Subject: Re: Consistent vnlock dead locks in low memory conditions
To: Stephen Jones <firstname.lastname@example.org>
From: Manuel Bouyer <email@example.com>
Date: 06/23/2006 00:26:00
On Thu, Jun 22, 2006 at 02:57:30PM -0700, Stephen Jones wrote:
> On Jun 22, 2006, at 2:15 PM, Manuel Bouyer wrote:
> >Depends on RAM available, and other things I guess
> >>Would that really cause a deadlock?
> >in my experience, yes.
> What sort of 'other things'
Network activity for example, and more generally all things that want
to allocate memory in kernel.
> and can you please tell me more about
> your experiences?
Well, I don't have much. I just noticed that if maxvnode is set too high,
the kernel ends up deadlocking, with free memory near 0.
> >If you didn't change the default value then it's probably OK.
> They usually all show up as allocated in systat within a few hours of
> system running no matter what the setting is set to. Is that
> probably OK too?
Yes, this is OK.
> >>BUFCACHE I have set to 6 (%) in the kernel, can that be set
> >>via sysctl as well?
> >Yes, it's vm.bufcache.
> What I really meant to ask is it dynamic though? Can it only be
> increased without a reboot
> or drop to single user mode?
I think it can be decreased as well. But 6% isn't that much
> >If it's a NFS server it doens't matter; this is for clients.
> This particular server is an NFS client as well. I typically see
> vnlock deadlocks
> on clients and rarely on a server. However, this case is a bit of
> exception as it only
> has one served file system and its nfsd are the first to deadlock.
Hum, I didn't notice this on clients (although I've seen other issues for
which I have PR open).
Manuel Bouyer <firstname.lastname@example.org>
NetBSD: 26 ans d'experience feront toujours la difference