Subject: Re: nfs performance
To: Jan Schaumann <jschauma@netmeister.org>
From: Greg A. Woods <woods@weird.com>
List: netbsd-users
Date: 11/14/2003 12:09:43
[ On Thursday, November 13, 2003 at 22:04:57 (-0500), Jan Schaumann wrote: ]
> Subject: nfs performance
>
> I think I'm finally fed up having to rely on a commercial vendor for our
> main file server (currently an Irix machine), and given that it's about
> 8 years old, I'll be looking at a replacement at some point in the
> not-too-distant future anyway.  Now I would really like to replace it
> with a NetBSD machine, but I seem to recall that supposedly NetBSD does
> not perform as well as an NFS server compared to Irix.

NetBSD-1.6 should perform relatively well as a NFS server, provided you
choose a decent hardware platform with a good and sufficiently fast disk
subsystem, as well as choosing well supported and decently performing
network interface adapter(s).

> Does anybody out there have any reasonable benchmark tests or any other
> input wrt to NetBSD's NFS performance?

The Postmark benchmark (in NetBSD's pkgsrc under benchmarks/postmark)
can be tuned to emulate many types of filesystem access.  I've used it
with good results as a sanity check of NFS function and performance.

Postmark is a reasonably portable C program and it should be easy enough
to port to whichever client system(s) you have in production.  If you
normally have a day of "downtime" per week (e.g. a weekend day) then you
could perhaps swap in a test server on your network and run postmark
simultaneously on as many as your existing client systems as possible,
then re-run the tests on the current production server for comparison.
To really excercise NFS and predict scalability you'd need to run
postmark on varying numbers of clients, culminating with as many
simultaneous clients as you could manage.

-- 
						Greg A. Woods

+1 416 218-0098                  VE3TCP            RoboHack <woods@robohack.ca>
Planix, Inc. <woods@planix.com>          Secrets of the Weird <woods@weird.com>