Some implementors do (which i think is wrong) stupid things from time
to time in the name of performance.
"wrong" and "stupid" are funny terms for open source developers to use
when talking about someones efforts to improve performance (even if you
do not agree with the procedures)...
-----Original Message-----
From: ethereal-dev-bounces@xxxxxxxxxxxx
[mailto:ethereal-dev-bounces@xxxxxxxxxxxx] On Behalf Of Pia Sahlberg
Sent: Thursday, March 25, 2004 10:36 AM
To: ethereal-dev@xxxxxxxxxxxx
Subject: RE: [Ethereal-dev] TCP/IP Retransmission
>(part of formally closing the TCP connection).
>Many TCP services do not close a connection properly so the reset
>connection is not always generated - if the TCP testing on ethereal was
>done using such a service this could cause it to be missed in the
>analyzer.
Well, technically speaking RST is NOT part of formally closing a
session.
The proper way to close a session is by using the normal 3-way FIN
handshake.
RST segments are used to indicate that an unrecoverable error has
occured and that the session is thus destroyed.
Some implementors do (which i think is wrong) stupid things from time to
time in the name of performance. Some implementations do use RST as a
quick way to shut down sessions. Broken and abusing tcp but thats the
way they do it.
Attached is a small patch to stop ethereal from flagging RST or FIN
segments as zero windows or dup acks (which they are not)
there are still bugs in the analysis:
1, basic window updates should NOT be flagged as dupacks, but that
would take more brainsurgery to fix than this simple patch above.
2, rfc2988 implementation
3, detection of partial acks !!!! (important) 4, it can not detect and
distinguish between tahoe/reno/newreno
_________________________________________________________________
Personalise your mobile chart ringtones and polyphonics. Go to
http://ringtones.com.au/ninemsn/control?page=/ninemsn/main.jsp
____________________________
This email has been scanned.
