On Tue, 20 May 2008 13:36:26 -0700
Greg Banks <gnb-cP1dWloDopni96+mSzHFpQC/[email protected]> wrote:
> Greg Banks wrote:
> > [...]
> > I'll send out the patch I mentioned when we
> > chatted at Cthon, maybe you can consider the issue it fixes.
> >
> >
> Here's the patch. You probably don't want to do it the same way, but
> the intent is to delay the allocation of the svc_rqst structure until
> the nfsd is running on the correct cpu so that the allocation comes from
> the correct local memory on NUMA systems. Absent the patch, most of the
> nfsds spend a lot of time updating remote memory allocated on node0.
>
Thanks Greg. I'll make sure I incorporate a similar fix in the next
iteration of the patch. Rather than delaying the allocation like this,
I wonder if we can just figure out the node from the cpumask and then
do a kmalloc_node()?
--
Jeff Layton <[email protected]>