Internet Info 1997 December/Internet_Info_CD-ROM_Walnut_Creek_December_1997.iso/ietf/intserv/intserv-minutes-94mar.txt

Text File | 1994-06-06 | 11KB | 238 lines

CURRENT_MEETING_REPORT_

Reported by Craig Partridge/BBN

Minutes of the Integrated Services Working Group (INTSERV)

This group met as a BOF in Seattle.


First Meeting

The first meeting was an organizational meeting, describing the
motivations for the group, its organization, and timeline.

The goal of the working group (which is still awaiting approval) is to
make the Internet friendly to real-time applications (e.g., multimedia
conferencing) by enhancing the Internet architecture to support
integrated services.

The proposed working group organization calls for Craig Partridge to be
the working group chair, and John Wroclawski, Scott Shenker and Dave
Clark to be vice-chairs.  It was explained that the large management
team is intended to ensure that the necessary outreach to various
affected communities is made.

The working group timeline calls for delivery of several RFCs over the
next two years.  The question was raised about whether anything with
delivery two years into the future was ``research'' and thus not within
IETF scope.  The consensus answer seemed to be yes, it sometimes takes
that long, and that this agenda was appropriately conservative (i.e., we
might finish faster) rather than wildly optimistic.


Second Meeting

This meeting was devoted to trying to investigate what requirements
integrated services might place upon IPng.  Partridge explained that
this was definitely putting the cart before the horse, given that the
group had barely even had a chance to talk about proposed integrated
services architectures, but at the request of the IPng Directorate, it
was agreed to hold a discussion.  To launch discussion, two working
group co-chairs gave talks.

John Wroclawski gave a talk about flow IDs.  One of John's key points
was that there is a range of tradeoffs involved in how to identify a
flow.  The problem is locating some state in a router that tells you
what special handling a particular packet requires.  (A series of
packets which receives special treatment is what constitutes a flow.)
John also pointed out that we already do this state lookup or
classification today -- when we look up a route.  What is making the
future different is that we will want to do several classifications on a
single packet (finding state based on issues such as security,
accounting, flow control, and destination), and that while there are
logically several lookups being done here into multiple databases, for
efficient forwarding, in practice all these virtual lookups need to be
reduced to one actual lookup.

At one extreme one can locate the state by looking at enough fields in a
packet header (e.g., IP source, destination, protocol ID, protocol
ports, TOS field, etc.).  to uniquely identify a particular flow of
packets.  This may well be what we do for IPv4, since it lacks any
specific flow IDs.  At the other extremely, there is a flow ID in each
packet which points to some state.  Another issue is what to do if,
after you look up the flow ID, you find you do not have state.  If one
examines the header, one might be able to figure out what state should
be given to the packet.  If one uses flow IDs, the flow ID can either be
a hint (if missing, you can probably puzzle out from the headers what to
do with the packet), or as Noel Chiappa suggested, it can be a ``key''
which if missing means a router should discard the packet because it
will probably do the ``wrong thing'' (e.g., route to the wrong place).

John's conclusions were:


   o It is highly desirable for IPng to support a mechanism for fast
     multi-axis classification of all packets.  This requirement is
     general, not just required for integrated services.

   o The mechanism chosen for the IPng protocol should not assume any
     particular definition of flow or choice of classification criteria.


John's talk engendered a long discussion, with various people debating
which type of flow ID semantics were really desired, whether multiple
applications' traffic should be combined into one flow, or if a single
application's traffic might be split into multiple flows and even if
flows were needed.  (For an example of multiple applications in one
flow, consider several video channels aggregated together, perhaps to
yield a flow with less variable bandwidth demands.  For an example of
multiple flows for one application, consider multi-layer encoded video,
where different receivers may only want certain layers, and thus
different layers are sent as different flows.)

Dave Clark then gave a talk based on part of the IAB retreat report that
discussed authentication issues for IPng and flows in particular.  It is
clearly undesirable for an entity ``Bob'' to be able to access the
resources reserved for a flow that has been established for ``Alice,''
provided Alice is actually using the resources.  (If Alice is not using
them, we would presumably like to see the resources available to others,
as needed.)  Dave talked about a number of ways that one could try to
inexpensively authenticate packets as belonging to a particular flow.
Dave pointed out that, unlike the needs in some security contexts, the
authentication mechanisms do not have to be perfect, they simply need to
keep unauthorized access to a manageable level.  (For details of the
various proposed mechanisms, see the IAB retreat report, available as an
Internet-Draft.)

After Dave's talk, there was some discussion of how to relate
authentication needs to the flow ID issues raised by John's talk.

At the end of the meeting, the proto-working group concluded that there
were two requirements it would like to place on IPng:


  1. That an IPng have some mechanism to locate per-datagram
     classification information (e.g., flow state) and that the
     mechanism should be consistent with forwarding at the media speeds
     expected in the future.

  2. That an IPng should have mechanisms to hinder Bob from using or
     disrupting resources that Alice has been granted and is using.


Craig Partridge was tasked to write up these points and circulate a
draft to the working group, for consideration as submission as a white
paper to the IPng Directorate.


Attendees

Brian Adamson            [e-mail]
Kevin Altis              [e-mail]
Randall Atkinson         [e-mail]
Richard Bantel           [e-mail]
William Barns            [e-mail]
Stephen Batsell          [e-mail]
Steven Berson            [e-mail]
Richard Binder           [e-mail]
David Borman             [e-mail]
Ute Bormann              [e-mail]
Robert Braden            [e-mail]
Michael Bradley          [e-mail]
David Bridgham           [e-mail]
Scott Brim               [e-mail]
Theodore Brunner         [e-mail]
Steve Buchko             [e-mail]
John Burruss             [e-mail]
Ken Carlberg             [e-mail]
John Carlson             [e-mail]
Greg Celmainis           [e-mail]
Wun Chao                 [e-mail]
Ping Chen                [e-mail]
Luo-Jen Chiang           [e-mail]
Eric Chin-Li-Sun         [e-mail]
Chris Chiotasso          [e-mail]
Richard Cogger           [e-mail]
Richard Cornetti         [e-mail]
Jon Crowcroft            [e-mail]
David Crowe              [e-mail]
Sandip Dasgupta          [e-mail]
Bruce Davie              [e-mail]
Farokh Deboo             [e-mail]
Stephen Deering          [e-mail]
Chuck deSostoa           [e-mail]
Bruce Dugan              [e-mail]
Julio Escobar            [e-mail]
Roger Fajman             [e-mail]
Robert Fink              [e-mail]
William Fink             [e-mail]
Sally Floyd              [e-mail]
Paul Francis             [e-mail]
Mark Garrett             [e-mail]
Shawn Gillam             [e-mail]
Ramesh Govindan          [e-mail]
William Haggerty         [e-mail]
Mark Handley             [e-mail]
Dimitry Haskin           [e-mail]
Marco Hernandez          [e-mail]
Greg Hill                [e-mail]
Don Hoffman              [e-mail]
Kathy Huber              [e-mail]
Jim Hughes               [e-mail]
Phil Irey                [e-mail]
Kimio Ishii              [e-mail]
Ronald Jacoby            [e-mail]
Kyungran Kang            [e-mail]
Frank Kastenholz         [e-mail]
David Katinsky           [e-mail]
John Krawczyk            [e-mail]
Padma Krishnaswamy       [e-mail]
Mark Laubach             [e-mail]
Fong-Ching Liaw          [e-mail]
Arthur Lin               [e-mail]
Paul Lu                  [e-mail]
Charles Lynn             [e-mail]
Andrew Malis             [e-mail]
Tracy Mallory            [e-mail]
Allison Mankin           [e-mail]
Matt Mathis              [e-mail]
Keith McCloghrie         [e-mail]
David Meyer              [e-mail]
Greg Minshall            [e-mail]
Steve Minzer             [e-mail]
Erik Nordmark            [e-mail]
Joseph Pang              [e-mail]
Craig Partridge          [e-mail]
Amy Pearl                [e-mail]
John Penners             [e-mail]
Drew Perkins             [e-mail]
K. K. Ramakrishnan       [e-mail]
Ram Ramanathan           [e-mail]
Robert Reschly           [e-mail]
Allyn Romanow            [e-mail]
Hal Sandick              [e-mail]
Dallas Scott             [e-mail]
Joshua Seeger            [e-mail]
Scott Shenker            [e-mail]
Ming-Jye Sheu            [e-mail]
William Simpson          [e-mail]
Henry Sinnreich          [e-mail]
Jim Solomon              [e-mail]
Michael Speer            [e-mail]
Martha Steenstrup        [e-mail]
John Stewart             [e-mail]
Dan Tappan               [e-mail]
Fumio Teraoka            [e-mail]
Susan Thomson            [e-mail]
Claudio Topolcic         [e-mail]
Aleks Totic              [e-mail]
Curtis Villamizar        [e-mail]
Abel Weinrib             [e-mail]
Rick Wilder              [e-mail]
Linda Winkler            [e-mail]
John Wroclawski          [e-mail]
Lixia Zhang              [e-mail]