Re: [Xen-devel] NUMA_BALANCING and Xen PV guest regression in 3.20-rc0

[Mel Gorman <mgorman@suse.de>]

On Mon, Feb 23, 2015 at 03:13:48PM +0000, Dario Faggioli wrote:
> Hi everyone,
> 
> On Thu, 2015-02-19 at 17:01 +0000, Mel Gorman wrote:
> > On Thu, Feb 19, 2015 at 01:06:53PM +0000, David Vrabel wrote:
> 
> > I cannot think of a reason why this would fail for NUMA balancing on bare
> > metal. The PAGE_NONE protection clears the present bit on p[te|md]_modify
> > so the expectations are matched before or after the patch is applied. So,
> > for bare metal at least
> > 
> > Acked-by: Mel Gorman <mgorman@suse.de>
> > 
> > I *think* this will work ok with Xen but I cannot 100% convince myself.
> > I'm adding Wei Liu to the cc who may have a Xen PV setup handy that
> > supports NUMA and may be able to test the patch to confirm.
> > 
> I'm not Wei, but I've been able to test a kernel with David's patch in
> the following conditions:
> 

Thanks very much, it's appreciated.

>  1. as Dom0 kernel, when Xen does not have any virtual NUMA support
>  2. as DomU PV kernel, when Xen does not have any virtual NUMA support
>  3. as DomU PV kernel, when Xen _does_ _have_ virtual NUMA support
>     (i.e., Wei's code)
> 
> Cases 1. and 2. have been, I believe, tested by David already, but
> anyways... :-)
> 
> Case 3. worked well for me, as the following commands show. In fact,
> with this in guest config file:
> 
> <SNIP>
>
> And this:
> 
>     root@test-pv:~# grep numa /proc/vmstat 
>     numa_hit 65987
>     numa_miss 0
>     numa_foreign 0
>     numa_interleave 14473
>     numa_local 58642
>     numa_other 7345
>     numa_pte_updates 596
>     numa_huge_pte_updates 0
>     numa_hint_faults 479
>     numa_hint_faults_local 420
>     numa_pages_migrated 51
> 
> So, yes, I would say this wok with Xen, is that correct, Mel?
> 

Yes, these stats indicate that NUMA balancing is active. There is no
much activity but it's there.

> I'll give it a try at running more complex stuff like 'perf bench numa'
> inside the guest and see what happens...
> 

Thanks.

-- 
Mel Gorman
SUSE Labs