Re: [PATCH v3 4/8] mm, page_alloc: count movable pages when stealing from pageblock

[Vlastimil Babka]

On 03/16/2017 02:53 AM, Joonsoo Kim wrote:
> On Tue, Mar 07, 2017 at 02:15:41PM +0100, Vlastimil Babka wrote:
>> When stealing pages from pageblock of a different migratetype, we count how
>> many free pages were stolen, and change the pageblock's migratetype if more
>> than half of the pageblock was free. This might be too conservative, as there
>> might be other pages that are not free, but were allocated with the same
>> migratetype as our allocation requested.
> 
> I think that too conservative is good for movable case. In my experiments,
> fragmentation spreads out when unmovable/reclaimable pageblock is
> changed to movable pageblock prematurely ('prematurely' means that
> allocated unmovable pages remains). As you said below, movable allocations
> falling back to other pageblocks don't causes permanent fragmentation.
> Therefore, we don't need to be less conservative for movable
> allocation. So, how about following change to keep the criteria for
> movable allocation conservative even with this counting improvement?
> 
> threshold = (1 << (pageblock_order - 1));
> if (start_type == MIGRATE_MOVABLE)
>         threshold += (1 << (pageblock_order - 2));
> 
> if (free_pages + alike_pages >= threshold)
>         ...

That could help, or also not. Keeping more pageblocks marked as unmovable also
means that more unmovable allocations will spread out to them all, even if they
would fit within less pageblocks. MIGRATE_MIXED was an idea to help in this
case, as truly unmovable pageblocks would be preferred to the mixed ones.

Can't decide about such change without testing :/

> Thanks.
>