Location: Sheraton Centre, Montreal, Canada
Date: Sunday, 23 June 1996
Meeting Chair: Elise Gerich & Geoff Huston
Minutes: Geoff Huston & Henry Clark
Participants:
Name Organization E-mail address ----------------------- -------------------- --------------------------- Steven Bakker DANTE S.Bakker@dante.org.uk Vincent Berkout DANTE V.Berkhout@dante.org.uk Suresh Bhogavilli ISI suresh@isi.edu Nevil Brownlee Univ of Auckland n.brownlee@auckland.ac.nz Brian Carpenter CERN brian@dxcoms.cern.ch Chris Chaundy Connect chris@connect.com.au Kilnam Chon KAIST chon@cosmos.kaist.ac.kr Kimberly Claffy NLANR kc@kasine.nlanr.net Henry Clark BBN Planet henryc@bbnplanet.com Steve Corbato Univ Washington corbato@cac.washington.edu John Curran BBN Planet jcurran@bbnplanet.com Joanne Davis Connect joanne@connect.com.au Sean Doran Sprint smd@sprint.net Havard Eidnes NORDUnet he@runit.sintef.no Johnny Eriksson EBONE bygg@sunet.se Peter Ford MCI pford@mci.net Vince Fuller BBN Planet vaf@bbnlanet.com Elise Gerich Merit epg@merit.edu Barry Greene SingTel barry@singnet.com.sg Frode Greisen EBONE frode.greisen@uni-c.dk Craig Haney Condo Consulting craig@kludge.net John Hawkinson BBN Planet jhawk@bbnplanet.com Kim Hubbard Network Solutions kimh@internic.net Geoff Huston Telstra gih@telstra.net Hugh Irvine Connect hugh@connect.com.au Marijke Kaat SURFnet Marijke-kaat@sec.nl Tamotsu Kanoh KDD kanoh@ats.sjk.kdd.co.jp Akira Kato WIDE Project kato@wide.ad.jp Sean Kennedy BBN PLanet liam@bbnplanet.com Ed Kern Digex ejk@digex.net Dorian Kim CICNet dorian@cic.net Mark Kosters Network Solutions markk@internic.net Hock-Koon Lim Pacific Internet lim@pacific.net.sg Lars-Johan Liman Ebone NOC liman@ebone.net Peter Lothberg STUPI roll@stupi.se Bill Manning ISI bmanning@isi.edu Jun Matsukata NACSIS jm@rd.nacsis.ac.jp Keith Mitchell Linx keith@linx.net Tracie Monk DynCorp/FNC tmonk@snap.org Yasumori Motonaga KDD ya-motonaga@kdd.co.jp Mike O'Dell UUNET mo@uu.net Carol Orange RIPE NCC orange@ripe,net Michael Patton BBN map@bbn.com Dave Pokorney Univ of Florida dp@ufl.edu Jon Postel ISI postel@isi.edu Laina Raveendran GETIT laina@singnet.com.sg Bernhard Stockman Telia boss@it.kth.se Duane Wessels SDSC wessels@nlanr.net Bill Woodcock Zocalo woody@zocalo.net David Woodgate Telstra davidw@telstra.net Lixia Zhang UCLA lixia@parc.xerox.com
The agenda was distributed prior to the meeting following a request for items.
The first part of the discussion was related to the logical location of root nameservers in respect to provider and exchange topology. It was noted that while the placing of a root server on an exchange was a topological neutral location both in terms of traffic flow and in terms of relation to potential inter-provider settlement structures, the location did hamper effective management of the root name service. Placing a root server within a provider network offers improved management capability and places the onus on the provider to provide high quality connectivity to the server, and was generally considered to be a more stable deployment structure.
The second part of the discussion examined the distribution of root name servers across the Internet. It was noted that protocol restrictions effectively limit the number of distinct root name servers to a maximum of 13. Current 9 root name servers are deployed (8 US, 1 EU) with the 13 slots filled by multiple IP addresses for some servers. Increasing the geographical distribution of root servers was discussed. It was noted that root traffic at one US sites was recorded as 300kbps incoming and 1Mbps outgoing. The recorded traffic levels at KTH, Sweden were lower than this with 400kbps outgoing traffic levels recorded. It was noted that the older versions of named used a strict round robin model of root server selection, imposing a high US directed traffic flow on any non-US deployed root name servers. Recent versions of named bias the round robin model with consideration of the RTT of the server, which does perform some localisation of root DNS traffic.
The concept of reuse of a single network number to deploy multiple root nameservers with routing used to determine "closeness" was discussed, and generally thought to be a high risk approach with the possibility of strange failure modes bringing down large parts of the operational DNS.
The IEPG proposed to draft the operational requirements of root name servers in an IEPG Operational Note (Action: Bill Manning).
The IEPG proposed to draft an IEPG Operational Note, proposing to IANA an experimental deployment of 2 additional name servers, with proposed locations in the UK (Linx) and Japan (WIDE Project) and proposed timing, duration and objectives of the experiment to be documented (Action: Bill Manning).
SQUID is an outcome of the Harvest activity and uses the Internet Cache Protocol (ICP) which implements a cache proxy protocol using a hierarchy of cache servers and adaptive cache referral mechanisms which take RTT into account together with total elapsed query time. The current deployment of SQUID servers includes 6 NLANR servers in operation on the vBNS, using 128Mb DEC Alpha platforms with 8G disk subsystems with a forced 3 day TTL within the cache, using a single process implementation of the cache server Scaling is being examined with the current servers experiencing 10% - 15% load levels while servicing 100,000 - 200,000 requests per day. Other issues relevant to the scaling of caches include "cache friendly" resources, cache hit referrals to the original source and copyright and liability issues. Cache hits rates are noted at 15% - 20% within the hierarchy, but this increases to 35% - 40% at the user-facing edge of the cache system.
This presentation was an overview of the work being undertaken in the IETF RTFM Working Group, and further information can be found at http://www.auckland.ac.nz/net/Internet/rtfm/TOP.html
As usual this topic generated large volumes of comment and a wide range of views, ranging from the view that this problem which required common formal settlement structures, through to informal settlement through content brokerage. Phone settlement structures were noted, and the IETF IPPM work was also noted, as well as Brian's work in attempting to define a language set and associated metrics which could be applicable to inter-provider settlements.
The group did highlight that issue that inter-provider settlement structures, if they become widely adopted will have a major impact on edge pricing to the end customer, and it was noted that this aspect of edge pricing is explored in a paper by Shenker, Clark, Estrin and Hezog (one of a number of papers located at ftp://parcftp.xeox.com/net-research)
It was noted that routing flap levels are decreasing at the core, but further effort is required to stabilise route flaps at the periphery of the network and avoid propagating the flap inward to the core. It was considered that the most effective way of undertaking this was to applying dampening of a flapping route in direct proportion to the prefix size, thus passing a flapping /8 prefix on directly while applying heavy dampening factor to a /24 prefix. One vendor's code (cisco) allows a route map to be applied to a matching prefix length, allowing this mechanism to be implemented directly.
The additional topic covered in this section was the issue of a "soft" reconfiguration for BGP, allowing BGP route prefixes to be withdrawn without resetting the BGP session (and generating a flap).
It was agreed to draft up the preferred approach to route flap dampening as an IEPG operational note, and review techniques and route flap metrics at the next IEPG meeting.
The proposed engineering is to use a dual FDDI ring as a core switch structure, and use a number of co-location points, each provisioned with access to both FDDI rings. In addition all co-location points are interconnected via an SDH ring to allow private interconnect traffic. The intention is to provision this across Stockholm using city-provided dark fibre. Annual connection fee is currently envisaged to be of the order of USD 30,000 per port per year.
The proposed organisational structure is envisaged to use the IEPG, the EOF, and the Swedish Network Users Society (SNUS) within a hierarchical organisational structure, and operate the exchange as a Working Group of SNUS, which in turn will execute contracts with providers.
The technical focus at the provider level, looking at the Internet from an international perspective was seen as valuable, and the distinction between the IETF Operational Requirements Area Directorate was seen as a distinction between the IETF ORAD generic focus on tools and techniques that may be useful to providers in terms of technology development, whereas the IEPG focus is more one of current operational imperatives in today's Internet.
The IEPG wanted to use the publication mechanism of the IEPG Operational Note as a means of disseminating operational information, and the Root DNS Server experiment proposal and Route Flap Dampening Deployment were considered to be issues which should be published as IEPG Operational Notes immediately. The role of the EOF, NANOG and APRICOT regional groups in considering operational issues of local significance were considered to be very important, allowing the IEPG to pick topics which were common across the regional operational agendas.
The group elected to hold meetings at each IETF on the Sunday prior to the IETF meeting.
--- End of Minutes