Subject: Re: NFS problem.
To: None <current-users@NetBSD.org>
From: Matthias Scheler <email@example.com>
Date: 12/10/2005 12:19:49
In article <firstname.lastname@example.org>,
email@example.com (Michael van Elst) writes:
> On the other hand, TCP isn't exactly fast with 5% packet loss either.
> The problem with some older machines however was not some arbitrary
> packet loss but a very specific packet loss caused by back-to-back
> sending of IP fragments.
I'm not sure what you mean by "back-to-back". If you are talking are
refering to packets sent in burst I agree.
> By using TCP or by reducing rsize/wsize to 1k the problem goes away.
In my case (SPARCstation LX with 10Mb/s vs. Pentium 4 with 1Gb/s) using
TCP is enough. I don't have to reduce any of the NFS transaction sizes.
Matthias Scheler http://scheler.de/~matthias/