. 864 It might be expected given this distribution mechanism that the 865 maximum latency between the publication of a new revision of the root 866 zone and the availability of the corresponding Yeti-Root zone on any 867 Yeti-Root server would be 20 minutes, since in normal operation at 868 least one DM should serve that Yeti-Zone within 20 minutes of root 869 zone publication. In practice, this was not observed. 871 In one case a Yeti-Root server running Bundy 1.2.0 on FreeBSD 872 10.2-RELEASE was found to lag root zone publication by as much as ten 873 hours, which upon investigation was due to software defects that were 874 subsequently corrected. 876 More generally, Yeti-Root servers were observed routinely to lag root 877 zone publication by more than 20 minutes, and relatively often by 878 more than 40 minutes. Whilst in some cases this might be assumed to 879 be a result of connectivity problems, perhaps suppressing the 880 delivery of NOTIFY messages, it was also observed that Yeti-Root 881 servers receiving a NOTIFY from one DM would often send SOA queries 882 and AXFR requests to a different DM. If that DM was not yet serving 883 the new revision of the Yeti-Root zone, a delay in updating the Yeti- 884 Root server would naturally result. 886 4.3. DNSSEC KSK Rollover 888 At the time of writing, the Root Zone KSK is expected to undergo a 889 carefully-orchestrated rollover as described in [ICANN2016]. ICANN 890 has commissioned various tests and has published an external test 891 plan [ICANN2017]. 893 Three related DNSSEC KSK rollover exercises were carried out on the 894 Yeti DNS testbed, somewhat concurrent with the planning and execution 895 of the rollover in the root zone. Brief descriptions of these 896 exercises are included below. 898 4.3.1. Failure-Case KSK Rollover 900 The first KSK rollover that was executed on the Yeti DNS testbed 901 deliberately ignored the 30-day hold-down timer specified in 902 [RFC5011] before retiring the outgoing KSK. 904 It was confirmed that clients of some (but not all) validating Yeti 905 resolvers experienced resolution failures (received SERVFAIL 906 responses) following this change. Those resolvers required 907 administrator intervention to install a functional trust anchor 908 before resolution was restored. 910 4.3.2. KSK Rollover vs. BIND9 Views 912 The second Yeti KSK rollover was designed with similar phases to the 913 ICANN's KSK rollover roll, although with modified timings to reduce 914 the time required to complete the process. The "slot" used in this 915 rollover was ten days long, as follows: 917 +--------------+----------+----------+ 918 | | 19444 | New Key | 919 +--------------+----------+----------+ 920 | slot 1 | pub+sign | | 921 | slot 2,3,4,5 | pub+sign | pub | 922 | slot 6,7 | pub | pub+sign | 923 | slot 8 | revoke | pub+sign | 924 | slot 9 | | pub+sign | 925 +--------------+----------+----------+ 927 During this rollover exercise, a problem was observed on one Yeti 928 resolver that was running BIND 9.10.4-p2 [KROLL-ISSUE]. That 929 resolver was configured with multiple views serving clients in 930 different subnets at the time that the KSK rollover began. DNSSEC 931 validation failures were observed following the completion of the KSK 932 rollover, triggered by the addition of a new view, intended to serve 933 clients from a new subnet. 935 BIND 9.10 requires "managed-keys" configuration to be specified in 936 every view, a detail that was apparently not obvious to the operator 937 in this case and which was subsequently highlighted by ISC in their 938 general advice relating to KSK rollover in the root zone to users of 939 BIND 9 . When the "managed-keys" configuration is 941 present in every view that is configured to perform validation, trust 942 anchors for all views are updated during a KSK rollover. 944 4.3.3. Large Responses during KSK Rollover 946 Since a KSK rollover necessarily involves the publication of outgoing 947 and incoming public keys simultaneously, an increase in the size of 948 DNSKEY responses is expected. The third KSK rollover carried out on 949 the Yeti DNS testbed was accompanied by a concerted effort to observe 950 response sizes and their impact on end-users. 952 As described in Section 3.2.2, in the Yeti DNS testbed each DM can 953 maintain control of its own set of ZSKs, which can undergo rollover 954 independently. During a KSK rollover where concurrent ZSK rollovers 955 are executed by each of three DMs the maximum number of apex DNSKEY 956 RRs present is eight (incoming and outcoming KSK, plus incoming and 957 outgoing of each of three ZSKs). In practice, however, such 958 concurrency did not occur; only the BII ZSK was rolled during the KSK 959 rollover, and hence only three DNSKEY RRSet configurations were 960 observed: 962 o 3 ZSK and 2 KSK, DNSKEY response of 1975 octets; 964 o 3 ZSK and 1 KSK, DNSKEY response of 1414 octets; and 966 o 2 ZSK and 1 KSK, DNSKEY response of 1139 octets. 968 RIPE Atlast probes were used as described in Section 4.1.1 to send 969 DNSKEY queries directly to Yeti-Root servers. The numbers of queries 970 and failures were recorded and categorised according to the response 971 sizes at the time the queries were sent. A summary of the results is 972 as follows: 974 +---------------+----------+---------------+--------------+ 975 | Response Size | Failures | Total Queries | Failure rate | 976 +---------------+----------+---------------+--------------+ 977 | 1139 | 274 | 64252 | 0.0042 | 978 | 1414 | 3141 | 126951 | 0.0247 | 979 | 1975 | 2920 | 42529 | 0.0687 | 980 +---------------+----------+---------------+--------------+ 982 The general approach illustrated briefly here provides a useful 983 example of how the design of the Yeti DNS testbed, separate from the 984 Root Server System but constructed as a live testbed on the Internet, 985 facilitates the use of general-purpose active measurement facilities 986 such as RIPE Atlas probes as well as internal passive measurement 987 such as packet capture. 989 4.4. Capture of Large DNS Response 991 Packet capture is a common approach in production DNS systems where 992 operators require fine-grained insight into traffic in order to 993 understand production traffic. For authoritative servers, capture of 994 inbound query traffic is often sufficient, since responses can be 995 synthesised with knowledge of the zones being served at the time the 996 query was received. Queries are generally small enough not to be 997 fragmented, and even with TCP transport are generally packed within a 998 single segment. 1000 The Yeti DNS testbed has different requirements; in particular there 1001 is a desire to compare responses obtained from the Yeti 1002 infrastructure with those received from the Root Server System in 1003 response to a single query stream (e.g. using YmmV as described in 1004 Appendix D). Some Yeti-Root servers were capable of recovering 1005 complete DNS messages from within nameservers, e.g. using dnstap; 1006 however, not all servers provided that functionality and a consistent 1007 approach was desirable. 1009 The requirement passive capture of responses from the wire together 1010 with experiments that were expected (and in some cases designed) to 1011 trigger fragmentation and use of TCP transport led to the development 1012 of a new tool, PcapParser, to perform fragment and TCP stream 1013 reassembly from raw packet capture data. A brief description of 1014 PcapParser is included in Appendix D. 1016 4.5. Automated Hints File Maintenance 1018 Renumbering events in the Root Server System are relatively rare. 1019 Although each such event is accompanied by the publication of an 1020 updated hints file in standard locations, the task of updating local 1021 copies of that file used by DNS resolvers is manual, and the process 1022 has an observably-long tail: for example, in 2015 J-Root was still 1023 receiving traffic at its old address some thirteen years after 1024 renumbering [Wessels2015]. 1026 The observed impact of these old, deployed hints file is minimal, 1027 likely due to the very low frequency of such renumbering events. 1028 Even the oldest of hints file would still contain some accurate root 1029 server addresses from which priming responses could be obtained. 1031 By contrast, due to the experimental nature of the system and the 1032 fact that it is operated mainly by volunteers, Yeti-Root Servers are 1033 added, removed and renumbered with much greater frequency. A tool to 1034 facilitate automatic maintenance of hints files was therefore 1035 created, [hintUpdate]. 1037 The automated procedure followed by the hintUpdate tool is as 1038 follows. 1040 1. Use the local resolver to obtain a response to the query ./IN/NS; 1042 2. Use the local resolver to obtain a set of IPv4 and IPv6 addresses 1043 for each name server; 1045 3. Validate all signatures obtained from the local resolvers, and 1046 confirm that all data is signed; 1048 4. Compare the data obtained to that contained within the currently- 1049 active hints file; if there are differences, rotate the old one 1050 away and replace it with a new one. 1052 This tool would not function unmodified when used in the Root Server 1053 System, since the names of individual Root Servers (e.g. A.ROOT- 1054 SERVERS.NET) are not signed. All Yeti-Root Server names are signed, 1055 however, and hence this tool functions as expected in that 1056 environment. 1058 4.6. Root Label Compression in Knot 1060 [RFC1035] specifies that domain names can be compressed when encoded 1061 in DNS messages, being represented as one of 1063 1. a sequence of labels ending in a zero octet; 1065 2. a pointer; or 1067 3. a sequence of labels ending with a pointer. 1069 The purpose of this flexibility is to reduce the size of domain names 1070 encoded in DNS messages. 1072 It was observed that Yeti-Root Servers running Knot 2.0 would 1073 compress the zero-length label (the root domain, often represented as 1074 ".") using a pointer to an earlier example. Although legal, this 1075 encoding increases the encoded size of the root label from one octet 1076 to two; it was also found to break some client software, in 1077 particular the Go DNS library. Bug reports were filed against both 1078 knot and the Go DNS library, and both were resolved in subsequent 1079 releases. 1081 5. Conclusions 1083 Yeti DNS was designed and implemented as a live DNS root system 1084 testbed. It serves a root zone ("Yeti-Root" in this document) 1085 derived from the root zone root zone published by the IANA with only 1086 those structural modifications necessary to ensure its function in 1087 the testbed system. The Yeti DNS testbed has proven to be a useful 1088 platform to address many questions that would be challenging to 1089 answer using the production Root Server System, such as those 1090 included in Section 2. 1092 Indicative findings following from the construction and operation of 1093 the Yeti DNS testbed include: 1095 o Operation in a pure IPv6-only environment; confirmation of a 1096 significant failure rate in the transmission of large responses 1097 (~7%), but no other persistent failures observed. Two cases in 1098 which Yeti-Root servers failed to retrieve the Yeti-Root zone due 1099 to fragmentation of TCP segments; mitigated by setting a TCP MSS 1100 of 1220 octets (see Section 4.1.1). 1102 o Successful operation with three autonomous Yeti-Root zone signers 1103 and 25 Yeti-Root servers, and confirmation that IXFR is not an 1104 appropriate transfer mechanism of zones that are structurally 1105 incongruent across different transfer paths (see Section 4.2). 1107 o ZSK size increased to 2048 bits and multiple KSK rollovers 1108 executed to exercise RFC 5011 support in validating resolvers; 1109 identification of pitfalls relating to views in BIND9 when 1110 configured with "managed-keys" (see Section 4.3). 1112 o Use of natural (non-normalised) names for Yeti-Root servers 1113 exposed some differences between implementations in the inclusion 1114 of additional-section glue in responses to priming queries; 1115 however, despite this inefficiency, Yeti resolvers were observed 1116 to function adequately (see Section 3.5). 1118 o It was observed that Knot 2.0 performed label compression on the 1119 root (empty) label. This results in an increased encoding size 1120 for references to the root label, since a pointer is encoded as 1121 two octets whilst the root label itself only requires one (see 1122 Section 4.6). 1124 o Some tools were developed in response to the operational 1125 experience of running and using the Yeti DNS testbed: DNS fragment 1126 and DNS ATR for large DNS responses, a BIND9 patch for additional 1127 section glue, YmmV and IPv6 defrag for capturing and mirroring 1128 traffic. In addition a tool to facilitate automatic maintenance 1129 of hints files was created (see Appendix D). 1131 The Yeti DNS testbed was used only by end-users whose local 1132 infrastructure providers had made the conscious decision to do so, as 1133 is appropriate for an experimental, non-production system. However, 1134 the service quality reported from end-users of the system was high, 1135 in the main case indistinguishable from that of the production Root 1136 Server System. 1138 The experience gained during the operation of the Yeti DNS testbed 1139 suggested several topics worthy of further study: 1141 o Priming Truncation and TCP-only Yeti-Root servers: observe and 1142 measure the worst-possible case for priming truncation by 1143 responding with TC=1 to all priming queries received over UDP 1144 transport, forcing clients to retry using TCP. This should also 1145 give some insight into the usefulness of TCP-only DNS in general. 1147 o KSK ECDSA Rollover: one possible way to reduce DNSKEY response 1148 sizes is to change to an elliptic curve signing algorithm. While 1149 in principle this can be done separately for the KSK and the ZSK, 1150 the RIPE NCC has done research recently and discovered that some 1151 resolvers require that both KSK and ZSK use the same algorithm. 1152 This means that an algorithm roll also involves a KSK roll. 1153 Performing an algorithm roll at the root would be an interesting 1154 challenge. 1156 o Sticky Notify for zone transfer: the non-applicability of IXFR as 1157 a zone transfer mechanism in the Yeti DNS testbed could be 1158 mitigated by the implementation of a sticky preference for master 1159 server for each slave, such that an initial AXFR response could be 1160 followed up with IXFR requests without compromising zone integrity 1161 in the case (as with Yeti) that equivalent but incongruent 1162 versions of a zone are served by different masters. 1164 o Key distribution for zone transfer credentials: the use of a 1165 shared secret between slave and master requires key distribution 1166 and management whose scaling properties are not ideally suited to 1167 systems with large numbers of transfer clients. Other approaches 1168 for key distribution and authentication could be considered. 1170 6. IANA Considerations 1172 This document requests no action of the IANA. 1174 7. Acknowledgments 1176 The editors would like to acknowledge the contributions of the 1177 various and many subscribers to the Yeti DNS Project mailing lists, 1178 including the following people who were involved in the 1179 implementation and operation of the Yeti DNS testbed itself: 1181 Tomohiro Ishihara, Antonio Prado, Stephane Bortzmeyer, Mickael 1182 Jouanne, Pierre Beyssac, Joao Damas, Pavel Khramtsov, Ma Yan, 1183 Otmar Lendl, Praveen Misra, Carsten Strotmann, Edwin Gomez, Remi 1184 Gacogne, Guillaume de Lafond, Yves Bovard, Hugo Salgado-Hernandez, 1185 Li Zhen, Daobiao Gong, Runxia Wan. 1187 The editors also acknowledge the assistance of the Independent 1188 Submissions Editorial Board, and of the following reviewers whose 1189 opinions helped improve the clarity of this document: 1191 Subramanian Moonesamy, Joe Abley,Paul Mockapetris 1193 8. References 1195 [hintUpdate] 1196 "Hintfile Auto Update", 2015, 1197 . 1199 [I-D.andrews-tcp-and-ipv6-use-minmtu] 1200 Andrews, M., "TCP Fails To Respect IPV6_USE_MIN_MTU", 1201 draft-andrews-tcp-and-ipv6-use-minmtu-04 (work in 1202 progress), October 2015. 1204 [I-D.muks-dns-message-fragments] 1205 Sivaraman, M., Kerr, S., and D. Song, "DNS message 1206 fragments", draft-muks-dns-message-fragments-00 (work in 1207 progress), July 2015. 1209 [I-D.song-atr-large-resp] 1210 Song, L., "ATR: Additional Truncated Response for Large 1211 DNS Response", draft-song-atr-large-resp-00 (work in 1212 progress), September 2017. 1214 [I-D.taylor-v6ops-fragdrop] 1215 Jaeggli, J., Colitti, L., Kumari, W., Vyncke, E., Kaeo, 1216 M., and T. Taylor, "Why Operators Filter Fragments and 1217 What It Implies", draft-taylor-v6ops-fragdrop-02 (work in 1218 progress), December 2013. 1220 [ICANN2010] 1221 "DNSSEC Key Management Implementation for the Root Zone", 1222 May 2010, . 1226 [ICANN2016] 1227 "Root Zone KSK Rollover Plan", 2016, 1228 . 1231 [ICANN2017] 1232 "2017 KSK Rollover External Test Plan", July 2016, 1233 . 1236 [IPv6-frag-DNS] 1237 "Dealing with IPv6 fragmentation in the DNS", August 2017, 1238 . 1241 [ISC-TN-2003-1] 1242 Abley, J., "Hierarchical Anycast for Global Service 1243 Distribution", March 2003, 1244 . 1246 [ITI2014] "Identifier Technology Innovation Report", May 2014, 1247 . 1250 [KROLL-ISSUE] 1251 "A DNSSEC issue during Yeti KSK rollover", 2016, 1252 . 1255 [RFC1034] Mockapetris, P., "Domain names - concepts and facilities", 1256 STD 13, RFC 1034, DOI 10.17487/RFC1034, November 1987, 1257 . 1259 [RFC1035] Mockapetris, P., "Domain names - implementation and 1260 specification", STD 13, RFC 1035, DOI 10.17487/RFC1035, 1261 November 1987, . 1263 [RFC1995] Ohta, M., "Incremental Zone Transfer in DNS", RFC 1995, 1264 DOI 10.17487/RFC1995, August 1996, 1265 . 1267 [RFC1996] Vixie, P., "A Mechanism for Prompt Notification of Zone 1268 Changes (DNS NOTIFY)", RFC 1996, DOI 10.17487/RFC1996, 1269 August 1996, . 1271 [RFC2826] Internet Architecture Board, "IAB Technical Comment on the 1272 Unique DNS Root", RFC 2826, DOI 10.17487/RFC2826, May 1273 2000, . 1275 [RFC2845] Vixie, P., Gudmundsson, O., Eastlake 3rd, D., and B. 1276 Wellington, "Secret Key Transaction Authentication for DNS 1277 (TSIG)", RFC 2845, DOI 10.17487/RFC2845, May 2000, 1278 . 1280 [RFC5011] StJohns, M., "Automated Updates of DNS Security (DNSSEC) 1281 Trust Anchors", STD 74, RFC 5011, DOI 10.17487/RFC5011, 1282 September 2007, . 1284 [RFC5890] Klensin, J., "Internationalized Domain Names for 1285 Applications (IDNA): Definitions and Document Framework", 1286 RFC 5890, DOI 10.17487/RFC5890, August 2010, 1287 . 1289 [RFC6219] Li, X., Bao, C., Chen, M., Zhang, H., and J. Wu, "The 1290 China Education and Research Network (CERNET) IVI 1291 Translation Design and Deployment for the IPv4/IPv6 1292 Coexistence and Transition", RFC 6219, 1293 DOI 10.17487/RFC6219, May 2011, 1294 . 1296 [RFC6891] Damas, J., Graff, M., and P. Vixie, "Extension Mechanisms 1297 for DNS (EDNS(0))", STD 75, RFC 6891, 1298 DOI 10.17487/RFC6891, April 2013, 1299 . 1301 [RFC7720] Blanchet, M. and L-J. Liman, "DNS Root Name Service 1302 Protocol and Deployment Requirements", BCP 40, RFC 7720, 1303 DOI 10.17487/RFC7720, December 2015, 1304 . 1306 [RFC7872] Gont, F., Linkova, J., Chown, T., and W. Liu, 1307 "Observations on the Dropping of Packets with IPv6 1308 Extension Headers in the Real World", RFC 7872, 1309 DOI 10.17487/RFC7872, June 2016, 1310 . 1312 [RFC8109] Koch, P., Larson, M., and P. Hoffman, "Initializing a DNS 1313 Resolver with Priming Queries", BCP 209, RFC 8109, 1314 DOI 10.17487/RFC8109, March 2017, 1315 . 1317 [RRL] Vixie, P. and V. Schryver, "Response Rate Limiting (RRL)", 1318 June 2012, . 1320 [RSSAC001] 1321 "Service Expectations of Root Servers", December 2015, 1322 . 1325 [RSSAC023] 1326 "History of the Root Server System", November 2016, 1327 . 1330 [TNO2009] Gijsen, B., Jamakovic, A., and F. Roijers, "Root Scaling 1331 Study: Description of the DNS Root Scaling Model", 1332 September 2009, 1333 . 1336 [Wessels2015] 1337 Wessels, D., "Thirteen Years of Old J-Root", 2015, 1338 . 1342 Appendix A. Yeti-Root Hints File 1344 The following hints file (complete and accurate at the time of 1345 writing) causes a DNS resolver to use the Yeti DNS testbed in place 1346 of the production Root Server System and hence participate in 1347 experiments running on the testbed. 1349 Note that some lines have been wrapped in the text that follows in 1350 order to fit within the production constraints of this document. 1351 Wrapped lines are indicated with a blackslash character ("\"), 1352 following common convention. 1354 . 3600000 IN NS bii.dns-lab.net 1355 bii.dns-lab.net 3600000 IN AAAA 240c:f:1:22::6 1356 . 3600000 IN NS yeti-ns.tisf.net 1357 yeti-ns.tisf.net 3600000 IN AAAA 2001:559:8000::6 1358 . 3600000 IN NS yeti-ns.wide.ad.jp 1359 yeti-ns.wide.ad.jp 3600000 IN AAAA 2001:200:1d9::35 1360 . 3600000 IN NS yeti-ns.as59715.net 1361 yeti-ns.as59715.net 3600000 IN AAAA \ 1362 2a02:cdc5:9715:0:185:5:203:53 1363 . 3600000 IN NS dahu1.yeti.eu.org 1364 dahu1.yeti.eu.org 3600000 IN AAAA \ 1365 2001:4b98:dc2:45:216:3eff:fe4b:8c5b 1366 . 3600000 IN NS ns-yeti.bondis.org 1367 ns-yeti.bondis.org 3600000 IN AAAA 2a02:2810:0:405::250 1368 . 3600000 IN NS yeti-ns.ix.ru 1369 yeti-ns.ix.ru 3600000 IN AAAA 2001:6d0:6d06::53 1370 . 3600000 IN NS yeti.bofh.priv.at 1371 yeti.bofh.priv.at 3600000 IN AAAA 2a01:4f8:161:6106:1::10 1372 . 3600000 IN NS yeti.ipv6.ernet.in 1373 yeti.ipv6.ernet.in 3600000 IN AAAA 2001:e30:1c1e:1::333 1374 . 3600000 IN NS yeti-dns01.dnsworkshop.org 1375 yeti-dns01.dnsworkshop.org \ 1376 3600000 IN AAAA 2001:1608:10:167:32e::53 1377 . 3600000 IN NS yeti-ns.conit.co 1378 yeti-ns.conit.co 3600000 IN AAAA \ 1379 2604:6600:2000:11::4854:a010 1380 . 3600000 IN NS dahu2.yeti.eu.org 1381 dahu2.yeti.eu.org 3600000 IN AAAA 2001:67c:217c:6::2 1382 . 3600000 IN NS yeti.aquaray.com 1383 yeti.aquaray.com 3600000 IN AAAA 2a02:ec0:200::1 1384 . 3600000 IN NS yeti-ns.switch.ch 1385 yeti-ns.switch.ch 3600000 IN AAAA 2001:620:0:ff::29 1386 . 3600000 IN NS yeti-ns.lab.nic.cl 1387 yeti-ns.lab.nic.cl 3600000 IN AAAA 2001:1398:1:21::8001 1388 . 3600000 IN NS yeti-ns1.dns-lab.net 1389 yeti-ns1.dns-lab.net 3600000 IN AAAA 2001:da8:a3:a027::6 1390 . 3600000 IN NS yeti-ns2.dns-lab.net 1391 yeti-ns2.dns-lab.net 3600000 IN AAAA 2001:da8:268:4200::6 1392 . 3600000 IN NS yeti-ns3.dns-lab.net 1393 yeti-ns3.dns-lab.net 3600000 IN AAAA 2400:a980:30ff::6 1394 . 3600000 IN NS \ 1395 ca978112ca1bbdcafac231b39a23dc.yeti-dns.net 1396 ca978112ca1bbdcafac231b39a23dc.yeti-dns.net \ 1397 3600000 IN AAAA 2c0f:f530::6 1398 . 3600000 IN NS \ 1399 3e23e8160039594a33894f6564e1b1.yeti-dns.net 1400 3e23e8160039594a33894f6564e1b1.yeti-dns.net \ 1401 3600000 IN AAAA 2803:80:1004:63::1 1402 . 3600000 IN NS \ 1403 3f79bb7b435b05321651daefd374cd.yeti-dns.net 1404 3f79bb7b435b05321651daefd374cd.yeti-dns.net \ 1405 3600000 IN AAAA 2401:c900:1401:3b:c::6 1406 . 3600000 IN NS \ 1407 xn--r2bi1c.xn--h2bv6c0a.xn--h2brj9c 1408 xn--r2bi1c.xn--h2bv6c0a.xn--h2brj9c \ 1409 3600000 IN AAAA 2001:e30:1c1e:10::333 1410 . 3600000 IN NS yeti1.ipv6.ernet.in 1411 yeti1.ipv6.ernet.in 3600000 IN AAAA 2001:e30:187d::333 1412 . 3600000 IN NS yeti-dns02.dnsworkshop.org 1413 yeti-dns02.dnsworkshop.org \ 1414 3600000 IN AAAA 2001:19f0:0:1133::53 1415 . 3600000 IN NS yeti.mind-dns.nl 1416 yeti.mind-dns.nl 3600000 IN AAAA 2a02:990:100:b01::53:0 1418 Appendix B. Yeti-Root Server Priming Response 1420 Here is the reply of a Yeti root name server to a priming request. 1421 The authoritative server runs NSD. 1423 ... 1424 ;; Got answer: 1425 ;; ->>HEADER<<- opcode: QUERY, status: NOERROR, id: 62391 1426 ;; flags: qr aa rd; QUERY: 1, ANSWER: 26, AUTHORITY: 0, ADDITIONAL: 7 1427 ;; WARNING: recursion requested but not available 1429 ;; OPT PSEUDOSECTION: 1430 ; EDNS: version: 0, flags: do; udp: 1460 1431 ;; QUESTION SECTION: 1432 ;. IN NS 1434 ;; ANSWER SECTION: 1435 . 86400 IN NS bii.dns-lab.net. 1436 . 86400 IN NS yeti.bofh.priv.at. 1437 . 86400 IN NS yeti.ipv6.ernet.in. 1438 . 86400 IN NS yeti.aquaray.com. 1439 . 86400 IN NS yeti.jhcloos.net. 1440 . 86400 IN NS yeti.mind-dns.nl. 1441 . 86400 IN NS dahu1.yeti.eu.org. 1442 . 86400 IN NS dahu2.yeti.eu.org. 1443 . 86400 IN NS yeti1.ipv6.ernet.in. 1444 . 86400 IN NS ns-yeti.bondis.org. 1445 . 86400 IN NS yeti-ns.ix.ru. 1446 . 86400 IN NS yeti-ns.lab.nic.cl. 1447 . 86400 IN NS yeti-ns.tisf.net. 1448 . 86400 IN NS yeti-ns.wide.ad.jp. 1449 . 86400 IN NS yeti-ns.datev.net. 1450 . 86400 IN NS yeti-ns.switch.ch. 1451 . 86400 IN NS yeti-ns.as59715.net. 1452 . 86400 IN NS yeti-ns1.dns-lab.net. 1454 . 86400 IN NS yeti-ns2.dns-lab.net. 1455 . 86400 IN NS yeti-ns3.dns-lab.net. 1456 . 86400 IN NS xn--r2bi1c.xn--h2bv6c0a.xn--h2brj9c. 1457 . 86400 IN NS yeti-dns01.dnsworkshop.org. 1458 . 86400 IN NS yeti-dns02.dnsworkshop.org. 1459 . 86400 IN NS 3f79bb7b435b05321651daefd374cd.yeti-dns.net. 1460 . 86400 IN NS ca978112ca1bbdcafac231b39a23dc.yeti-dns.net. 1461 . 86400 IN RRSIG NS 8 0 86400 ( 1462 20171121050105 20171114050105 26253 . 1463 FUvezvZgKtlLzQx2WKyg+D6dw/pITcbuZhzStZfg+LNa 1464 DjLJ9oGIBTU1BuqTujKHdxQn0DcdFh9QE68EPs+93bZr 1465 VlplkmObj8f0B7zTQgGWBkI/K4Tn6bZ1I7QJ0Zwnk1mS 1466 BmEPkWmvo0kkaTQbcID+tMTodL6wPAgW1AdwQUInfy21 1467 p+31GGm3+SU6SJsgeHOzPUQW+dUVWmdj6uvWCnUkzW9p 1468 +5en4+85jBfEOf+qiyvaQwUUe98xZ1TOiSwYvk5s/qiv 1469 AMjG6nY+xndwJUwhcJAXBVmGgrtbiR8GiGZfGqt748VX 1470 4esLNtD8vdypucffem6n0T0eV1c+7j/eIA== ) 1472 ;; ADDITIONAL SECTION: 1473 bii.dns-lab.net. 86400 IN AAAA 240c:f:1:22::6 1474 yeti.bofh.priv.at. 86400 IN AAAA 2a01:4f8:161:6106:1::10 1475 yeti.ipv6.ernet.in. 86400 IN AAAA 2001:e30:1c1e:1::333 1476 yeti.aquaray.com. 86400 IN AAAA 2a02:ec0:200::1 1477 yeti.jhcloos.net. 86400 IN AAAA 2001:19f0:5401:1c3::53 1478 yeti.mind-dns.nl. 86400 IN AAAA 2a02:990:100:b01::53:0 1480 ;; Query time: 163 msec 1481 ;; SERVER: 2001:4b98:dc2:45:216:3eff:fe4b:8c5b#53 1482 ;; WHEN: Tue Nov 14 16:45:37 +08 2017 1483 ;; MSG SIZE rcvd: 1222 1485 Appendix C. Active IPv6 Prefixes in Yeti DNS testbed 1486 +----------------------+---------------------------------+----------+ 1487 | Prefix | Originator | Location | 1488 +----------------------+---------------------------------+----------+ 1489 | 240c::/28 | BII | CN | 1490 | 2001:6d0:6d06::/48 | MSK-IX | RU | 1491 | 2001:1488::/32 | CZ.NIC | CZ | 1492 | 2001:620::/32 | SWITCH | CH | 1493 | 2001:470::/32 | Hurricane Electric, Inc. | US | 1494 | 2001:0DA8:0202::/48 | BUPT6-CERNET2 | CN | 1495 | 2001:19f0:6c00::/38 | Choopa, LLC | US | 1496 | 2001:da8:205::/48 | BJTU6-CERNET2 | CN | 1497 | 2001:62a::/31 | Vienna University Computer | AT | 1498 | | Center | | 1499 | 2001:67c:217c::/48 | AFNIC | FR | 1500 | 2a02:2478::/32 | Profitbricks GmbH | DE | 1501 | 2001:1398:4::/48 | BII | CN | 1502 | 240c::/28 | NIC Chile | CL | 1503 | 2001:4490:dc4c::/46 | NIB (National Internet | IN | 1504 | | Backbone) | | 1505 | 2001:4b98::/32 | Gandi | FR | 1506 | 2a02:aa8:0:2000::/52 | T-Systems-Eltec | ES | 1507 | 2a03:b240::/32 | Netskin GmbH | CH | 1508 | 2801:1a0::/42 | Universidad de Ibague | CO | 1509 | 2a00:1cc8::/40 | ICT Valle Umbra s.r.l. | IT | 1510 | 2a02:cdc0::/29 | ORG-CdSB1-RIPE | IT | 1511 +----------------------+---------------------------------+----------+ 1513 Appendix D. Tools developed for Yeti DNS testbed 1515 Various tools were developed to support the Yeti DNS testbed, a 1516 selection of which are described briefly below. 1518 YmmV ("Yeti Many Mirror Verifier") is designed to make it easy and 1519 safe for a DNS administrator to capture traffic sent from a resolver 1520 to the Root Server System and to replay it towards Yeti-Root servers. 1521 Responses from both systems are recorded and compared, and 1522 differences are logged. See . 1524 PcapParser is a module used by YmmV which reassembles fragmented IPv6 1525 datagrams and TCP segments from a PCAP archive and extracts DNS 1526 messages contained within them. See . 1529 DNS-layer-fragmentation implements DNS proxies that perform 1530 application-level fragmentation of DNS messages, based on 1531 [I-D.muks-dns-message-fragments]. The idea with these proxies is to 1532 explore splitting DNS messages in the protocol itself, so they will 1533 not by fragmented by the IP layer. See . 1536 DNS_ATR is an implementation of DNS ATR, as described in 1537 [I-D.song-atr-large-resp]. DNS_ATR acts as a proxy between resolver 1538 and authoritative servers, forwarding queries and responses as a 1539 silent and transparent listener. Responses that are larger than a 1540 nominated threshold (1280 octets by default) trigger additional 1541 truncated responses to be sent immediately following the large 1542 response. See . 1544 Appendix E. Controversy 1546 The Yeti DNS Project, its infrastructure and the various experiments 1547 that have been carried out using that infrastructure, have been 1548 described by people involved in the project in many public meetings 1549 at technical venues since its inception. The mailing lists using 1550 which the operation of the infrastructure has been coordinated are 1551 open to join, and their archives are public. The project as a whole 1552 has been the subject of robust public discussion. 1554 Some commentators have expressed concern that the Yeti DNS Project 1555 is, in effect, operating an alternate root, challenging the IAB's 1556 comments published in [RFC2826]. Other such alternate roots are 1557 considered to have caused end-user confusion and instability in the 1558 namespace of the DNS by the introduction of new top-level labels or 1559 the different use of top-level labels present in the Root Server 1560 System. The coordinators of the Yeti DNS Project do not consider the 1561 Yeti DNS Project to be an alternate root in this sense, since by 1562 design the namespace enabled by the Yeti-Root Zone is identical to 1563 that of the Root Zone. 1565 Some commentators have expressed concern that the Yeti DNS Project 1566 seeks to influence or subvert administrative policy relating to the 1567 Root Server System, in particular in the use of DNSSEC trust anchors 1568 not published by the IANA and the use of Yeti-Root Servers in regions 1569 where governments or other organisations have expressed interest in 1570 operating a Root Server. The coordinators of the Yeti-Root project 1571 observe that their mandate is entirely technical and has no ambition 1572 to influence policy directly; they do hope, however, that technical 1573 findings from the Yeti DNS Project might act as a useful resource for 1574 the wider technical community. 1576 Appendix F. About This Document 1578 This section (and sub-sections) has been included as an aid to 1579 reviewers of this document, and should be removed prior to 1580 publication. 1582 F.1. Venue 1584 The authors propose that this document proceed as an Independent 1585 Submission, since it documents work that, although relevant to the 1586 IETF, has been carried out externally to any IETF working group. 1587 However, a suitable venue for discussion of this document is the 1588 dnsop working group. 1590 Information about the Yeti DNS project and discussion relating to 1591 particular experiments described in this document can be found at 1592 . 1594 This document is maintained in GitHub at . 1597 F.2. Revision History 1599 F.2.1. draft-song-yeti-testbed-experience-00 through -03 1601 Change history is available in the public GitHub repository where 1602 this document is maintained: . 1605 F.2.2. draft-song-yeti-testbed-experience-04 1607 Substantial editorial review and rearrangement of text by Joe Abley 1608 at request of BII. 1610 Added what is intended to be a balanced assessment of the controversy 1611 that has arisen around the Yeti DNS Project, at the request of the 1612 Independent Submissions Editorial Board. 1614 Changed the focus of the document from the description of individual 1615 experiments on a Root-like testbed to the construction and 1616 motivations of the testbed itself, since that better describes the 1617 output of the Yeti DNS Project to date. In the considered opinion of 1618 this reviewer, the novel approaches taken in the construction of the 1619 testbed infrastructure and the technical challenges met in doing so 1620 are useful to record, and the RFC series is a reasonable place to 1621 record operational experiences related to core Internet 1622 infrastructure. 1624 Note that due to draft cut-off deadlines some of the technical 1625 details described in this revision of the document may not exactly 1626 match operational reality; however, this revision provides an 1627 indicative level of detail, focus and flow which it is hoped will be 1628 helpful to reviewers. 1630 F.2.3. draft-song-yeti-testbed-experience-05 1632 Added commentary on IPv6-only operation, IPv6 fragmentation, 1633 applicability to and use by IPv4-only end-users and use of multiple 1634 signers. 1636 F.2.4. draft-song-yeti-testbed-experience-06 1638 Conclusion; tools; editorial changes. 1640 Authors' Addresses 1642 Linjian Song (editor) 1643 Beijing Internet Institute 1644 2508 Room, 25th Floor, Tower A, Time Fortune 1645 Beijing 100028 1646 P. R. China 1648 Email: songlinjian@gmail.com 1649 URI: http://www.biigroup.com/ 1651 Dong Liu 1652 Beijing Internet Institute 1653 2508 Room, 25th Floor, Tower A, Time Fortune 1654 Beijing 100028 1655 P. R. China 1657 Email: dliu@biigroup.com 1658 URI: http://www.biigroup.com/ 1660 Paul Vixie 1661 TISF 1662 11400 La Honda Road 1663 Woodside, California 94062 1664 US 1666 Email: vixie@tisf.net 1667 URI: http://www.redbarn.org/ 1668 Akira Kato 1669 Keio University/WIDE Project 1670 Graduate School of Media Design, 4-1-1 Hiyoshi, Kohoku 1671 Yokohama 223-8526 1672 JAPAN 1674 Email: kato@wide.ad.jp 1675 URI: http://www.kmd.keio.ac.jp/ 1677 Shane Kerr 1678 Antoon Coolenlaan 41 1679 Uithoorn 1422 GN 1680 NL 1682 Email: shane@time-travellers.org