Date:	Thu, 28 Jun 2012 17:24:25 -0400
From:	Rik van Riel <riel@...hat.com>
To:	Andrew Morton <akpm@...ux-foundation.org>
CC:	linux-mm@...ck.org, linux-kernel@...r.kernel.org,
	Mel Gorman <mel@....ul.ie>, jaschut@...dia.gov,
	minchan@...nel.org, kamezawa.hiroyu@...fujitsu.com
Subject: Re: [PATCH -mm v2] mm: have order > 0 compaction start off where
 it left

On 06/28/2012 04:59 PM, Andrew Morton wrote:
> On Thu, 28 Jun 2012 13:55:20 -0400
> Rik van Riel<riel@...hat.com>  wrote:
>
>> Order>  0 compaction stops when enough free pages of the correct
>> page order have been coalesced. When doing subsequent higher order
>> allocations, it is possible for compaction to be invoked many times.
>>
>> However, the compaction code always starts out looking for things to
>> compact at the start of the zone, and for free pages to compact things
>> to at the end of the zone.
>>
>> This can cause quadratic behaviour, with isolate_freepages starting
>> at the end of the zone each time, even though previous invocations
>> of the compaction code already filled up all free memory on that end
>> of the zone.
>>
>> This can cause isolate_freepages to take enormous amounts of CPU
>> with certain workloads on larger memory systems.
>>
>> The obvious solution is to have isolate_freepages remember where
>> it left off last time, and continue at that point the next time
>> it gets invoked for an order>  0 compaction. This could cause
>> compaction to fail if cc->free_pfn and cc->migrate_pfn are close
>> together initially, in that case we restart from the end of the
>> zone and try once more.
>>
>> Forced full (order == -1) compactions are left alone.
>
> Is there a quality of service impact here?  Newly-compactable pages
> at lower pfns than compact_cached_free_pfn will now get missed, leading
> to a form of fragmentation?

The compaction side of the zone always starts at the
very beginning of the zone.  I believe we can get
away with this, because skipping a whole transparent
hugepage or non-movable block is 512 times faster than
scanning an entire block for target pages in
isolate_freepages.

>> @@ -463,6 +474,8 @@ static void isolate_freepages(struct zone *zone,
>>   		 */
>>   		if (isolated)
>>   			high_pfn = max(high_pfn, pfn);
>> +		if (cc->order>  0)
>> +			zone->compact_cached_free_pfn = high_pfn;
>
> Is high_pfn guaranteed to be aligned to pageblock_nr_pages here?  I
> assume so, if lots of code in other places is correct but it's
> unobvious from reading this function.

Reading the code a few more times, I believe that it is
indeed aligned to pageblock size.

>> --- a/mm/internal.h
>> +++ b/mm/internal.h
>> @@ -118,8 +118,10 @@ struct compact_control {
>>   	unsigned long nr_freepages;	/* Number of isolated free pages */
>>   	unsigned long nr_migratepages;	/* Number of pages to migrate */
>>   	unsigned long free_pfn;		/* isolate_freepages search base */
>> +	unsigned long start_free_pfn;	/* where we started the search */
>>   	unsigned long migrate_pfn;	/* isolate_migratepages search base */
>>   	bool sync;			/* Synchronous migration */
>> +	bool wrapped;			/* Last round for order>0 compaction */
>
> This comment is incomprehensible :(

Agreed.  I'm not sure how to properly describe that variable
in 30 or so characters :)

It denotes whether the current invocation of compaction,
called with order > 0, has had free_pfn and migrate_pfn
meet, resulting in free_pfn being reset to the top of
the zone.

Now, how to describe that briefly?

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

Powered by blists - more mailing lists