Subject: Re: Patch to use a larger TCP initial window for local nets
To: Jason R Thorpe <firstname.lastname@example.org>
From: John Hawkinson <jhawk@MIT.EDU>
Date: 03/01/2003 20:53:24
Jason R Thorpe <email@example.com> wrote on Thu, 27 Feb 2003
at 13:42:46 -0800 in <20030227214246.GA3653@yeah-baby.shagadelic.org>:
> This patch adds makes TCP use a larger initial window (slow-start)
> for hosts on a local network. There seems to be little worry of
> congesting a router when on a local net :-)
You say that tongue-in-cheekedly, but of course there is always a real
worry of congesting the local link. It's a fallacy many people seem to
run into, that local links are congestion-free.
In any case, this change is consistent with RFC3390 (Oct 2002), which
explicitly permits up to a 4 segment initial window (for all traffic,
not just local), so I think this is sane. (I do think it's important
to discuss, or at least mention, impact of standards on proposed changes
we make to our TCP stack).
Would you please document this in doc/CHANGES?
Speaking of which, I'm reminded of email I sent you in November of
2001 (<200111040347.WAA05985@multics.mit.edu>), asking you to document
net.inet.tcp.delack_ticks in doc/CHANGES, as well as congestion window
monitoring. While congestion window monitoring is no longer the
default, it still some awareness from users who don't read
source-changes ("Why has my TCP changed? There's nothing in
doc/CHANGES?!"; or "What new TCP features has NetBSD added?").