Subject: Re: Initial RTO too big in TCP? No PSH flag in FINs?
To: Lipinski, Michal <Michal.Lipinski@intel.com>
From: Mark Allman <mallman@grc.nasa.gov>
List: tech-net
Date: 04/10/2001 10:08:55
> I've noticed that the initial RTO (retransmission timeout) in TCP
> is set to 6 seconds in NetBSD 1.x. Whereas two documents RFC-1122
> and RFC-2988, state that RTO should be initialized to 3
> seconds. Why is NetBSD different?

Good point.  

6 seconds is not non-standard as it is more conservative than the
standard allows, which is explicitly allowed.  However, I think an
initial RTO closer to 3 seconds is probably the way to go.  I looked
at the question of whether 3 seconds was a reasonable choice in the
following paper:

    Mark Allman. A Web Server's View of the Transport Layer. ACM
    Computer Communication Review, 30(5), October 2000. 
    http://roland.grc.nasa.gov/~mallman/papers/webobs-ccr.ps

The conclusion from the paper is:

    An additional note about the RTT distribution is that slightly over
    2\% of the connections in our dataset observed at least one RTT over
    3~seconds.  Furthermore, slightly more than 1\% of the connections
    averaged RTTs of over 3~seconds.  This indicates that TCP's minimum
    initial retransmission timeout (RTO) of 3~seconds as specified in
    \cite{rfc1122,rfc2988} continues to be a conservative choice.

allman


---
Mark Allman -- BBN/NASA GRC -- http://roland.grc.nasa.gov/~mallman/