Subject: Re: nfsd and process load sharing
To: Jason Thorpe <thorpej@nas.nasa.gov>
From: Simon Burge <simonb@netbsd.org>
List: tech-kern
Date: 06/06/1999 15:39:46
Jason Thorpe wrote:
> On Fri, 4 Jun 1999 10:00:08 +0200
> Frank van der Linden <frank@wins.uva.nl> wrote:
>
> > I could change the code to put free nfsds at the back of the queue so that
> > nfsds are used round-robin if it makes you feel happy ;-)
>
> Actually, the current scheme is much better. It allows you to see
> how many nfsds you *actually* need.
>
> I.e. Simon could probably prune it down to one or two and his system would
> still be just fine.
Since this machine's _only_ job is to NFS serve a ccd, I'll not worry
about it. I chose 20 nfsd based on an old rule of thumb that said you
use (2 x num-disks) as the number of nfsd's.
While we're talking about performance, I remember a discussion about ccd
tuning from a a while back but can't find the reference any more. This
is a 1.3.3/pmax with 10 2GB disks in a ccd and "options CCDNBUF=16" in
the config file. Here's the output of "ccdconfig -s" (the machines been
up for ~100 days) - look ok?
Statistics for ccd0:
freelist count: 160
freelist hiwater: 160
freelist misses: 1657
ccdbuf allocations: 10990878
Michael (if you're listening) - as Mr FAQ, is there a place this info
could live?
Simon.