Minutes of Power in 9.1.0 meeting

[Richard A. Smith]

Greg Smith wrote:

>  > * Richard to determine how to address the no regressions requirement and
>  > how to measure the success of the feature in terms of Amps used.

I've been working on such tests off and on since October when the report 
of 8.2 regressions first popped up.

And while I've learned a lot since then I still don't have a test thats 
feasible for short term testing on a generic grab-off-the-shelf XO.  I 
have an idea for a 1 hour test that I think might remove enough 
variability to be useful but it will need some run hour testing to see.

I think it makes more sense right now to invest the time in coding up 
tests that run in lab reading the instrumented XO.  That is after all 
one of the reasons we got that equipment.  Repeated runs of the same 
test for variability still need to be performed.

Other issues are that some tests will need to be long runtime tests and 
unless we wire up a new system for the tinderbox to use It will block 
the tinderbox.  It won't be hard to wire up a new unit for tinderbox as 
I think the only requirement is 1 IO hookup to strobe the power button.

In whatever case, WLAN has to be disabled.  The WLAN power draw varies 
enough in that unless you are in very quiet RF environments its very 
difficult to make a comparison between any given 2 runs as more or less 
power.

Using the instrumented XO the wlan power reading could be subtracted out 
of the total, but repeatability tests need to be performed to see if 
that removes enough variability to make judgments on regressions.  If so 
then it _might_ be possible to add some workload tests involving 
browsing or sharing into the mix and get meaningful results.

On a different note, one test we might think about running is the 
closest thing the industry has to a "standard" battery life test.  It's 
specified on a lot of the netbook specs.

It's defined here: http://it.jeita.or.jp/mobile/e/index.html

However, I'm also seeing that a lot of vendors are choosing not to use 
this test because it generally results in a number higher than what the 
typical user will actually get.

-- 
Richard Smith  <richard at laptop.org>
One Laptop Per Child