Re: a question about the deployment of SACK and NewReno TCP (fwd)

[Lloyd Wood <l.wood@eim.surrey.ac.uk>]

end2end thread that may be of interest.

L.

<L.Wood@surrey.ac.uk>PGP<http://www.ee.surrey.ac.uk/Personal/L.Wood/>

---------- Forwarded message ----------
Date: Fri, 17 Mar 2000 14:56:01 -0500
From: David.Eckhardt@cs.cmu.edu
To: end2end-interest@ISI.EDU
Cc: Sally Floyd <floyd@aciri.org>, Mark Allman <mallman@grc.nasa.gov>,
     davide+@cs.cmu.edu
Subject: Re: a question about the deployment of SACK and NewReno TCP 

Mark Allman wrote:

>RFC 1323 extensions:
>    Window scaling:     46082 (~41%)
>    Timestamps:         41376 (~37%)
>    WS + TS:            41323 (~37%)

But is this good news?  In late 1997 I was bitten by the fact that
NetBSD tried to turn on these extensions for every TCP connection,
and that this disabled VJ header compression, leading to totally
horrible TELNET character-echo latency over my (then-14.4kbps)
modem link.  Since then "sysctl -w net.inet.tcp.rfc1323=0" has
occupied a place of honor in my /etc/rc.local.

Is header compression smarter now, or are modems fast enough that
we don't notice it isn't working?

It would be fascinating to learn what percentage of the window-scaled
TCP connections end up transferring more than, say, 4 megabytes
(0.1% of the un-scaled sequence number space), before being torn
down.

Dave Eckhardt