Subject: Re: Strange network performance problem with -current and 1.6_BETA1
To: Markus W Kilbinger <email@example.com>
From: Manuel Bouyer <firstname.lastname@example.org>
Date: 06/11/2002 22:21:46
On Tue, Jun 11, 2002 at 01:04:35AM +0200, Markus W Kilbinger wrote:
> >>>>> "Andrew" == Andrew Gillham <email@example.com> writes:
> Andrew> http://mail-index.netbsd.org/netbsd-announce/1998/04/30/0000.html
> >>>>> "Manuel" == Manuel Bouyer <firstname.lastname@example.org> writes:
> >> From sysctl(2)
> Manuel> tcp.cwm
> Manuel> Returns 1 if use of the
> Manuel> Hughes/Touch/Heidemann Congestion
> Manuel> Window Monitoring algorithm is
> Manuel> enabled. [...]
> >> I didn't change anything in my config relating
> >> this... So it's a new default in 1.6*?
> Manuel> Yes.
> Thanks for the details!
> But: Why is it so contra productive in my case, especially if nat
> connections are involved? It squeezes data transfer rates down to
> values where my users think it's completely broken! Or does it only
> work if _everybody_ is using it?
I'm not sure as I don't know exactly what this algorithm is, but maybe
it fails when some router on the path have too small buffer to hold a
burst (maybe your NAT, depending on the ethernet adapters it uses ?).
Maybe with some tcpdump on both sides of the NAT router a TCP expert could
understand what's happening ...
Manuel Bouyer <email@example.com>