CURRENT_MEETING_REPORT_

Reported by Robert Braden/Information Sciences Institute

Minutes of the Real-time Packet Forwarding and Admission
Control BOF (REALTIME)

The demand for multimedia communication, and the success of IETF
audio/videocasts, will soon create an urgent requirement for resource
reservation and control in the Internet.  From an architectural
viewpoint, this represents a new Internet service model.  Such a service
model should include, in an integrated fashion, both real-time and
link-sharing services along with the traditional best-effort datagram
services.  Research in DARTnet has developed (a) an integrated service
model for the Internet, and (b) a particular set of mechanisms to
realize this model.

To provide end-to-end service suitable for realtime applications, the
routers must all implement the same service model, although there may be
alternative mechanisms.  The group therefore proposes that the service
model be standardized.  This BOF will describe the service model and the
realization, and suggest the service model as a candidate for Internet
standardization.

The session was opened by Scott Shenker, who noted that Bob Braden, one
of the BOF chairs, could not be present due to concerns about his home
in the current LA fires.

Scott delivered a talk on the necessity of explicit service models in
the Internet.  He described the taxonomy of service models that are
proposed in the related Internet-Draft.  He stressed that this proposal
is not for a closed set of service models, but for a growing set of
explicit models.  He also mentioned dissenting opinions from the loyal
opposition, who would prefer an approach without explicit service models
or explicit service reservations.

Scott continued by presenting and discussing three fundamental
questions.


  1. Why do we need a service model?  The answer is efficiency.  He
     argued that implementing an explicit mechanism is more effective
     than expending bandwidth in improving overall service, and this
     mechanism can only be introduced in the context of a model that
     defines the objective of the mechanism.

  2. Should the service model be explicit or implicit?  He advocated
     that binding between application and service should be determined
     by the application, outside the network, rather than internally to
     the network.

  3. Is admission control necessary?  He argued that, at least for heavy
     users, occasional blocking is a much more economical approach than
     vast over-provisioning.


He then took a straw poll which, while informal, suggested that there
was not wide dissension to his conclusions to the above questions.

A discussion of service models followed.  There was support for the
concept of admission control.

David Clark then gave a talk on the way in which routers must be
constructed in order to realize the service models discussed earlier.
He used the guaranteed real-time service as an example of behavioral
characterization of router functionality.

More discussion followed.  It was observed that behavioral
characterization of functionality is a very difficult intellectual
problem, and that it was important that the community not get bogged
down in this exercise.  We must start to implement and deploy routers
and get real experience at the same time we work towards a formal
performance characterization.

It was observed that while the presentation had emphasized router
requirements, there will also be a need for ``subnetwork requirements''
or ``link level requirements.''  In the past, the Internet has demanded
relatively little of its subnet technology, but these QOS requirements
will change that.  ATM, in particular, must fit into this architecture
in a harmonious way.

The BOF ended at 9:30, due to exhaustion of all parties.  There was an
informal assessment that a working group ought to be formed.


Attendees

Anthony Alles            aalles@cisco.com
Robert Braden            braden@isi.edu
Scott Brim               Scott_Brim@cornell.edu
Theodore Brunner         tob@thumper.bellcore.com
Stephen Casner           casner@isi.edu
John Chang               jrc@uswest.com
David Clark              ddc@lcs.mit.edu
Ed Ellesson              ellesson@vnet.ibm.com
Roger Fajman             raf@cu.nih.gov
Mark Garrett             mwg@faline.bellcore.com
Robert Gilligan          Bob.Gilligan@Eng.Sun.Com
Daniel Grossman          dan@merlin.dev.cdx.mot.com
Robert Hinden            hinden@eng.sun.com
Phil Irey                pirey@relay.nswc.navy.mil
Van Jacobson             van@ee.lbl.gov
Ronald Jacoby            rj@sgi.com
Frank Kastenholz         kasten@ftp.com
Mark Laubach             laubach@hpl.hp.com
Bryan Lyles              lyles@parc.xerox.com
Dan Magorian             magorian@ni.umd.edu
Andrew Malis             malis@maelstrom.timeplex.com
David Marlow             dmarlow@relay.nswc.navy.mil
Thomas Maslen            maslen@eng.sun.com
Keith McCloghrie         kzm@hls.com
Greg Minshall            minshall@wc.novell.com
Karen O'Donoghue         kodonog@relay.nswc.navy.mil
Steve Parker             sparker@ossi.com
Drew Perkins             ddp@fore.com
Allyn Romanow            allyn.romanow@eng.sun.com
Hal Sandick              sandick@vnet.ibm.com
Henning Schulzrinne      hgs@research.att.com
Andrew Smith             asmith@synoptics.com
Michael Speer            michael.speer@sun.com
Abel Weinrib             abel@bellcore.com