Subject: Re: tcp_do_rfc1323 vs Linux - any solid understanding?
To: None <current-users@NetBSD.ORG>
From: Kevin Sullivan <ksulliva@kludge.psc.edu>
List: current-users
Date: 06/11/1996 16:18:46
Ken Hornstein (kenh@cmf.nrl.navy.mil) wrote:
: >The problems generally apply two fold when you have two NetBSD systems. One
: >NetBSD and one (for example) Solaris system are nowhere near as bad in
: >performance terms (i have no recorded benchmarks... yet) as two NetBSD
: >systems.

: I'm sorry, but this has not been my experience at all.  I first noticed
: these problems when connecting to Solaris and Irix boxes.

Actually I could believe this.  I've definately seen this sort of
thing when I had a PPP line with VJ header compression in only one
direction; TCP was not nearly as slow as when the VJ compression was
in both directions.  Likewise I could believe that Solaris understands
rfc1323 but does not use it by default.  In a NetBSD<->Solaris
connection the delays would be in one direction, while in a
NetBSD<->NetBSD connection the delays would be in both directions.
The problem is not a "sub-optimal" implementation, just a poor
default.

Turning TCP timestamps off by default sounds like a good solution.
The only possible loss I could see is when a NetBSD system is directly
connected to something faster than ethernet, and in that case the
admin could turn it back on.

	-Kevin