Kernel GPF

[Patrick O'Callaghan]

On Tue, 2010-02-09 at 12:19 +0000, Bryn M. Reeves wrote:
> On Sun, 2010-02-07 at 17:32 -0430, Patrick O'Callaghan wrote:
> > After reducing my RAM to 2GB via the mem= boot option (see parallel
> > branch of this thread) I don't seem to be getting memory errors, but I
> > still have problems, apparently with NFS. I've posted a trace from
> > dmesg to http://fpaste.org/eEh6/
> 
> That sounds a bit like:
> 
> http://article.gmane.org/gmane.linux.nfs/30419
> 
> and:
> 
> http://article.gmane.org/gmane.linux.nfs/30437
> 
> There's a patch in the first link which I don't think is in the kernel
> you're running; could be worth a look if you are seeing this regularly.

Only a couple of times, but of course I'll keep an eye on it.

> > The scenario is that I'm using rsync to an NFS-mounted directory as a
> > backup method. I had previously tried rsyncing directly to the server
> > (an Iomega ix2-200 desktop NAS), but it's unbelievably slow in this
> > configuration. I measured 100-300Kbps doing it this way -- which would
> > take well over a day to run my initial full backup job), versus at least
> > an order of magnitude faster running rsync over NFS. I conjecture that
> > the NAS cpu just isn't up to calculating the rsync checksums fast enough
> > to keep up with a 100Mbps LAN.
> 
> Is the box under memory pressure while doing this workload with mem=2g?
> That would tend to support the idea that you're seeing the above problem
> since it's been reported to occur under low memory conditions.

Very possibly. I was (rashly) trying to run a VBox VM at the same time
and I could see more than 50% of swap was committed. Load average was
over 20!

poc