Experiences from Root Testbed in the Yeti DNS Project

The information below is for an old version of the document
Document Type Active Internet-Draft (individual)
Last updated 2016-05-23
Stream (None)
Intended RFC status (None)
Formats pdf htmlized (tools) htmlized bibtex
IETF conflict review conflict-review-song-yeti-testbed-experience
Stream Stream state (No stream defined)
Consensus Boilerplate Unknown
RFC Editor Note (None)
IESG IESG state I-D Exists
Telechat date
Responsible AD (None)
Send notices to (None)
Internet Engineering Task Force                                  L. Song
Internet-Draft                                                   S. Kerr
Intended status: Informational                                    D. Liu
Expires: November 24, 2016                    Beijing Internet Institute
                                                            May 23, 2016

         Experiences from Root Testbed in the Yeti DNS Project


   This document reports and discusses issues in DNS root services,
   based on experiences from the experiments in the Yeti DNS project.
   These issues include IPv6-only operation, the root DNS server naming
   scheme, DNSSEC KSK rollover, root server renumbering, multiple root
   zone signer, and so on.  This project was founded in May 2015 and has
   since built a live root DNS server system testbed with volunteer root
   server and resolver operations.

   REMOVE BEFORE PUBLICATION: Although this document is submitted as an
   independent submission, comments are welcome in the IETF DNSOP (DNS
   Operations) working group mailing list.  The source of the document
   is currently placed at GitHub [xml-file].

Status of This Memo

   This Internet-Draft is submitted in full conformance with the
   provisions of BCP 78 and BCP 79.

   Internet-Drafts are working documents of the Internet Engineering
   Task Force (IETF).  Note that other groups may also distribute
   working documents as Internet-Drafts.  The list of current Internet-
   Drafts is at http://datatracker.ietf.org/drafts/current/.

   Internet-Drafts are draft documents valid for a maximum of six months
   and may be updated, replaced, or obsoleted by other documents at any
   time.  It is inappropriate to use Internet-Drafts as reference
   material or to cite them other than as "work in progress."

   This Internet-Draft will expire on November 24, 2016.

Copyright Notice

   Copyright (c) 2016 IETF Trust and the persons identified as the
   document authors.  All rights reserved.

Song, et al.            Expires November 24, 2016               [Page 1]
Internet-DrExperiences from Root Testbed in the Yeti DNS Proj   May 2016

   This document is subject to BCP 78 and the IETF Trust's Legal
   Provisions Relating to IETF Documents
   (http://trustee.ietf.org/license-info) in effect on the date of
   publication of this document.  Please review these documents
   carefully, as they describe your rights and restrictions with respect
   to this document.  Code Components extracted from this document must
   include Simplified BSD License text as described in Section 4.e of
   the Trust Legal Provisions and are provided without warranty as
   described in the Simplified BSD License.

Table of Contents

   1.  Introduction  . . . . . . . . . . . . . . . . . . . . . . . .   2
   2.  Problem Statement . . . . . . . . . . . . . . . . . . . . . .   3
   3.  Yeti Testbed and Experiment Setup . . . . . . . . . . . . . .   5
     3.1.  Distribution Master . . . . . . . . . . . . . . . . . . .   6
       3.1.1.  Yeti root zone SOA SERIAL . . . . . . . . . . . . . .   7
       3.1.2.  Timing of Root Zone Fetch . . . . . . . . . . . . . .   7
       3.1.3.  Information Synchronization . . . . . . . . . . . . .   8
     3.2.  Yeti Root Servers . . . . . . . . . . . . . . . . . . . .   9
     3.3.  Yeti Resolvers and Experimental Traffic . . . . . . . . .   9
   4.  Experiments in Yeti Testbed . . . . . . . . . . . . . . . . .  10
     4.1.  Root Naming Scheme  . . . . . . . . . . . . . . . . . . .  11
     4.2.  Multiple-Signers with Multi-ZSK . . . . . . . . . . . . .  12
       4.2.1.  MZSK lab experiment . . . . . . . . . . . . . . . . .  12
       4.2.2.  MZSK Yeti experiment  . . . . . . . . . . . . . . . .  13
     4.3.  Root Renumbering Issue and Hint File Update . . . . . . .  13
     4.4.  DNS Fragments . . . . . . . . . . . . . . . . . . . . . .  14
     4.5.  The KSK Rollover Experiment in Yeti . . . . . . . . . . .  14
   5.  Other Technical findings and bugs . . . . . . . . . . . . . .  15
     5.1.  IPv6 fragments issue  . . . . . . . . . . . . . . . . . .  15
     5.2.  Root compression issue  . . . . . . . . . . . . . . . . .  16
     5.3.  SOA update delay issue  . . . . . . . . . . . . . . . . .  16
   6.  IANA Considerations . . . . . . . . . . . . . . . . . . . . .  17
   7.  Acknowledgments . . . . . . . . . . . . . . . . . . . . . . .  17
   8.  References  . . . . . . . . . . . . . . . . . . . . . . . . .  17
   Appendix A.  The Yeti root server in hint file  . . . . . . . . .  20
   Authors' Addresses  . . . . . . . . . . . . . . . . . . . . . . .  21

1.  Introduction

   [RFC1034] says the domain name space is a tree structure.  The top
   level of the tree for the unique identifier system is the DNS root
   system.  It has been operational for 25+ years.  It is pivotal to
   making the current Internet useful.  So it is considered somewhat
   ossified for stability reasons.  It is hard to test and implement new
   ideas evolving to a more advanced level to counter challenges like
   IPv6-only operation, DNSSEC key/algorithm rollover[RFC4986], scaling

Song, et al.            Expires November 24, 2016               [Page 2]
Internet-DrExperiences from Root Testbed in the Yeti DNS Proj   May 2016

   issues, and so on.  In order to make the test more practical, it is
   also necessary to involve users' environments which are highly
   diversified, in order to study the effect of the changes in question.

   To benefit Internet development as a whole, the Yeti Project was
   proposed to build a parallel, experimental, live IPv6 DNS root system
   to discover the limits of DNS root name service and deliver useful
   technical output.  Possible research agenda will be explored on this
   testbed, covering several aspects (but not limited to):

   o  IPv6-only operation

   o  DNSSEC key rollover

   o  Renumbering issues

   o  Scalability issues

   o  Multiple zone file signers

   Starting from May 2015, three coordinators began to build this live
   experimental environment and called for participants.  At the time of
   writing, there are 25 Yeti root servers with 14 operators, and
   experimental traffic from volunteers, universities, DNS vendors,
   mirrored traffic non-Yeti traffic, and RIPE Atlas probes.  Some
   experiments have been proposed and have been verified in lab tests.

   Note that the Yeti DNS project has complete fealty to IANA as the DNS
   name space manager.  All IANA top-level domain names will be
   precisely expressed in the Yeti DNS system, including all TLD data
   and meta-data[Root-Zone-Database].  So, the Yeti DNS project is not
   an "alternative root" in the usual sense of that term.  It is
   expected to inform the IANA community by peer-reviewed science as to
   future possibilities to consider for the IANA root DNS system.

   In order to let people know the technical activities in Yeti DNS
   project, this document reports and discusses issues on root DNS
   services, based on experiences so far from the testbed construction
   and experiments in the Yeti DNS project.

2.  Problem Statement

   Some problems and policy concerns over the DNS Root Server system
   stem from centralization from the point of view of DNS content
   consumers.  These include external dependencies and surveillance

Song, et al.            Expires November 24, 2016               [Page 3]
Internet-DrExperiences from Root Testbed in the Yeti DNS Proj   May 2016

   o  External Dependency.  Currently, there are 12 DNS Root Server
      operators for the 13 Root Server letters, with more than 500
      instances deployed globally.  Yet compared to the number of
      connected devices, AS networks, and recursive DNS servers, the
      number of root instances is far from sufficient.  Connectivity
      loss between one autonomous network and the IANA root name servers
      usually results in loss of local service within the local network,
      even when internal connectivity is perfect

   o  Surveillance risk.  Even when one or more root name server anycast
      instances are deployed locally or in a nearby network, the queries
      sent to the root servers carry DNS lookup information which
      enables root operators or other parties to analyze the DNS query
      traffic.  This is a kind of information leakage[RFC7626] which is
      to some extent not acceptable to some policy makers

   People are often told that the current root system with 13 root
   servers is not able to be extended to alleviate the above concerns,
   because it is limited to 13 by the current DNS protocol[ROOT-FAQ].
   To the best of authors' knowledge, there is no scientific evidence to
   support this assertion.  It remains an open question.

   There are some technical issues in the areas of IPv6 and DNSSEC,
   which were introduced to the DNS root server system after it was
   created.  Renumbering DNS root servers also creates some technical

   o  IPv6-only capability.  Currently some DNS servers including root
      which support both A and AAAA (IPv4 and IPv6) records still do not
      respond to IPv6 queries.  IPv6 introduces larger minimum MTU (1280
      bytes) and a different fragmentation model[RFC2460].  It is not
      clear whether DNS can survive without IPv4 (in an IPv6-only
      environment), or what the impact of IPv6-only environment
      introduces to current DNS operations especially in the DNS root
      server system.

   o  KSK rollover.  Currently, IANA rolls the ZSK every six weeks but
      the KSK has never been rolled as of writing.  Is RFC5011 [RFC5011]
      widely supported by resolvers?  How about longer key or different
      encryption algorithm?  Is the DNS packet size limitation (512 or
      1280 bytes) should be respected during KSK rollover nowadays?
      There are many issues still unknown.

   o  Renumbering issue.  It is likely that root operators may change
      their IP addresses for root servers as well.  Currently resolver
      can use priming exchange[I-D.ietf-dnsop-resolver-priming] to
      update its memory in real time.  Or it may combine out-band way to
      periodicaly get the current list of NS server of Root.  However it

Song, et al.            Expires November 24, 2016               [Page 4]
Internet-DrExperiences from Root Testbed in the Yeti DNS Proj   May 2016

      is observed root renumbering is stil a concern which need
      coordination and interference from human labour which deserves
      exploring for idealer automation.

3.  Yeti Testbed and Experiment Setup

   To use the Yeti testbed operationally, the information that is
   required for correct root name service is a matching set of the

   o  a root "hints file"

   o  the root zone apex NS record set

   o  the root zone's signing key

   o  root zone trust anchor

   Although Yeti DNS project publishes strictly IANA information for TLD
   data and meta-data, it is necessary to use a special hint file and
   replace the apex NS RRset with Yeti authority name servers, which
   will enable the resolves to find and stick to the Yeti root system.
   In addition, unless IANA was to help Yeti sign its root zone with a
   different root set, it is necessary to use a different ZSK and KSK
   (the DNSSEC trust anchor) in Yeti system.

   Below is a figure to demonstrate the topology of Yeti and the basic
   data flow, which consists of the Yeti distribution master, Yeti root
   server, and Yeti resolver:

Song, et al.            Expires November 24, 2016               [Page 5]
Internet-DrExperiences from Root Testbed in the Yeti DNS Proj   May 2016

                           |   IANA Root Zone via   |
                         +-+   F.root-servers.net   +--+
                         | +-----------+------------+  |
   +-----------+         |             |               | IANA root zone
   |    Yeti   |         |             |               |
   |  Traffic  |      +--v---+     +---v--+      +-----v+
   | Collection|      |  BII |     | WIDE |      | TISF |
   |           |      |  DM  |     |  DM  |      |  DM  |
   +---+----+--+      +------+     +-+----+      +---+--+
       ^    ^         |              |               |
       |    |         |              |               |   Yeti root zone
       |    |         v              v               v
       |    |   +------+      +------+               +------+
       |    +---+ Yeti |      | Yeti |  . . . . . .  | Yeti |
       |        | Root |      | Root |               | Root |
       |        +---+--+      +---+--+               +--+---+
       |            |             |                      |
       | pcap       ^             ^                      ^ DNS lookup
       | upload     |             |                      |
       |                   +--------------------------+
       +-------------------+      Yeti Resolvers      |
                           |     (with Yeti Hint)     |

   Figure 1.  The topology of Yeti testbed

3.1.  Distribution Master

   As shown in figure 1, the Yeti Root system takes the IANA root zone
   and performs minimal changes needed to serve the zone from the Yeti
   root servers instead of the IANA root servers.  In Yeti, this
   modified root zone is generated by the Yeti Distribution Masters
   (DM), which provide it to the Yeti root servers.

   So the generation process is:

   o  DM downloads the latest IANA root zone at a certain time

   o  DM makes modifications to change from the IANA to Yeti root

   o  DM signs the new Yeti root zone

Song, et al.            Expires November 24, 2016               [Page 6]
Internet-DrExperiences from Root Testbed in the Yeti DNS Proj   May 2016

   o  DM publishes the new Yeti root zone to Yeti root servers

   While in principle this could be done by a single DM, Yeti uses a set
   of three DMs to avoid any sense that the Yeti project is run by a
   single organization.  The three Distribution Masters (DMs) can
   independently fetch the root zone from IANA, sign it and publish the
   latest zone data to Yeti root servers.

   In the same while, these DMs coordinate their work so that the
   resulting Yeti root zone is always consistent.  There are two aspects
   of coordination between three DMs: timing and information

3.1.1.  Yeti root zone SOA SERIAL

   Consistency with IANA root zone except the apex record is one of most
   important point for the project.  As part of Yeti DM design, the Yeti
   SOA SERIAL which reflect the changes of yeti root zone is one factor
   to be considered.

   Currently IANA SOA SERIAL number for root zone is in the form of
   YYYYMMDDNN, like 2015111801.  In Yeti root system, IANA SOA SERIAL is
   directly copied in to Yeti SOA SERIAL.  So once the IANA root zone
   has changed with a new SOA SERIAL, a new version of the Yeti root
   zone is generated with the same SOA SERIAL.

   There is a case of Yeti DM operation that when a new Yeti root server
   added, DM operator change the Yeti root zone without change the SOA
   SERIAL which introduces inconsistency of Yeti root system.  To avoid
   inconsistency, the DMs hold on every changes to Yeti apex record and
   only new IANA SOA SERIAL will trigger the operation of adding these
   changes to Yeti root zone.

   An analysis of IANA convention shows IANA SOA SERIAL change 2 times
   every day (NN=00, 01).  And that since October 2007 the maximum of NN
   was 03 while 13 times it is observed that the versions with NN=02.
   So in the worst case, the changes of Yeti apex record is updated into
   Yeti root zone in less than 12 hours.

3.1.2.  Timing of Root Zone Fetch

   Yeti root system operators do not receive notify message from IANA
   when IANA root zone updates with a new SOA serial number.  So Yeti
   DMs check the root zone periodically.  At the time of writing, each
   Yeti DM checks to see if the IANA root zone has changed hourly, on
   the following schedule:

Song, et al.            Expires November 24, 2016               [Page 7]
Internet-DrExperiences from Root Testbed in the Yeti DNS Proj   May 2016

                         | DM Operator | Time    |
                         | BII         | hour+00 |
                         | WIDE        | hour+20 |
                         | TISF        | hour+40 |

   Note that Yeti DMs can check IANA root zone more frequently (every
   minute for example).  A test done by Yeti participant shows that the
   delay of IANA root zone update from the first IANA root server to
   last one is around 20 minute.  Once a Yeti DM fetch the new root
   zone, it will notify all the Yeti root server with a new SOA serial
   number.  So normally yeti root server will be notified in less than
   20 minute after new IANA root zone generated.  Ideally, if IANA DM
   notifies the Yeti DMs, Yeti root zone will be updated more timely.

3.1.3.  Information Synchronization

   Given three DMs operational in Yeti root system, it is necessary to
   prevent any inconsistency caused by human mistakes in operation.  The
   straight method is to share the same parameters to produce the Yeti
   root zone.  There parameters includes following set of files:

   o  the list of Yeti root servers, including:

      *  public IPv6 address and host name

      *  IPv6 addresses originating zone transfer

      *  IPv6 addresses to send DNS notify to

   o  the ZSKs used to sign the root

   o  the KSK used to sign the root

   o  the SERIAL when this information is active

   The theory of operation is that each DM operator runs a Git
   repository, containing files with the information needed to produce
   the Yeti root zone.  When a change is desired (such as adding a new
   server or rolling the ZSK), a DM operator updates the local Git
   repository.  A SOA SERIAL in the future is chosen for when the
   changes become active.  The DM operator then pushes the changes to
   the Git repositories of the other two DM operators.  When the SOA
   SERIAL of the root zone passes the number chosen, then the new
   version of the information is used.

Song, et al.            Expires November 24, 2016               [Page 8]
Internet-DrExperiences from Root Testbed in the Yeti DNS Proj   May 2016

3.2.  Yeti Root Servers

   In Yeti root system, authoritative servers donated and operated by
   Yeti volunteers are configured as a slave to the Yeti DM.  As the
   time of writing, there are 25 Yeti root servers distributed around
   the world, one of which use IDN as its name (see Yeti Hint file in
   Appendix A).  As one of operational research goal, all authoritative
   servers are required to work in an IPv6-only environment.  In
   addition, different from the IANA root, Yeti root server only serve
   the Yeti root zone, other than root-servers.org zone and .arpa zone.

   Since Yeti is a scientific research project, it needs to capture DNS
   traffic sent for later analysis.  Today some servers use dnscap,
   which is a DNS-specific tool to produce pcap files.  There are
   several versions of dnscap floating around; some people use the
   VeriSign one.  Since dnscap loses packets in some cases (tested on a
   Linux kernel), some people use pcapdump.  It requires the patch
   attached to this bug report [dnscap-bug-report]

   System diversity is also a requirement and observed for current 14
   Yeti root server.  Here are the results of a survey regarding the
   machine, operation system and DNS software:

   o  Machine: 11 out of 14 root server operator are using a VPS to
      provide service.

   o  OS: 6 operators use Linux (including Ubuntu, Debian, CentOS,
      ArchLinux). 5 operators use FreeBSD and 1 NetBSD. 3 other servers
      are unknown.

   o  DNS software: 18 our of 25 root server use BIND (varying from
      9.9.7 to 9.10.3). 4 of them use NSD (4.10 and 4.15).  The other 2
      servers use Knot (2.0.1 and 2.1.0).  And one use Bundy (1.2.0)

3.3.  Yeti Resolvers and Experimental Traffic

   In client side of Yeti project, there are DNS resolvers with IPv6
   support, updated with Yeti "hints" file to use the Yeti root servers
   instead of the IANA root servers and using Yeti KSK as trust anchor.
   The Yeti KSK rollover experiment is expected to change key often
   (typically every three months), it is required that resolver operator
   to configure your resolver compliant to RFC 5011 for automatic
   update.  For Yeti resolver, it is also interesting to try some
   mechanism end-system resolvers to signal to a server about their
   DNSSEC key status, like [I-D.wessels-edns-key-tag] and
   [I-D.wkumari-dnsop-trust-management] mentioned.

Song, et al.            Expires November 24, 2016               [Page 9]
Internet-DrExperiences from Root Testbed in the Yeti DNS Proj   May 2016

   Participants and volunteers are expected from individual researchers,
   labs of universities, companies and institutes, and vendors (for
   example, the DNS software implementers), developers of CPE devices &
   IoT devices, and middle box developers who can test their product and
   connect their own testbed into Yeti testbed.  Resolvers donated by
   Yeti volunteers are required to be configured with Yeti hint file and
   Yeti DNSSEC KSK.  It is required that Yeti resolver can speak both
   IPv4 and IPv6, given that not all the stub resolver and authoritative
   servers on the Internet are IPv6 capable.

   At the time of writing several universities and labs have joined us
   and contributed certain amount of traffic to Yeti testbed.  But it is
   far from the desired volume of experiment traffic.  So Yeti adopts
   two alternative ways to increase the experimental traffic in the Yeti
   testbed and check the functionality of Yeti root system.

   One approach is to mirror the real traffic by off-path method and
   reply it into Yeti testbed; this is implemented by one of the Yeti
   root server operators.  Another approach is to use some traffic
   generating tool such as RIPE Atlas probes to generate specific
   queries against Yeti servers.

4.  Experiments in Yeti Testbed

   The main goal of Yeti DNS Project is to act as an experimental
   network.  Experiments will be conducted on this network.  In order to
   make the findings that result from these experiments more rigorous,
   an experiment protocol is proposed.

   A Yeti experiment goes through four phases:

   o  Proposal.  The first step is to make a proposal.  It is discussed
      and if accepted by the Yeti participants then it can proceed to
      the next phase.

   o  Lab Test.  The next phase is to run a version of the experiment in
      a controlled environment.  The goal is to check for problems such
      as software crashes or protocol errors that may cause failures on
      the Yeti network, before putting onto the experimental network.

   o  Yeti Test.  The next phase actually running the experiment on the
      Yeti network.  Details of this will depend on the experiment.  It
      must be coordinated with the Yeti participants.

   o  Report of Findings.  When completed, a report of the findings of
      the experiment should be made.  It need not be an extensive

Song, et al.            Expires November 24, 2016              [Page 10]
Internet-DrExperiences from Root Testbed in the Yeti DNS Proj   May 2016

   In this section, we are going to introduce some experiments
   implemented and planned in Yeti project.

4.1.  Root Naming Scheme

   In root server history, the naming scheme for individual root servers
   was not fixed.  Current IANA Root server adopt [a-m].root-servers.net
   naming scheme to represent 13 servers which are labeled with letter
   from A to M.  One reason behind this naming scheme is that DNS label
   compression can be used to produce a smaller DNS response within 512
   bytes.  But in Yeti testbed there is a chance to design and test
   alternative naming schemes carrying following desired properties.

   o  All RRset of priming response from Root server should be
      authenticated with DNSSEC.  Not only the NS RR and DNSEKY, but
      also AAAA RR in additional section.  Currently root-servers.net is
      not signed.

   o  Root zone should be authoritative for root name so that AAAA RR
      will be included in prime response as authoritative results.
      Currently the IANA root zone is not authoritative for root name

   o  Support changes that produce larger response packets, in order to
      test the limits of the system as more root servers are added.

   Currently, there are two naming schemes as experiments in Yeti
   testbed.  One is to use separate and normal domains for root servers
   (Appendix A).  It intentionally produces larger packets for priming
   responses to avoid name compression efficiency.  Note that currently,
   the Yeti root has a priming response which is 1031 Bytes.The size of
   the response will increase when more servers are added into system.

   Besides, there is also a glue issue for this naming scheme in which
   the priming response dose not contain glue for all root severs.  It
   is documented as a technical findings [Yeti-glue-issue].  There are
   two approaches: one is to patch BIND 9 so that it includes the glue
   addresses in the additional section.  The other one is to add a zone
   file for each root server and answer for all of them at each Yeti
   server.  That means each Yeti root server would have a small zone
   file for "bii.dns-lab.net", "yeti-ns.wide.ad.jp", "yeti-ns.tisf.net",
   and so on.

   Another naming scheme under Yeti lab test is to use a special non-
   delegated TLD, like .yeti-dns for root server operated by BII.  The
   benefit of non-delegated TLD naming scheme are in two aspects: 1) the
   response to a priming query is protected by DNSSEC; 2) To meet
   political reason that the zone authoritative for root server is not

Song, et al.            Expires November 24, 2016              [Page 11]
Internet-DrExperiences from Root Testbed in the Yeti DNS Proj   May 2016

   delegated and belong to particular companies or organizations except

   The obvious concern of this naming scheme is the size of the signed
   response with RRSIG for each root server and optionally DNSSKEY RR.
   There is a Lab test result regarding the different size of priming
   response in Octet : 1) with no additional data, with RRISG in
   additional section , with DNSKEY+RRSIG in additional section (7 keys
   in MZSK experiment)

             | No additional data | RRSIG  | RRISG +DNSKEY  |
             | 753                | 3296   | 4004           |

   REMOVE BEFORE PUBLICATION: We found that modification of IANA root
   zone by adding new TLD is so controversial even for scientific
   purpose.  There are non-trivial discussions on this issue in yeti
   discuss mailing list, regarding the proposal .yeti-dns for root name
   or .local for new AS112 [I-D.bortzmeyer-dname-root].  It is argued
   that this kind of experiment should based on community consensus from
   technical bodies like IETF and be operated within a limited duration,
   such as 60 days.

4.2.  Multiple-Signers with Multi-ZSK

   According to the Problem statement of Yeti DNS project, more
   independent participants and operators of root system is desirable.
   As the name implies, multi-ZSK (MZSK) mode introduces different ZSKs
   sharing a single unique KSK, as opposed to the IANA root system
   (which uses a single ZSK to sign the root zone).  On the condition of
   good availability and consistency on root system, the Multi-ZSK
   proposal is designed to give each DM operator enough room to manage
   their own ZSK, by choosing different ZSK, length, duration, and so
   on; even the encryption algorithm may vary (although this may cause
   some problem with older versions of the Unbound resolver).

4.2.1.  MZSK lab experiment

   In the lab test phase, we simply setup two root servers (A and B) and
   a resolver switch between them (BIND only).  Root A and Root B use
   their own ZSK to sign the zone.  It is proved that Multi-ZSK works by
   adding multiple ZSK to the root zone.  As a result, the resolver will
   cache the key sets instead of single ZSK to validate the data no
   matter it is signed by Root A or Root B.  We also test Unbound who
   can pass the test with more than 10 DMs and 10 ZSKs.

Song, et al.            Expires November 24, 2016              [Page 12]
Internet-DrExperiences from Root Testbed in the Yeti DNS Proj   May 2016

   Although more DM and ZSK can be added into the test, adding more ZSKs
   to root zone enlarges the DNS response size for DNSKEY queries which
   may be a concern given the limitation of DNS packet size.  Current
   IANA root server operators are inclined to keep the packets size as
   small as possible.  So the number of DM and ZSK will be parameter
   which is decided based on operation experience.  In the current Yeti
   root testbed, there are 3 DMs, each with a separate ZSK.

4.2.2.  MZSK Yeti experiment

   After the lab test, the MZSK experiment is being conducted on the
   Yeti platform.  There are two phases:

   o  Phase 1.  In the first phase, we confirmed that using multiple ZSK
      works in the wild.  We insured that using the maximum number of
      ZSK continues to work in resolver side.  Here one of the DM (BII)
      created and added 5 ZSK using the existing synchronization
      mechanism.  (If all 3 ZSK are rolling then we have 6 total.  To
      get this number we add 5.)

   o  Phase 2.  In the second phase, we separate the management of the
      ZSK so that each DM will create and publish separate ZSK.  For
      this phase, modified zone generation protocol and software was
      used [Yeti-DM-Sync-MZSK], which allows the DM to sign without
      access to the private parts of all ZSK.  In this phase we roll all
      three ZSK separately.

   The MZSK experiment was finished by the end of 2016-04.  Almost
   everything appears to be working.  But there have been some findings
   [Experiment-MZSK-notes], including discovering that IPv6 fragmented
   packets are not forwarded on an Ethernet bridge with netfilter
   ip6_tables loaded on one authority server, and issue with IXFR
   falling back to AXFR due to multiple signers which will be detailed
   in a separate draft describing as a problem statement.

4.3.  Root Renumbering Issue and Hint File Update

   With the nearing renumbering of H root Server's IP address, there is
   a discussion of ways that resolvers can update their hint file.
   Traditional ways include using FTP protocol by doing a wget and using
   dig to double-check the servers' addresses manually.  Each way would
   depend on operators manual operation.  As a result, there are many
   old machines that have not updated their hint files.  As a proof,
   after done renumbering for thirteen years, there is an observation
   that the "Old J-Root" can still receive DNS query traffic

Song, et al.            Expires November 24, 2016              [Page 13]
Internet-DrExperiences from Root Testbed in the Yeti DNS Proj   May 2016

   This experiment proposal aims to find an automatic way for hint-file
   updating.  The already-completed work is a shell script tool which
   provides the function that update a hint-file in file system
   automatically with DNSSEC and trust anchor validation.

   The methodology is straightforward.  The tool first queries the NS
   list for "." domain and queries A and AAAA record for every domain on
   the NS list.  It requires DNSSEC validation for both signature and
   trust anchor for all the answers.  After getting all the answers, the
   tool compares the new hint file to the old one.  If there is a
   difference, it renames the old one with a time-stamp and replaces the
   old one with the new one.  Otherwise the tool deletes the new hint
   file and nothing will be changed.

   Note that in current IANA root system the servers named in the root
   NS record are not signed due to lack of incentive.  So the tool can
   not fully work in the production network.  In Yeti root system some
   of the names listed in the NS record are signed, which provides a
   test environment for such a proposal.

4.4.  DNS Fragments

   In consideration of new DNS protocol and operation, there is always a
   hard limit on the DNS packet size.  Take Yeti for example: adding
   more root servers, using the Yeti naming scheme, rolling the KSK, and
   Multi-ZSK increase the packet size.  The fear of large DNS packets
   mainly stem from two aspects: one is IP-fragments and the other is
   frequently falling back to TCP.

   In Yeti testbed, a mechanism is implemented which supports larger DNS
   packet working around the IP-layer fragment caused by middle box
   misbehavior (in IPv4) and IPv6 MTU limitation by splitting a single
   DNS message across multiple UDP datagrams.  This DNS fragments
   mechanism is documented in [I-D.muks-dns-message-fragments] as an
   experimental IETF draft.

4.5.  The KSK Rollover Experiment in Yeti

   The Yeti project provides a good basis to conduct a real-world
   experiment of a root KSK roll.  It is not a perfect analogy to the
   IANA root because all of the resolvers to the Yeti experiment are
   "opt-in", and are presumably run by administrators who are interested
   in the DNS and knowledgeable about it.  Still, it can inform the IANA
   root KSK roll.

   The IANA root KSK has not been rolled.  ICANN put together a design
   team to analyze the problem and make recommendations.  The design

Song, et al.            Expires November 24, 2016              [Page 14]
Internet-DrExperiences from Root Testbed in the Yeti DNS Proj   May 2016

   team put together a plan[ICANN-ROOT-ROLL].  The Yeti project can use
   this as a basis for an experimental KSK roll.  The experiment may not
   be identical, since the time-lines laid out in the current IANA plan
   are very long, and the Yeti project would like to conduct the
   experiment in a shorter time.

   The Yeti project would also like to conduct an experiment to try
   rolling the root KSK using a straightforward method, such as a
   double-DS approach outlined in [RFC6781].  If this ends up being
   adopted for the IANA root, then only a single Yeti experiment will
   need to be conducted.

   It's worthwhile to mention that in Yeti testbed there is a lesson
   when the KSK rollover was implemented at the first time.  It lasted
   for one month and has been held off afterwards.  In the first trial,
   it make old KSK inactive in one week after new key created, and
   delete it in another week, which is totally unaware of RFC5011.
   Because the hold-down timer is not correctly set in the server side,
   some clients (like Unbound) SERVFAILs (like dig without +cd) because
   the new key is still in AddPend state when old key is inactive.  The
   lesson from the first KSK trial is that both server and client should
   compliant to RFC5011 to set proper timer.

   One of the question is how can authority server know the resolver is
   ready for RFC5011.  Some drafts [I-D.wessels-edns-key-tag] and
   [I-D.wkumari-dnsop-trust-management] try to address the problem.  In
   addtion a compliant resolver implementation may fail without any
   complain if it is not carefully configured.  In the case of Unbound
   1.5.8, the key is only readable for DNS users.

5.  Other Technical findings and bugs

   Besides the experiments with specific goal and procedures, some
   unexpected bugs have been reported.  It is worthwhile to record them
   as technical findings from Yeti DNS Project.  Hopefully, these
   experiences can share and help.

5.1.  IPv6 fragments issue

   There are two cases in Yeti testbed reported that some Yeti root
   servers on VPS failed to pull the zone from Distribution Master via
   AXFR/IXFR.  Two facts have been revealed in both client side and
   server side after trouble shooting.

   One fact in client side is that some operation system on VPS can not
   handle IPv6 fragments correctly which causes failure when they are

Song, et al.            Expires November 24, 2016              [Page 15]
Internet-DrExperiences from Root Testbed in the Yeti DNS Proj   May 2016

   doing AXRF/IXFR in TCP.  The bug covers several OS and one VM
   platform (listed below).

                | OS                    | VM              |
                | NetBSD 6.1 and 7.0RC1 | VMware ESXI 5.5 |
                | FreeBSD10.0           |                 |
                | Debian 3.2            |                 |

   Another fact is from server side in which one TCP segment of AXRF/
   IXFR is fragmented in IP layer resulting in two fragmented packets.
   This weird behavior has been documented IETF
   draft[I-D.andrews-tcp-and-ipv6-use-minmtu].  It reports a situation
   that some implementations of TCP running over IPv6 neglect to check
   the IPV6_USE_MIN_MTU value when performing MSS negotiation and when
   constructing a TCP segment.  It will cause TCP MSS option set to 1440
   bytes, but IP layer will limit the packet less than 1280 bytes and
   fragment the packets which finally result two fragmented packets.

   The latter is not a technical error though, but it will cause the
   error in the former fact which deserves much attention in IPv6
   operation when VPS is already widely used.

5.2.  Root compression issue

   [RFC1035]specifies DNS massage compression scheme which allows a
   domain name in a message to be represented as either: 1) a sequence
   of labels ending in a zero octet, 2) a pointer, 3) or a sequence of
   labels ending with a pointer.  It is designed to save more room of
   DNS packet.

   However in Yeti testbed, it is found that Knot 2.0 server compresses
   even the root.  It means in a DNS message the name of root (a zero
   octet) is replaced by a pointer of 2 octets.  As well, it is legal
   but breaks some tools (Go DNS lib in this bug report) which does not
   expect such name compression for root.  Now both Knot and Go DNS lib
   have fixed that bug.

5.3.  SOA update delay issue

   It is observed one server on Yeti testbed have some bugs on SOA
   update with more than 10 hours delay.  It is running on Bundy 1.2.0
   on FreeBSD 10.2-RELEASE.  It is now fixed by checking DM's SOA status
   in regular base.  But it still need some work to find the bug in code
   path to improve the software.

Song, et al.            Expires November 24, 2016              [Page 16]
Internet-DrExperiences from Root Testbed in the Yeti DNS Proj   May 2016

6.  IANA Considerations

   This document requires no action from the IANA.

7.  Acknowledgments

   The editors fully acknowledge that this memo is based on joint work
   and discussions of many people in the mailing list of Yeti DNS
   project [Yeti-DNS-Project].  Some of them actually are co-authors of
   this memo but limited by the number of co-authors listed in the
   headline.  The people deserve the credit who help to construct the
   Yeti testbed and contribute to this document, so their effort is
   acknowledged here with a name list:

   Paul Vixie (project coordinator), Akira Kato(project coordinator),
   Tomohiro Ishihara, Antonio Prado, Stephane Bortzmeyer, Mickael
   Jouanne, Pierre Beyssac, Joao Damas, Pavel Khramtsov, Ma Yan, Otmar
   Lendl, Praveen Misra, Carsten Strotmann, Edwin Gomez, Remi Gacogne,
   Guillaume de Lafond, Yves Bovard, Hugo Salgado-Hernandez, Li Zhen,
   Daobiao Gong, Runxia Wan.

   Acknowledgment to all anonymous Yeti participants and volunteers who
   contribute Yeti resolvers to make the experimental testbed functional
   and workable.

8.  References

              "Unbound should test that auto-* files are writable",
              2016, <https://www.nlnetlabs.nl/bugs-script/

              Bortzmeyer, S., "pcaputils: IWBN to have an option to run
              a program after file rotation in pcapdump", 2009,

              "MZSK Experiment Notes", 2016, <https://github.com/shane-

              "Hintfile Auto Update", 2015, <https://github.com/BII-Lab/

Song, et al.            Expires November 24, 2016              [Page 17]
Internet-DrExperiences from Root Testbed in the Yeti DNS Proj   May 2016

              Andrews, M., "TCP Fails To Respect IPV6_USE_MIN_MTU",
              draft-andrews-tcp-and-ipv6-use-minmtu-04 (work in
              progress), October 2015.

              Bortzmeyer, S., "Using DNAME in the root for the
              delegation of special-use TLDs", draft-bortzmeyer-dname-
              root-00 (work in progress), April 2016.

              Koch, P., Larson, M., and P. Hoffman, "Initializing a DNS
              Resolver with Priming Queries", draft-ietf-dnsop-resolver-
              priming-07 (work in progress), March 2016.

              Sivaraman, M., Kerr, S., and D. Song, "DNS message
              fragments", draft-muks-dns-message-fragments-00 (work in
              progress), July 2015.

              Wessels, D., "The EDNS Key Tag Option", draft-wessels-
              edns-key-tag-00 (work in progress), July 2015.

              Kumari, W., Huston, G., Hunt, E., and R. Arends,
              "Signalling of DNS Security (DNSSEC) Trust Anchors",
              draft-wkumari-dnsop-trust-management-01 (work in
              progress), October 2015.

              "Root Zone KSK Rollover Plan", 2016,

              Wessels, D., "Thirteen Years of "Old J-Root"", 2015,

   [RFC1034]  Mockapetris, P., "Domain names - concepts and facilities",
              STD 13, RFC 1034, DOI 10.17487/RFC1034, November 1987,

   [RFC1035]  Mockapetris, P., "Domain names - implementation and
              specification", STD 13, RFC 1035, DOI 10.17487/RFC1035,
              November 1987, <http://www.rfc-editor.org/info/rfc1035>.

Song, et al.            Expires November 24, 2016              [Page 18]
Internet-DrExperiences from Root Testbed in the Yeti DNS Proj   May 2016

   [RFC2460]  Deering, S. and R. Hinden, "Internet Protocol, Version 6
              (IPv6) Specification", RFC 2460, DOI 10.17487/RFC2460,
              December 1998, <http://www.rfc-editor.org/info/rfc2460>.

   [RFC4986]  Eland, H., Mundy, R., Crocker, S., and S. Krishnaswamy,
              "Requirements Related to DNS Security (DNSSEC) Trust
              Anchor Rollover", RFC 4986, DOI 10.17487/RFC4986, August
              2007, <http://www.rfc-editor.org/info/rfc4986>.

   [RFC5011]  StJohns, M., "Automated Updates of DNS Security (DNSSEC)
              Trust Anchors", STD 74, RFC 5011, DOI 10.17487/RFC5011,
              September 2007, <http://www.rfc-editor.org/info/rfc5011>.

   [RFC6781]  Kolkman, O., Mekking, W., and R. Gieben, "DNSSEC
              Operational Practices, Version 2", RFC 6781,
              DOI 10.17487/RFC6781, December 2012,

   [RFC7626]  Bortzmeyer, S., "DNS Privacy Considerations", RFC 7626,
              DOI 10.17487/RFC7626, August 2015,

              Karrenberg, D., "DNS Root Name Server FAQ", 2007,

              "Root Zone Database",

              "XML source file of Yeti experience draft", 2016,

              "Yeti DM Synchronization for MZSK", 2016,

              "Website of Yeti DNS Project", <http://www.yeti-dns.org>.

              "Yeti Glue Issue", 2015,

Song, et al.            Expires November 24, 2016              [Page 19]
Internet-DrExperiences from Root Testbed in the Yeti DNS Proj   May 2016

Appendix A.  The Yeti root server in hint file

   REMOVE BEFORE PUBLICATION: Currently in Yeti testbed, there are cases
   that multiple servers run by single operator, like VeriSgin runs A
   and J.  It is allowed because we need more server to satisfy Yeti
   experiment requirement.  The name of those servers share common top
   domain name like yeti.eu.org, dns-lab.net, yeti-dns.net.  We
   intentionally pick five random labels (first 30 characters of
   SHA256([a-e])) to offset the effect of name compression.  According
   to the Yeti policy those servers will be reclaimed if qualified
   volunteers apply to host a Yeti server.

.                              3600000    IN   NS       bii.dns-lab.net
bii.dns-lab.net                3600000    IN   AAAA     240c:f:1:22::6
.                              3600000    IN   NS       yeti-ns.tisf.net
yeti-ns.tisf.net               3600000    IN   AAAA     2001:559:8000::6
.                              3600000    IN   NS       yeti-ns.wide.ad.jp
yeti-ns.wide.ad.jp             3600000    IN   AAAA     2001:200:1d9::35
.                              3600000    IN   NS       yeti-ns.as59715.net
yeti-ns.as59715.net            3600000    IN   AAAA     2a02:cdc5:9715:0:185:5:203:53
.                              3600000    IN   NS       dahu1.yeti.eu.org
dahu1.yeti.eu.org              3600000    IN   AAAA     2001:4b98:dc2:45:216:3eff:fe4b:8c5b
.                              3600000    IN   NS       ns-yeti.bondis.org
ns-yeti.bondis.org             3600000    IN   AAAA     2a02:2810:0:405::250
.                              3600000    IN   NS       yeti-ns.ix.ru
yeti-ns.ix.ru                  3600000    IN   AAAA     2001:6d0:6d06::53
.                              3600000    IN   NS       yeti.bofh.priv.at
yeti.bofh.priv.at              3600000    IN   AAAA     2a01:4f8:161:6106:1::10
.                              3600000    IN   NS       yeti.ipv6.ernet.in
yeti.ipv6.ernet.in             3600000    IN   AAAA     2001:e30:1c1e:1::333
.                              3600000    IN   NS       yeti-dns01.dnsworkshop.org
yeti-dns01.dnsworkshop.org     3600000    IN   AAAA     2001:1608:10:167:32e::53
.                              3600000    IN   NS       yeti-ns.conit.co
yeti-ns.conit.co               3600000    IN   AAAA     2607:ff28:2:10::47:a010
.                              3600000    IN   NS       dahu2.yeti.eu.org
dahu2.yeti.eu.org              3600000    IN   AAAA     2001:67c:217c:6::2
.                              3600000    IN   NS       yeti.aquaray.com
yeti.aquaray.com               3600000    IN   AAAA     2a02:ec0:200::1
.                              3600000    IN   NS       yeti-ns.switch.ch
yeti-ns.switch.ch              3600000    IN   AAAA     2001:620:0:ff::29
.                              3600000    IN   NS       yeti-ns.lab.nic.cl
yeti-ns.lab.nic.cl             3600000    IN   AAAA     2001:1398:1:21::8001
.                              3600000    IN   NS       yeti-ns1.dns-lab.net
yeti-ns1.dns-lab.net           3600000    IN   AAAA     2001:da8:a3:a027::6
.                              3600000    IN   NS       yeti-ns2.dns-lab.net
yeti-ns2.dns-lab.net           3600000    IN   AAAA     2001:da8:268:4200::6
.                              3600000    IN   NS       yeti-ns3.dns-lab.net

Song, et al.            Expires November 24, 2016              [Page 20]
Internet-DrExperiences from Root Testbed in the Yeti DNS Proj   May 2016

yeti-ns3.dns-lab.net           3600000    IN   AAAA     2400:a980:30ff::6
.                              3600000    IN   NS       ca978112ca1bbdcafac231b39a23dc.yeti-dns.net
ca978112ca1bbdcafac231b39a23dc.yeti-dns.net           3600000    IN   AAAA     2c0f:f530::6
.                              3600000    IN   NS       3e23e8160039594a33894f6564e1b1.yeti-dns.net
3e23e8160039594a33894f6564e1b1.yeti-dns.net           3600000    IN   AAAA     2803:80:1004:63::1
.                              3600000    IN   NS       2e7d2c03a9507ae265ecf5b5356885.yeti-dns.net
2e7d2c03a9507ae265ecf5b5356885.yeti-dns.net           3600000    IN   AAAA     2400:8901:e001:39::6
.                              3600000    IN   NS       18ac3e7343f016890c510e93f93526.yeti-dns.net
18ac3e7343f016890c510e93f93526.yeti-dns.net           3600000    IN   AAAA     2a05:78c0:0:2::3:6
.                              3600000    IN   NS       3f79bb7b435b05321651daefd374cd.yeti-dns.net
3f79bb7b435b05321651daefd374cd.yeti-dns.net           3600000    IN   AAAA     2a03:f80:ed15:149:154:153:136:6
.                              3600000    IN   NS       xn--r2bi1c.xn--h2bv6c0a.xn--h2brj9c
xn--r2bi1c.xn--h2bv6c0a.xn--h2brj9c 3600000    IN   AAAA     2001:e30:1c1e:10::333
.                              3600000    IN   NS       yeti1.ipv6.ernet.in
yeti1.ipv6.ernet.in            3600000    IN   AAAA     2001:e30:187d::333

Authors' Addresses

   Linjian Song
   Beijing Internet Institute
   2508 Room, 25th Floor, Tower A, Time Fortune
   Beijing  100028
   P. R. China

   Email: songlinjian@gmail.com
   URI:   http://www.biigroup.com/

   Shane Kerr
   Beijing Internet Institute
   2/F, Building 5, No.58 Jinghai Road, BDA
   Beijing  100176

   Email: shane@biigroup.cn
   URI:   http://www.biigroup.com/

   Dong Liu
   Beijing Internet Institute
   2508 Room, 25th Floor, Tower A, Time Fortune
   Beijing  100028
   P. R. China

   Email: dliu@biigroup.com
   URI:   http://www.biigroup.com/

Song, et al.            Expires November 24, 2016              [Page 21]