Re: [PATCH 1/3] mm, numa: rework do_pages_move

[Michal Hocko]

On Fri 05-01-18 14:50:04, Anshuman Khandual wrote:
> On 01/05/2018 02:44 PM, Michal Hocko wrote:
> > On Fri 05-01-18 09:22:22, Anshuman Khandual wrote:
> > [...]
> >> Hi Michal,
> >>
> >> After slightly modifying your test case (like fixing the page size for
> >> powerpc and just doing simple migration from node 0 to 8 instead of the
> >> interleaving), I tried to measure the migration speed with and without
> >> the patches on mainline. Its interesting....
> >>
> >> 					10000 pages | 100000 pages
> >> 					--------------------------
> >> Mainline				165 ms		1674 ms
> >> Mainline + first patch (move_pages)	191 ms		1952 ms
> >> Mainline + all three patches		146 ms		1469 ms
> >>
> >> Though overall it gives performance improvement, some how it slows
> >> down migration after the first patch. Will look into this further.
> > 
> > What are you measuring actually? All pages migrated to the same node?
> 
> The mount of time move_pages() system call took to move these many
> pages from node 0 to node 8. Yeah they migrated to the same node.
> 
> > Do you have any profiles? How stable are the results?
> 
> No, are you referring to perf record kind profile ? Results were
> repeating.

Yes. I am really wondering because there souldn't anything specific to
improve the situation with patch 2 and 3. Likewise the only overhead
from the patch 1 I can see is the reduced batching of the mmap_sem. But
then I am wondering what would compensate that later...

-- 
Michal Hocko
SUSE Labs