Hi All,

I got inputs from couple of folks for this testing. Based on this, I have the opinion that the issue exists even with x86 platforms as well but with lesser severity (worstcase 89%), most probably due to optimized governor parameters (esp cpufreq_transition_latency). Also I figured out that there is an interesting patch from David C Niemi (http://kerneltrap.org/mailarchive/linux-kernel/2010/10/6/4628889/thread) which helps in reducing frequent OPP changes when there is high load in the system which helped to improve ondemand performance.

After optimizing cpufreq_transition_latency for omap along with applying above patch, cpufreq-bench results on OMAP are much better. Attached files have the results. cpufreq-bench results without optimization on OMAP. Round 1 - 41.11% Round 2 - 41.61% Round 3 - 40.79% Round 4 - 41.17% Round 5 - 52.58%

Time spent in different P states: 300M - 12.26% 600M - .28% 800M - 0% 1000M - 87.33%

cpufreq-bench results with optimization on OMAP. Round 1 - 90.24% Round 2 - 94.48% Round 3 - 96.06% Round 4 - 96.6% Round 5 - 86.89%

Time spent in different P states: 300M - 3.26% 600M - 0.4% 800M - 0% 1000M - 96.33%

Special thanks to Mark and Vincent for their testing and feedbacks. Attached docs have detailed report log.

Regards Vishwa

On Tue, Nov 16, 2010 at 11:27 PM, Mark Wilcox m_p_wilcox@yahoo.co.uk wrote:

Hi Vishwanath,

...

I am trying to investigate ondemand governor’s limitation in Linux kernel. As part of that, I have found a tool called cpufreq-bench which can be used to determine the performance degradation due to ondemand governor compared to performance governor.

I would need some of your help to run this test bench on some of recent X86 platforms that have support for many P states. (My PC is little old which has only 2 p-states (2.8GHz & 3.4 GHz), so performance degradation is not much visible). tarball with cpufreqbench and readme is available at http://people.linaro.org/~amitk/cpufreq.tgz

I'm just a lurker on the list at the moment but thought I could easily do this to help out.  Log attached.

A little feedback on the instructions - cpufreq-bench's Makefile doesn't have an install target although the instructions say to do make; make install, I copied the binary manually.

Copying these results to the list as requested but is that really necessary for everyone?  Could you summarise the results for us all when you've got enough please?

Mark

---

From: Vishwanath Sripathy vishwanath.sripathy@linaro.org To: "linaro-dev@lists.linaro.org" linaro-dev@lists.linaro.org Sent: Tue, 16 November, 2010 13:07:46 Subject: Re: cpufreq-bench test on X86 platforms

Hi All,

tarball with cpufreqbench and readme is available at http://people.linaro.org/~amitk/cpufreq.tgz

Vishwanath On Tue, Nov 16, 2010 at 3:37 PM, Sripathy, Vishwanath vishwanath.bs@ti.com wrote:

...

All,

I am trying to investigate ondemand governor’s limitation in Linux kernel. As part of that, I have found a tool called cpufreq-bench which can be used to determine the performance degradation due to ondemand governor compared to performance governor.

I have used this tool on OMAP platforms and able to see the issue. Now I would like to demonstrate this issue on x86 platform to conclude that it’s a generic governor problem and nothing specific to ARM based SOCs.

I would need some of your help to run this test bench on some of recent X86 platforms that have support for many P states. (My PC is little old which has only 2 p-states (2.8GHz & 3.4 GHz), so performance degradation is not much visible).

More details on the tool and test procedure can be found at

https://blueprints.edge.launchpad.net/linaro-pm-wg/+spec/cpufreq-ondemand-go...

Pls ping me if you need any assistance in testing.

Looking forward for your test results.

Regards

Vishwanath

---

linaro-dev mailing list linaro-dev@lists.linaro.org http://lists.linaro.org/mailman/listinfo/linaro-dev

---

linaro-dev mailing list linaro-dev@lists.linaro.org http://lists.linaro.org/mailman/listinfo/linaro-dev