The Linux Kernel Is Still On A Power Binge

The power regressions just haven't been mentioned recently since I've been out of the office since late last month due to UDS Budapest, LinuxTag 2011, and beer drinking with Phoronix readers in Bavaria and around Germany. Now that I'm back to the usual workload, I've run some more kernel tests to verify the increased power consumption is still there with the latest upstream kernel. Sadly, the issue is still there.
There's many more realizing as well that they are affected by this bug based upon the Ubuntu bug report and user experiences within the Phoronix Forums, Ubuntu Forums, etc.
With the Linux 2.6.39 kernel using the Linus Torvalds tree as of last night (17 May), the issue is still very much present when tested on one of the Lenovo ThinkPad notebooks with an Intel Core 2 Duo CPU.
Earlier this month while I was away there was some speculation that the cause of at least one of these power regressions was due to some SMP / tick-less changes in the recent kernels. In particular, this LKML thread from last September was referenced by a user commenting on the Ubuntu bug report. The thought was for SMP (multi-core) systems the load balancing tick was keeping the system up excessively.
Beyond testing the very latest kernel Git to verify its existence, I also ran a test this morning of the kernel in its default SMP kernel configuration and then again when it was booted with the nosmp flag at boot-time (and separately, with maxcpus=1 as well just to ensure no differences) so that the dual-core Intel CPU was only being treated as single-threaded.
This didn't help things and the kernel is still on a power binge. Testing continues.
Additionally, if anyone has any recommendations about UPS power systems that have a USB interface and support monitoring the power consumption under Linux via a sysfs node or similar, knowing that would be much appreciated. (Or any UPS vendors willing to send one out to Phoronix would be really great!) This would greatly expedite the testing process due to much faster desktop/server/workstation hardware around here and would allow to see the full scope of hardware affected by this major Linux power problem rather than relying upon notebook battery consumption monitoring.
14 Comments