Subject: Re: NFS timeouts (sorta weird)
To: Peter Seebach <seebs@plethora.net>
From: John Franklin <franklin@elfie.org>
List: current-users
Date: 12/17/2002 18:31:42
On Tue, Dec 17, 2002 at 02:20:13PM -0600, Peter Seebach wrote:
> In message <Pine.NEB.4.44.0212170945010.16877-100000@elation.immanent.net>, Fre
> derick Bruckman writes:
> >I ran into a similar problem on a low-memory NetBSD-1.6K server (only
> >16mb ,while I was shopping for SIMMS to replace some I'd identified as
> >bad.) In my case, "mountd" and "rpcbind" were being swapped out
> >persistently, and evidently taking to too long to swap back in. Upping
> >"vm.execmin" to '20' took care of it. You could see from running "top"
> >on the server, if you have the same problem.
> 
> Hmm.  That's not it, but I have to retract an earlier claim.  Retrying
> makes it go.
> 
> So, "mount /usr" may fail, but if I ^C and try again, it works after a
> couple of tries.  Same with "umount /usr".  Watching tcpdump doesn't explain
> this.

I had a problem with a NetBSD 1.6 NFS server being slow to write out.
Under 1.6, the buf cache (systat bufcache) is used for both some NFS
buffers (request and answer cache ? ) which are charged to the root
partition in systat.  These same buffers are used for vnodes and
metadata buffers.

Unless you've changed it in your kernel config file (options BUFCACHE),
then it is 5% of total memory.  Your NFS server may be trying to get
enough bufs to stat the inodes and create the NFS request cache.

jf
-- 
John Franklin
franklin@elfie.org
ICBM: 3543'56"N 7853'27"W