On Thu, Jun 22, 2006 at 09:59:30AM -0700, Stephen Jones wrote:
> I have a file server that used to operate fairly reliably with 1GB of  
> RAM.  A few weeks ago the memory tested bad and the system would not  
> boot, however the only known spares available were two 128MB modules  
> giving me 256MB of ram.  Since bringing the server up with this  
> amount of ram, I am able to achieve multiple vnlock-deadlocks
> scenarios on a daily basis.  These cannot be recovered from and the  
> system must be hardware reset.  The kernel is 2.1 and the a single  
> file system is being served via NFS with imapd, teapop and sendmail  
> processes accessing it.  It appears that the nfsd processes are the  
> first to vnlock deadlock, which then cascade and are unstoppable.  An
> interesting thing to note is that the system seems stable when it is  
> freshly booted and barely using swap, but
> once it gets deeper into swap the dead lock becomes eminent.  Since  
> this problem only occurred every 2 to 3 weeks
> with 1GB of ram and now is happening 2 to 3 times a day with 256MB of  
> ram, I thought it might be a good time to start having a kernel  
> expert investigate.

Did you do some tuning ? Large buffer cache, or change uvm settings ?
A too high kern.maxvnodes can cause this, for example.

-- 
Manuel Bouyer <bouyer@antioche.eu.org>
     NetBSD: 26 ans d'experience feront toujours la difference
--