TCPSAT Minutes, December 8, 1997, Washington, DC

Reported by Dan Glover (Daniel.Glover@lerc.nasa.gov)


Charts and documents are available from the TCPSAT Web page at
http://tcpsat.lerc.nasa.gov


Aaron Falk opened the meeting at 9:30am and reviewed the agenda
which was accepted without comments.  Aaron asked for a show of
hands on how many attendees were on the mailing list and estimated
about 80% were.


Aaron reviewed the status of the working group.  He stated his
intention to forward the standard mechanisms draft for consideration
as an informational RFC if consensus was reached.  The research
issues draft was not ready yet, so he presented a schedule change
showing a meeting of the working group at the next IETF meeting.
There was no comment on the proposed schedule which Aaron
interpreted as general agreement.


Mark Allman gave a presentation on the standard mechanisms draft.
He stated that he intended to provide a pointer to the PSC
implementations information Web page, but did not intend to include
details on specific implementations in the draft.


Luis Sanchez commented that security issues could be better
addressed and agreed to write a paragraph on IP security issues
related to TCP over satellite.  A question on UDP was dismissed as
out-of-scope by Aaron.  A question on LEO satellites not having high
delay*bandwidth was answered by pointing out that the draft list of
satellite characteristics was a general list and that any particular
satellite would have a subset of those characteristics.  There was a
comment that asymmetric bandwidth cases did not appear to be
addressed in detail.


There was a comment that bigger segments did not make sense in a
high error environment.  Questions were raised about Path MTU
Discovery, that there had been past problems with routers that
didn't support it, that some firewalls do not support it, that some
implementations are broken over GRE tunnels, and that the discovery
process could add significant time to a transmission.


Aaron asked if there was general consensus that the standard
mechanisms draft should be forwarded for consideration as an
informational RFC after some modification to the Path MTU Discovery
section.  There was a general nodding of heads and no dissent was
noted.


Mark then gave a presentation on the research issues draft.  He
pointed out that the first draft of the document is incomplete.  He
asked for volunteers to help write the sections that have already
been identified or any new sections using the same format as in the
present draft.


Someone volunteered to write up the benefits of header compression
for asymmetric channels.  Comments were made that on asymmetric
channels TCP rate limiting is needed to avoid bursts for byte
counting, and that ACKs could be lost on uplink.


Aaron asked for general comments on the state of the research draft.
Mark complained that the draft is nowhere near complete.


TCP for Transactions was discussed at some length.  Some felt that
the benefits of T/TCP were outweighed by the complexity of
implementation.  Someone commented that Web servers would benefit
from T/TCP, but was answered by another comment that HTTP 1.1 brings
the same benefits as persistant connections by using one connection
in place of many.  A comment was made that denial of service using
T/TCP would make it unsuitable for Web servers.  There was
discussion of the pros and cons of leaving T/TCP in the draft.


Comments were made that NACK and SCPS needed to be added to the list
of research issues in the draft and that Van Jacobson's comments
about byte counting at the last TCPSAT meeting should be addressed
in the draft.


Aaron closed the meeting.  The roster sheet had only made it around
about halfway by this point, so Aaron urged attendees to sign the
blue sheet on their way out.