Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1312168 > unrolled thread

Re: [PATCH 3/4] sched: introduce synchronized idle injection

Back to article view | Back to linux.kernel

This discussion starts older than the indexed window; earlier articles aren't shown. The article labeled Started by below is the oldest one visible, not the original post.

Contents

  Re: [PATCH 3/4] sched: introduce synchronized idle injection Ingo Molnar <mingo@kernel.org> - 2016-01-19 13:10 +0100
    Re: [PATCH 3/4] sched: introduce synchronized idle injection Jacob Pan <jacob.jun.pan@linux.intel.com> - 2016-01-19 19:10 +0100

#1312168 — Re: [PATCH 3/4] sched: introduce synchronized idle injection

* Jacob Pan <jacob.jun.pan@linux.intel.com> wrote:

> Note: PPW_RAPL= performance per watt measured by RAPL package power
> meter.
> 
> If we just look at 20% forced idle, I can see significantly more deep
> package C-states(PC7) residency.
> 		Pkg%pc2 	Pkg%pc6 	Pkg%pc7 
> w/o patch	3.17		4.39   		13.21 
> w/ patch	1.15   		1.43            17.24 

Can this be translated to: "this patch improved power efficiency (watts per 
second) by XYZ percent?", with XYZ filled in accordingly?

Thanks,

	Ingo

[toc] | [next] | [standalone]

#1312418

On Tue, 19 Jan 2016 13:06:05 +0100
Ingo Molnar <mingo@kernel.org> wrote:

> > If we just look at 20% forced idle, I can see significantly more
> > deep package C-states(PC7) residency.
> > 		Pkg%pc2 	Pkg%pc6 	Pkg%pc7 
> > w/o patch	3.17		4.39   		13.21 
> > w/ patch	1.15   		1.43            17.24   
> 
> Can this be translated to: "this patch improved power efficiency
> (watts per second) by XYZ percent?", with XYZ filled in accordingly?

With CPU bound workload and 20% idle injection ratio, this patch
improves performance per watt by 3.2% on Ivy Bridge system with
Package C7 as the deepest idle.

The above packages C-state residency explains why, i.e. more deeper PC7
when tick is stopped.

With newer CPUs, this improvements should be more visible in that
low power package C-states are getting deeper/less power. I will test
on Skylake client as soon as I have time.

Thanks,

Jacob

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web