Re: NMBCLUSTERS warning - sysctl fix?

[Manuel Bouyer <bouyer%antioche.eu.org@localhost>]

On Mon, Mar 24, 2008 at 08:02:58PM -0500, Donald Lee wrote:
> I got a brief hang today on my production server running NetBSD 2.1 PPC:
> 
> Mar 24 13:57:07 mercy /netbsd: WARNING: mclpool limit reached; increase 
> NMBCLUSTERS 
> Mar 24 13:58:07 mercy /netbsd: WARNING: mclpool limit reached; increase 
> NMBCLUSTERS 
> Mar 24 14:00:08 mercy last message repeated 2 times
> Mar 24 14:05:09 mercy last message repeated 5 times
> 
> In the NetBSD docs, there is some advice:
> 
>       http://netbsd.org/docs/kernel/#mclpool-limit
> 
> What I have done is set up /etc/sysctl.conf, and put:
> 
>       # Server gets busy sometimes.  Bump up network resources
>       # number of "mbuf clusters"?  Default of 1024 too small.
>       kern.mbuf.nmbclusters=4096
> 
> in it.  This seems to work fine, but I apparently have to bump it up again.
> 
> The docs (above) suggest the "sysctl -w", and also suggest patching the
> kernel.  At the end of the section it says:
> 
>       Combining this, and patching the binary, would mean no need to
>       build a new kernel or reboot.
> 
> Is there some reason that the sysctl.conf approach is insufficient?  Do I need
> to set sysctl *and* patch the kernel for some reason?

It depends on how the MCL pool is implemented on a particular architecture.
Some can grow it on the fly, come can't. I think alpha can grow
it on the fly (so the sysctl method will work) but i386/amd64 can't
(so need the kenrel patch method). I don't know in which category
macppc is though, but sysctl -w should return you an error if
it can't be grown on the fly.

-- 
Manuel Bouyer, LIP6, Universite Paris VI.           
Manuel.Bouyer%lip6.fr@localhost
     NetBSD: 26 ans d'experience feront toujours la difference
--