latency-policy call for review

[Richard Hughes]

On Wed, 2008-07-16 at 14:44 -0400, Dave Jones wrote:
> On Wed, Jul 16, 2008 at 12:51:36PM -0400, Bill Nottingham wrote:
>  > Richard Hughes (hughsient at gmail.com) said: 
>  > > Anyone want to review this package? It's designed to make it easy to set
>  > > system latency power tunables.
>  > 
>  > How does it coexist with, or obsolete, a separate cpuspeed service?
> 
> last one to run stomps over the other.

Exactly, it's not actually a great friend at the moment.

> this one also seems to lack a lot of the logic in cpuspeeds initscript
> to determine if we're actually capable of running the ondemand governor.
> 
> If the CPU is capable of running ondemand in a manner that it doesn't
> introduce performance regressions, we choose it already in cpuspeed.
> latency policy seems to be trying to second guess all of that.

Well, it's a lot more blunt. In the long term I would like to get the
latency information from the kernel, so we can make some sort of sane
estimation of the latencies involved.

> My vote would be to rip out all of this from latencypolicy, and have lp
> set the sampling rate sysfs knobs.

I guess you meant "have cpuspeed set the knobs" in which case that might
be sane. I would really like the latency information from the kernel so
we can still load ondemand even if the latency is horrible when we
configure for maximum latency for maximum powersaving.

Richard.