CURRENT_MEETING_REPORT_

Reported by Deborah Estrin/Information Sciences Institute

Minutes of the Source Demand Routing Working Group (SDR)

Tony Li started the first meeting with a walk through of the current
specification to collect comments.  There was minimal discussion.  The
specification appears stable.

USC is working on their second implementation which will be available
for distribution in November or December.  A first prototype release was
made available during the summer.

The remainder of the first session was spent discussing the agenda for
the second day's meeting and some future issues, in particular, the
relationship between route setup and reservation setup, and acquiring
information to compute SDRP routes not made available by BGP/IDRP.

The second day's discussion addressed:


   o Construction of SDRP routes

     The issue of how to evolve this functionality over time and what is
     a useful starting point was discussed as well as the issue of what
     is needed longer term:  query/search techniques, extract proposal
     for using IDRP/BGP.

   o Indirect provider extensions to IDRP

     Prior to the meeting, Yakov Rekhter proposed using SDRP to extend
     IDRP/BGP functionality to allow overriding of source-specific
     preferences.  However Yakov had to attend the ROLC meeting in
     Braden's absence and was unable to present this in detail.

   o Mapping packets to SDRP routes

     This must be done at the first SDRP speaker using local criteria
     and packet information.  There are questions about how to specify
     the criteria, policy, and how to apply it efficiently so as not to
     downgrade performance for SDRP and non-SDRP packets.

   o Open issues

     Design issues that are most open and in need of attention are
     longer term techniques for inter-domain and intra-domain SDRP route
     construction, managing feedback from SDRP route failures for use in
     route selection, and the possible use of anycast addresses.

     Open usage/deployment issues are about the best places to start
     deploying SDRP (e.g., in conjunction with upcoming real time
     experiments, perhaps?), use for other tunneling requirements, and
     the need to dispel beliefs that encapsulation can never be
     efficient.

   o Volunteers were requested to review the setup specification, finish
     the usage draft (Sue Hares and Peter Ford are currently responsible
     for this), participate in experiments (Merit and USC are currently
     participating), and participate in route construction design work.

   o A status report on the current implementation noted that the two
     unimplemented features were MTU discovery and policy checking by
     transit domains.


Several of these issues were not discussed in adequate detail because of
Deborah Estrin's sudden departure in response to the LA fires.  We will
return to them via e-mail discussions and during the March IETF.


Attendees

J. Noel Chiappa          jnc@lcs.mit.edu
Osmund deSouza           osmund.desouza@att.com
Deborah Estrin           estrin@usc.edu
Christian Huitema        Christian.Huitema@sophia.inria.fr
Tony Li                  tli@cisco.com
Yakov Rekhter            yakov@watson.ibm.com