Subject: nfs performance
To: None <firstname.lastname@example.org>
From: Arto Huusko <email@example.com>
Date: 02/10/2004 18:37:59
I've been very puzzled with my NFS performance ever since I changed the
server machine (the client has been the same all the time). Mainly,
the performance has been good, but when moving large files everything
(that is: all network traffic between client and server) stopped
for a long while, and sometimes never recovered.
So: I have a quite beefy i386 client, with VIA Rhine NIC. The server
is AlphaStation 500/266 with rtk NIC. The NICs are crap, and I suspect
that's the problem, but:
With this combination, copying over 100kb files from server to client,
and things grind to halt. I earlier used rtk NIC in the i386, which
seemed to recover after some time. The vr just halts, or I'm not
The server was previously a SparcStation 5 with 100mbit hme. It
worked fine (but the transfer speed was of course in general a
lot lower than from the alpha). And before that the server was
100MHz pentium, with the same rtk nic I'm using now in the alpha.
It worked okay too.
Now, finally fed up with this, I tweaked with my nfs client
parameters. I now mount all my NFS volumes with a read size of
2048 bytes, which seems to work snappily enough. 4096, and things
won't work anymore. Oh, and I've been using UDP all the time.
Just wondering what's the problem here? Is the fact that the
alpha can flood the poor vr/rtk NICs so badly, where the previous
servers I had weren't fast enough? Should I just buy some quality
NICs to both machines (I have one 100mbit tulip, but it doesn't
seem to do fdx reliably...)