The problem I am seeing is that when I transfer large-ish files (files
over ~40MB) from the Solaris NFS Server to the Linux NFS server, the
Solaris machine begins to stop responding on the network properly.
CPU load is not spiking, but ping times creep up almost exponentially
the more connections from the Linux NFS client.
The Linux machine is a Pentium 4 running Fedora Core 6.
The Solaris machine is a Ultra Sparc 10 running Solaris 8.
Both machines are on a 100Mbps network.
The interesting part of this is that if I go from Solaris to Solaris,
there is little to no effect on the network throughput on the Solaris
NFS server. It seems that Linux NFS has some hidden overhead that I
cannot compensate for. I've tried forcing NFS Version 2, I've tried
various different settings for the rsize and wsize options in Linux,
but nothing can get me to a satisfactory result.
Has anyone else seen this? Does anyone know of anything that I might
be missing?
--
Jason Williams
|