CURRENT_MEETING_REPORT_

Reported by Jim Barnes/Xylogics

Minutes of the TCP Multiplexing BOF (TMUX)


Agenda

   o Introduction
   o How we got here
   o TMux Overview and implementation experience
   o Issues from the mailing list
   o What now


Peter Cameron presented a short overview of the TMux protocol and gave a
summary of the implementation experiences to date.  TMux has been
implemented in Unix System V.4 streams and BSD 4.3 systems.  The total
number of implementations to date is six.  Peter noted that since the
interface between the IP and TCP layers is not well defined,
implementing a portable TMux module can be difficult.  A problem with
FTP traffic was also noted when there was a single FTP session.  The
implementations of TMux now do not attempt to multiplex FTP traffic.
Only Telnet and Rlogin data is multiplexed now.

The results from some performance tests simulating multiple Telnet
sessions were reviewed.  The following discussion resulted in requests
for additional performance information including perceived response
times for TMux versus non-TMux situations.  Peter took the action item
to send additional performance numbers to the mailing list.

After the presentation, a general discussion followed.  The following
significant points were raised:


   o Since we want to prevent intermediate nodes from fragmenting and
     reconstructing TMux frames, the ``do not fragment'' flag should be
     set.

   o The document needs to include an applicability statement.

   o If the TMux implementation begins to see timeouts with exactly one
     datagram in the packet (that is, there is little traffic to
     multiplex), TMux should be turned off.

   o A packet with the IP OPTIONS field is not a candidate for
     multiplexing with TMux.

   o Check the test implementations to make sure everything that was
     done to overcome an implementation problem is reflected in the
     draft document.


Discussion then moved on to consider specific points raised on the
mailing list.


   o Length field:  The consensus is that the length field should be 16
     bits.

   o Checksums:  After considerable discussion, no real consensus was
     reached, so the checksum field will stay in.


A lengthy message from Don Eastlake was posted to the mailing list just
before the BOF. The issues raised and the consensus reached during the
BOF were as follows:


   o The document is too terminal server centric.  The consensus of the
     BOF attendees was that TMux was a simple solution for a very
     specific problem.  The problem definition should remain tightly
     focussed.

   o Type of Service concerns.  TMux should ensure that all datagrams
     within the multiplexed packet have the same TOS.

   o Broadcast packets.  Only unicast addresses should be allowed.

   o Larger limit on the maximum size of TMux datagrams.  The maximum
     size of 30 will be replaced with information gained during
     implementation.  This max datagram size probably should be
     configurable.

   o Use TMux only in congested situations.  Agreed.

   o The section on security needs clarification.  Agreed.


The attendees were asked if there were any blocking issues that would
prevent TMux from being put on the standards track.  No one raised any
such issue and the consensus was that TMux could be recommended to the
IESG for further action.


Attendees

Steve Alexander          stevea@lachman.com
Jim Barnes               barnes@xylogics.com
David Borman             dab@cray.com
Peter Cameron            cameron@xylint.co.uk
Lida Carrier             lida@apple.com
Alan Clegg               abc@concert.net
David Crocker            dcrocker@mordor.stanford.edu
Chuck de Sostoa          chuckd@cup.hp.com
Marc Hasson              marc@mentat.com
Jeff Hilgeman            jeffh@apertus.com
William Kwan             kwan@rabbit.com
David Lapp               lapp@waterloo.hp.com
Walter Lazear            lazear@gateway.mitre.org
Allison Mankin           mankin@cmf.nrl.navy.mil
Matt Mathis              mathis@psc.edu
Marjo Mercado            marjo@cup.hp.com
William Palter           palter@tgv.com
Kanan Shah               kshah@cmf.nrl.navy.mil
Kitty Shih               kmshih@novell.com
Hoe Trinh                htrinh@vnet.ibm.com
John Vollbrecht          jrv@merit.edu
Walter Wimer             walter.wimer@andrew.cmu.edu
Weiping Zhao             zhao@nacsis.ac.jp