Internet Info 1997 December/Internet_Info_CD-ROM_Walnut_Creek_December_1997.iso/ietf/tcplw/tcplw-minutes-91nov.txt

Text File | 1993-02-17 | 5KB | 108 lines

CURRENT_MEETING_REPORT_


Reported by David Borman/Cray

Minutes of the TCP Large Windows Working Group (TCPLW)

The TCP Large Windows Working Group met for the first time in over a
year and a half.  Since the last time the group met, RFCs 1072 and 1185
had been proposed as Internet Standards, but problems with the
specifications were discovered, and a new version, combining the two
documents, was drafted and modified to address the problems.  This
document was circulated for comments and discussion, and the Working
Group was scheduled to review the current status of the document, and
hopefully come up with something that could be recommended for the
standards track.

Bob Braden gave an overview of the proposed options.  (The overheads
that were presented are attached to this report.)  One of the
assumptions made when the options were being designed was that vendors
would not want to change the TCP header.

Discussion was then broken up into several segments, to allow discussion
of each proposed option, and the merits of each proposed option versus
other ways of solving the same problems.

It was agreed that the Window Scale options was well defined, and
discussion of a window scale option versus a 32 bit window option in
each packet was deferred until later in the meeting.

The timestamp option was then discussed, just as a timestamp.  The PAWS
discussion was deffered to later.  It was agreed that the definition was
sufficient.

The next item for discussion was the SACK option.  The first item of
discussion was whether the offset and length values should be 16 or 32
bits.  There was also discussion about if the SACK option needed to have
more than one SACK field.  The decisions were that the fields would be
32 bit fields; it simplifies the specification and processing of the
option, and eliminates any dependencies on the window-scale option.  It
was also felt that nothing was gained by limiting the number of SACKs in
a single SACK option.

The other item of discussion on SACK was whether SACK is advisory or a
real acknowledgement.  It was agreed that it was a real ACK; hence once
a piece of data has been SACKed, the receiver of the data has committed
to accepting the data, and the sender is free to discard its copy of the
data that it was saving for possible re-transmission.

Having approved the three options as being adequately defined, with the
agreed upon changes, the discussion turned back to window scale option
vs.  a 32 bit window in each packet, and the PAWS mechanism.  With the
caveat that the WS option would have no effect on other options (done by
expanding the SACK option to 32 bits), and that the initial value for

                                   1





the shift is recommended to be based on the size of the receivers
buffer, it was agreed to go with the window scale option.

The final discussion was on PAWS, Protection Against Wrapped Sequence.
First, the question was, did the PAWS mechanism provided adequate
protection, and was it clearly defined?  The answer was ``yes''.  The
final discussion then rested on PAWS vs.  expanding the sequence space
to 64 bits.  Since PAWS works, and 64 bit sequence space eats up option
space, and there were some concerns about the extra processing needed to
deal with a 64 bit sequence space, it was decided to accept the PAWS
method instead of expanding the sequence space.

Bob Braden will incorporate the changes into the document, and get it
published as an Internet Draft.  Editoral comments on the document are
to be sent to Bob.

Attendees

David Borman             [e-mail]
Robert Braden            [e-mail]
Randy Butler             [e-mail]
Richard Cherry           [e-mail]
Jim DeMarco              [e-mail]
Joseph Godsil            [e-mail]
Olafur Gudmundsson       [e-mail]
Kenneth Hayward          [e-mail]
Frank Heath              [e-mail]
Jean-Michael Jouanigot   [e-mail]
Darren Kinley            [e-mail]
Ron Mackey               [e-mail]
Donald Merritt           [e-mail]
Glenn Michel             [e-mail]
Sean O'Malley            [e-mail]
Manoel Rodrigues         [e-mail]
Miguel Sasson            [e-mail]
Tim Seaver               [e-mail]
John Seligson            [e-mail]
Mike Spengler            [e-mail]
Kathleen Wilde           [e-mail]
Johnathan Wilson         [e-mail]
Nancy Yeager             [e-mail]



                                   2