The major Linux power regression situation seems to only be getting worse at this point. Following the Mobile Users Beware: Linux Has Major Power Regression
and The Tests Showing Ubuntu 11.04 On A Power Consumption Binge
, a variety of feedback has come in. There's the usual FUD that it's "Moronix" benchmarks and the like, but the fact of the matter is it's a very real problem and it's about to bite Ubuntu 11.04 and other Linux distributions planning to ship with 2.6.38 kernel or later.
There's now a stream of independent users reporting problems now that they know to look for power consumption issues with Ubuntu 11.04 and the Linux 2.6.38 kernel. To this Launchpad bug report
, which is about increased power usage in Natty (11.04), there's now over a dozen people reporting to be affected by this bug. The bug report itself was started by a Canonical software engineer.
Within the Phoronix Forums and elsewhere are also other reports of the Linux 2.6.38 kernel consuming excessive power compared to the previous releases. This is also not an Ubuntu-specific problem as I laid out with it being in the upstream Linux kernel. As such, to no surprise there's also been Arch Linux reports of the problem too.
This problem isn't limited to the mobile spectrum of hardware, but desktop users have begun reporting increased power usage according to their AC power meters. Increased power consumption on servers and workstations is also to be expected. There's also been desktop Intel Sandy Bridge reports of increased power usage too, so even the newest hardware is impacted. As my earlier tests have shown, even older Pentium M and Atom hardware are affected.
Reports have also come in that due to this regression there's been increased operating temperatures, and as part of this regression, significantly higher wake-ups per second when using a new kernel.
After using the Phoronix Test Suite and its PTS Commercial components to bisect the kernel (nearly complete), it's to no surprise that it's affecting so many hardware components. This is a regression in the core Linux kernel code -- not some hardware driver. As soon as the Phoronix Test Suite software is fully finished I'll report on the specific cause either tomorrow or Monday, but the results may surprise many users and it's not due to some trivial change or code that's resulting in better performance or anything along those lines. In fact, the fix may be far from trivial due to these invasive changes.
If there's any ISVs, IHVs, or other distribution stakeholders that happen to be working over Easter weekend and are able to reproduce this power issue, please contact me and I'll gladly provide the rest of the details that have yet to be released (MichaelLarabel.com
, michaellarabel on FreeNode, etc). I'm also updating my Twitter feed
with bits as discoveries are made.
I'm also looking for any Ubuntu users experiencing this issue and willing to test some Debian kernel packages as soon as I am certain of the individual commit causing the regression (right now it's narrowed down to sub-system / the merged branch during the 2.6.38 merge window).