Re: [PATCH 1/5] vmstat: Implement refresh_vm_stats()

[KOSAKI Motohiro]

(6/1/12 8:24 AM), Anton Vorontsov wrote:
> This function forcibly flushes per-cpu vmstat diff counters to the
> global counters.
> 
> Note that we don't try to flush percpu pagesets, the pcp will be
> still flushed once per 3 seconds.
> 
> Signed-off-by: Anton Vorontsov<anton.vorontsov@xxxxxxxxxx>

No.

This is insane. Your patch improved vmevent accuracy a _bit_. But instead of,
decrease a performance of large systems. That's no good deal. 99% user never
uses vmevent.

MOREOVER, this patch don't solve vmevent accuracy issue AT ALL. because of,
a second is enough big to make large inaccuracy. Modern cpus are fast. Guys,
the fact is, user land monitor can't use implicit batch likes vmstat. That's
a reason why perf don't use vmstat. I suggest you see perf APIs. It may bring
you good inspiration.

--
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/