Re: nfsd slows with age

On Fri, 27 Aug 2004, Bob Willcox wrote:

> On Fri, Aug 27, 2004 at 11:00:14PM +0200, Claus Guttesen wrote:
> > > I have noticed that on my NFS file server system the
> > > nfsd process
> > > gradually starts using more and more CPU time ...
> > > Stopping and restarting nfsd seems to fix things
> > back to
> > > normal.
> > >
> > > Has anyone else seen this behavior?
> >
> > I'm running a nfs-server with current as of Feb. 18'th
> > 2004, serving approx. 5 TB for some web-servers. It's
> > also acting as a rsync- and syslog-server. The
> > cpu-usage hardly goes above 10 % while doing nfs,
> > except when doing internal copying or rsyncing.
> > Longest uptime was 118 days before I took it down to
> > add more disks. Works _very_ solid (knock-knock) I
> > must say.
> >
> > Are you using tcp or udp?
>
> I'm using udp. My (rather few) clients use amd to automount stuff (home
> dirs, etc.).
>
> The server system it'self is really lightly loaded as it's for my
> home network with usually only me accessing it (via nfs to my home
> directory from my workstation). I don't know if the slowdown is related
> to activity or not, guess I could run some tests to try to determine
> this.

I'd keep an eye on automounter and rpc.statd.  Automounter on linux at
least is very irresponsible and frequent restarts cause dangling mounts
which get umount attempts every timeout interval. I haven't tried it on
FreeBSD but it may have the same misbehavior.  rpc.statd likes to grow
without bound, at least on 4.x; you may need to restart it periodically.

I've had to bounce statd once or twice on my workstation, which also
serves the ports tree to a set of build servers.

-- 
Doug White                    |  FreeBSD: The Power to Serve
dwhite_at_gumbysoft.com          |  www.FreeBSD.org