Metrics for comparing power savings of patches...

Metrics for comparing power savings of patches...

[Scott Thompson]

I've seen several patch submissions and comments with varying
claims of 'power savings' from one level to another on both the 
linux-pm and powertop lists.

Standard technique seems to be to plug computer directly into a
wattmeter and then provide some feedback as to the savings comparing
one patchgroup to what is currently available.

Different wattmeters, however, will have different response and 
accuracy depending on load for instant readings; usually the lower-
loads (such as laptops) will be less accurate.

However, I have seen that most wattmeters do a 'better' job at 
measuring total wattage over an interval than the 'instant 
reading'. I advise caution in advertising 'expected savings' for 
instant readings....

Also, if anyone knows of any free benchmarks, best practices, or 
guidelines for trying to do comparative power analysis for patches 
with wattmeters please let me know.

---------------------------------------
Scott Thompson / postfail@hushmail.com
---------------------------------------

Re: Metrics for comparing power savings of patches...

[Pavel Machek]

On Fri 2008-03-07 12:01:05, Scott Thompson wrote:
> I've seen several patch submissions and comments with varying
> claims of 'power savings' from one level to another on both the 
> linux-pm and powertop lists.
> 
> Standard technique seems to be to plug computer directly into a
> wattmeter and then provide some feedback as to the savings comparing
> one patchgroup to what is currently available.
> 
> Different wattmeters, however, will have different response and 
> accuracy depending on load for instant readings; usually the lower-
> loads (such as laptops) will be less accurate.

notebooks usually have internal wattmeters. cat /proc/acpi/batt*/*/*.

> However, I have seen that most wattmeters do a 'better' job at 
> measuring total wattage over an interval than the 'instant 
> reading'. I advise caution in advertising 'expected savings' for 
> instant readings....
> 
> Also, if anyone knows of any free benchmarks, best practices, or 
> guidelines for trying to do comparative power analysis for patches 
> with wattmeters please let me know.

Intel has something on lesswatts.org.
-- 
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html

Re: Metrics for comparing power savings of patches...

[Arjan van de Ven]

Pavel Machek wrote:
>> However, I have seen that most wattmeters do a 'better' job at 
>> measuring total wattage over an interval than the 'instant 
>> reading'. I advise caution in advertising 'expected savings' for 
>> instant readings....
>>
>> Also, if anyone knows of any free benchmarks, best practices, or 
>> guidelines for trying to do comparative power analysis for patches 
>> with wattmeters please let me know.
> 
> Intel has something on lesswatts.org.

the most "accurate" method is "minutes before the battery is empty" as long
as you do measurements shortly after eachother on the same battery and take care
of how full the battery is charged when you start the rundown.

However.. this takes forever to do any kind of experiment.

Re: Metrics for comparing power savings of patches...

[Wes Felter]

Scott Thompson wrote:

> Also, if anyone knows of any free benchmarks, best practices, or 
> guidelines for trying to do comparative power analysis for patches 
> with wattmeters please let me know.

The gold standard appears to be the SPECpower methodology. 
http://www.spec.org/power_ssj2008/docs/SPECpower-Methodology.pdf 
Unfortunately, I've heard their certified power meters are quite expensive.

Wes Felter - wesley@felter.org