Message-Id: <200210221211.g9MCBuF01138@masaka.cs.ohiou.edu> From: "Oriana Riva" <oriana.riva@cs.helsinki.fi> Subject: Re: tcptrace initial window parameter Date: Tue, 22 Oct 2002 14:40:03 +0300
Hi!
Thanks for your suggestions.
I have finally arrived at the conclusion that the initial window parameter
in the output of tcptrace is not the real Initial Window used by TCP, but
only an estimation of that. Actually I think this is the number of packets
sent by the TCP sender, at the beginning of the connection, without
receiving any acknowledgments, so a sort of estimation of this parameter.
What do you think?
I have sent 4 TCP flows (generated with TTCP) and in one of them the
initial window indicated by the tcptrace was 1 packet. Analysing the
connections I have seen that the connection was actually sending the first
packet and after the reception of the ACK for this, was sending 3 more
packets, without respecting the TCP Slow Start!! (If really the initial
window were 1 it would be allowed to send 2 more packets).
I don't know but the Initial Window can't change from one connection to
other, also for this reason I think that value is only an estimation.
Now I'm trying to find in the Kernel where the parameter of TCP Initial
Window is set and I suppose I will find for it a value of 3, the maximum
value I have found till now in my experiments. I hope to find this!;-) By
the way do you know where is indicated in the Kernel?
Thanks a lot for your help!
Oriana
----- Original Message -----
From: "Mark Allman" <mallman@grc.nasa.gov>
To: "Hong, Duke" <duke.hong@akamai.com>
Cc: "'Oriana Riva'" <riva@cs.helsinki.fi>; <tcptrace@tcptrace.org>
Sent: Monday, October 21, 2002 9:46 PM
Subject: Re: tcptrace initial window parameter
>
> > Where are you running the data to/from? Some TCP implementations
> > do use larger initial windows, up to 3, although 1 is the
> > standard.
>
> Actually, RFC 2581 (the standard) allows for 1 or 2 segments in the
> initial window. RFC 2414 allows for 3 and 4 segment initial windows
> as an experimental (non-standard) change.
>
> And, soon the allowed initial cwnd will be 1--4 segments (with 3 and
> 4 segment initial windows depending on segment size), per
> draft-ietf-tsvwg-initwin-04.txt which has been approved as a
> proposed standard by the IESG and is sitting in the RFC Editor's
> queue (http://www.rfc-editor.org/queue.html).
>
> allman
>
>
> --
> Mark Allman -- BBN/NASA GRC -- http://roland.grc.nasa.gov/~mallman/
>
----------------------------------------------------------------------------
To unsubscribe, send a message with body containing "unsubscribe tcptrace" to
majordomo@tcptrace.org.
This archive was generated by hypermail 2b30 : 10/22/02 EDT