Burning Through Power: Linux Regressions Found
Phoronix: Burning Through Power: Linux Regressions Found
For the multiple Linux kernel power regressions that I've talked about on Phoronix now for a number of weeks and have been affecting mobile Linux users en mass, I said I was looking for a better power measuring approach by using an AC power meter / UPS rather than a notebook battery to use in nailing these regressions. Using such a power meter would lead to a fully-automated process by the Phoronix Test Suite as no longer would I need to keep pulling the power plug from a laptop, could use much faster hardware, and allow for some other interesting possibilities. Well, last week I bought a power meter that plays with Linux. So now there's some news to share...
Michael is a living legend.
While I would say you've done good work (you have, if you've done what you've said), you've been saying you've found some of the power-leaks for weeks now, so saying it won't make it to users until 3.1 is a little disheartening.
If you disclosed these bugs individually as you found them rather than teasing and bundling them all up for the greatest publishing value, real users could be saving a lot of power right now. Chances are people might have had time to improve on some of your patches and get them into 3.0. Instead it's going to be another three+ month wait for the next stable kernel (assuming they're accepted in time).
You're doing the right thing but you're doing it the wrong way.
I think this is quite self-explanatory:
Originally Posted by oliw
Edit: Unless you care to sponsor Michaels work?
Originally Posted by Michael Larabel
impressive work... hell i'm a desktop user and i don't care about power consuming but this way to find and fix bugs are the gold and best way!
i'm sure Microsoft lose the battle only because of this kind of regression testing.
CyanogenMod (the popular Android ROM) has been using 2.6.37 again for weeks now since you disovered the power regression. They will be happy too :-)
This is great news. Thank you, Michael.
I admire your commitment to testing the Linux kernel. Some people say, your approaches were unscientific and testing was done too vague. However, I think you are doing it exactly the right way: continuously monitor performance (and if things are working properly) on a system with a common configuration, find a bug or a regression, reproduce it and hunt it down. Of course, there is still work to be done by the respective subsystem maintainer to look at all the gory details and to find a good solution. I wish, there were more guys like you doing such testing beyond "it's working, I do not need more".
btw: I would like to click on your ads, but I do not see any. Maybe the scripts do not work in Konqueror?
Tags for this Thread