Re: [PATCH] mm/vmstat: retrieve more accurate vmstat value

[Joonsoo Kim]

On Tue, Nov 24, 2015 at 09:36:09AM -0600, Christoph Lameter wrote:
> On Tue, 24 Nov 2015, Joonsoo Kim wrote:
> 
> > When I tested compaction in low memory condition, I found that
> > my benchmark is stuck in congestion_wait() at shrink_inactive_list().
> > This stuck last for 1 sec and after then it can escape. More investigation
> > shows that it is due to stale vmstat value. vmstat is updated every 1 sec
> > so it is stuck for 1 sec.
> 
> vmstat values are not designed to be accurate and are not guaranteed to be
> accurate. Comparing to specific values should not be done. If you need an
> accurate counter then please use another method of accounting like an
> atomic.

I think that maintaining duplicate counter to guarantee accuracy isn't
reasonable solution. It would cause more overhead to the system.

Although vmstat values aren't designed for accuracy, these are already
used by some sensitive places so it is better to be more accurate.
What this patch does is just adding current cpu's diff to global value
when retrieving in order to get more accurate value and this would not be
expensive. I think that it doesn't break any design principle of vmstat.

Thanks.
--
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/