Re: Dirty pages underflow on 3.14.23

[Vlastimil Babka]

On 01/08/2015 02:04 AM, Simon Kirby wrote:
> On Wed, Jan 07, 2015 at 10:48:10PM +0100, Vlastimil Babka wrote:
> 
>> On 01/07/2015 10:28 PM, Simon Kirby wrote:
>>
>> > Hmm...A possibly-related issue...Before trying this, after a fresh boot,
>> > /proc/vmstat showed:
>> > 
>> > nr_alloc_batch 4294541205
>> 
>> This can happen, and not be a problem in general. However, there was a fix
>> abe5f972912d086c080be4bde67750630b6fb38b in 3.17 for a potential performance
>> issue if this counter overflows on single processor configuration. It was marked
>> stable, but the 3.16 series was discontinued before the fix could be backported.
>> So if you are on single-core, you might hit the performance issue.
> 
> That particular commit seems to just change the code path in that case,
> but should it be underflowing at all on UP?

Yes, the code using this counter is either treating it as signed, or includes a
<0 tests. It's just the /proc output might be confusing here...

>> > Still, nr_alloc_batch reads as 4294254379 after MySQL restart, and now
>> > seems to stay up there.
>> 
>> Hm if it stays there, then you are probably hitting the performance issue. Look
>> at /proc/zoneinfo, which zone has the underflow. It means this zone will get
>> unfair amount of allocations, while others may contain stale data and would be
>> better candidates.
> 
> In this case, it has only 640MB, and there's only DMA and Normal. This is
> affecting Normal, and DMA is so small that it probably doesn't matter.
> 
> Simon-
> 
> --
> To unsubscribe, send a message with 'unsubscribe linux-mm' in
> the body to majordomo@xxxxxxxxxx  For more info on Linux MM,
> see: http://www.linux-mm.org/ .
> Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>
> 

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