FS#47086 - [linux] consider increasing the tick rate from 300 to 1000 in the kernel config
Attached to Project:
Arch Linux
Opened by John (graysky) - Monday, 16 November 2015, 20:23 GMT
Last edited by Doug Newgard (Scimmia) - Thursday, 01 March 2018, 14:41 GMT
Opened by John (graysky) - Monday, 16 November 2015, 20:23 GMT
Last edited by Doug Newgard (Scimmia) - Thursday, 01 March 2018, 14:41 GMT
|
Details
Increase the timer frequency from 300 Hz to 1000 Hz in the
kernel package. The help in the kernel docs in the nconfig
suggest that the default should be 1000 per my read:
"It is customary to have the timer interrupt run at 1000 Hz but 100 Hz may be more beneficial for servers and NUMA systems that do not need to have a fast response for user interaction and that may experience bus contention and cacheline bounces as a result of timer interrupts." It also seems that the 300 Hz tick rate is to blame for problems associated with higher power consumption on Sandy and Haswell chips whereas a setting of 1000 Hz fixes these problems. See references below. Thanks. 1. https://bbs.archlinux.org/viewtopic.php?id=193983 2. http://ubuntuforums.org/showthread.php?t=2266297&page=3 |
This task depends upon
Hibernation level is the same (C7)
But this is clearly not the same frequencies.
I think it is difficult to see the differences in consumption just when watching temperature in idle. Maybe the motherboard fan speeds to compensate or the temperature difference is too small to be seen.
I think the difference in power usage is ~10W.
"My recollection (suspect) as to how all this played out was that the 300 Kernel tick rate used by Arch linux as a culprit was a red herring.
The root cause issues were also present in the 250 Hz and 1000 Hz kernels. However, the 300 Hz kernel with a GUI desktop sometimes was rather optimal for manifesting the issue or issues."
http://ubuntuforums.org/showthread.php?t=2266297&page=3&p=13392458#post13392458
"John created a kernel using an Ubuntu config file. And supplied the perf record data. The results were the same. Therefore the conclusion is that this is a not a kernel configuration issue."
https://bugzilla.kernel.org/show_bug.cgi?id=93521#c53
Way back when intel_pstate was set as default my CPU(i7-2600k) started overheating and I replaced the CPU cooler, because I thought the heatpipe was damaged or something like that.
I only figured it out afterwards...
I tried again yesterday with intel_pstate and the problem still exists. CPU heats to ~60°C on an idle desktop and power draw is almost the same as under full load.
I guess I'll stay with acpi-cpufreq on that CPU.
Changing the interrupt frequency it's not a true fix for the higher power consumption but if it prevents the problem for us with haswell cpus with no side effects for the others please consider changing it to 250.