Re: [PATCH v3 00/18] periodic write-back timer optimization

[Artem Bityutskiy <Artem.Bityutskiy@nokia.com>]

Andi Kleen wrote:
> Artem Bityutskiy <Artem.Bityutskiy@nokia.com> writes:
> 
>> The patches attempt to optimize the periodic write-back and stop it when
>> there are no dirty data. IOW, we do not want the thread to wake up every
>> 5 seconds (by default), find there is nothing to do, and so on.
> 
> Is waking up every 5 seconds really a problem?

Yes, it is. In OMAP3 we may switch the core and most of peripherals
to off and we may stay in off and consume really few power. And waking
up from off every 5 seconds for no good reason is bad.

> The normal rule of thumb is iirc that longer sleep times than a few hundred
> ms give dimishing returns in terms of power saving.

I think our pm guys measured this on OMAP3 and it translates to
~hour of idle mode. And yes, this is bad for devices which
run from battery.

> A simple way might be simple to batch the timer better with other timers.

I agree. There is a lot of work in this direction. There are many
places where we could use range hrtimers or deferrable timers
and improve PM. But I consider this to be the second level of
optimization. The first level is to get rid of unneeded events
completely.

-- 
Best Regards,
Artem Bityutskiy (Артём Битюцкий)