From: Avi Kivity on
On 06/15/2010 04:55 PM, Dave Hansen wrote:
> This is a big RFC for the moment. These need a bunch more
> runtime testing.
>
> --
>
> We've seen contention in the mmu_shrink() function.

First of all, that's surprising. I tried to configure the shrinker so
it would stay away from kvm unless memory was really tight. The reason
is that kvm mmu pages can cost as much as 1-2 ms of cpu time to build,
perhaps even more, so we shouldn't drop them lightly.

It's certainly a neglected area that needs attention, though.

> This patch
> set reworks it to hopefully be more scalable to large numbers
> of CPUs, as well as large numbers of running VMs.
>
> The patches are ordered with increasing invasiveness.
>
> These seem to boot and run fine. I'm running about 40 VMs at
> once, while doing "echo 3> /proc/sys/vm/drop_caches", and
> killing/restarting VMs constantly.
>

Will drop_caches actually shrink the kvm caches too? If so we probably
need to add that to autotest since it's a really good stress test for
the mmu.

> Seems to be relatively stable, and seems to keep the numbers
> of kvm_mmu_page_header objects down.
>

That's no necessarily a good thing, those things are expensive to
recreate. Of course, when we do need to reclaim them, that should be
efficient.

We also do a very bad job of selecting which page to reclaim. We need
to start using the accessed bit on sptes that point to shadow page
tables, and then look those up and reclaim unreferenced pages sooner.
With shadow paging there can be tons of unsync pages that are basically
unused and can be reclaimed at no cost to future runtime.

--
error compiling committee.c: too many arguments to function

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo(a)vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
From: Dave Hansen on
On Wed, 2010-06-16 at 11:38 +0300, Avi Kivity wrote:
> On 06/15/2010 04:55 PM, Dave Hansen wrote:
> > These seem to boot and run fine. I'm running about 40 VMs at
> > once, while doing "echo 3> /proc/sys/vm/drop_caches", and
> > killing/restarting VMs constantly.
> >
>
> Will drop_caches actually shrink the kvm caches too? If so we probably
> need to add that to autotest since it's a really good stress test for
> the mmu.

I'm completely sure. I crashed my machines several times this way
during testing.

> > Seems to be relatively stable, and seems to keep the numbers
> > of kvm_mmu_page_header objects down.
> >
>
> That's no necessarily a good thing, those things are expensive to
> recreate. Of course, when we do need to reclaim them, that should be
> efficient.

Oh, I meant that I didn't break the shrinker completely.

> We also do a very bad job of selecting which page to reclaim. We need
> to start using the accessed bit on sptes that point to shadow page
> tables, and then look those up and reclaim unreferenced pages sooner.
> With shadow paging there can be tons of unsync pages that are basically
> unused and can be reclaimed at no cost to future runtime.

Sounds like a good next step.

-- Dave

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo(a)vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
From: Avi Kivity on
On 06/16/2010 06:03 PM, Dave Hansen wrote:
> On Wed, 2010-06-16 at 11:38 +0300, Avi Kivity wrote:
>
>> On 06/15/2010 04:55 PM, Dave Hansen wrote:
>>
>>> These seem to boot and run fine. I'm running about 40 VMs at
>>> once, while doing "echo 3> /proc/sys/vm/drop_caches", and
>>> killing/restarting VMs constantly.
>>>
>>>
>> Will drop_caches actually shrink the kvm caches too? If so we probably
>> need to add that to autotest since it's a really good stress test for
>> the mmu.
>>
> I'm completely sure.

Yes, easily seen from the code as well.

> I crashed my machines several times this way
> during testing.
>

Hopefully only with your patches applied?

I'll try to run autotest from time to time with drop_caches running in
the background. Looks like an excellent way to stress out the mmu.


--
error compiling committee.c: too many arguments to function

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo(a)vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/