On Thu, Mar 19, 2015 at 09:33:00PM +0000, Jorge Ramirez-Ortiz wrote:
On 03/19/2015 02:44 PM, Alan Bennett wrote:
On a similar note, I was going to run some of the LKP tests for a baseline last week. Of course the first one I chose, pigz[1], is a pretty good all-cores stress test and during the test run from I observed the thermal warnings and forced shutdowns after about 1-2 minutes. I was going to hit it with some freeze spray, but thought I had better not if I wanted to make sure the system was in good shape for demos.
https://git.kernel.org/cgit/linux/kernel/git/wfg/lkp-tests.git/tree/tests/pi... ..parallel implementation of gzip, is a fully functional replacement for gzip that exploits multiple processors and multiple cores to the hilt when compressing data...
Just received this from the Energy Aware Scheduler mailing list. I thought it might be of interest.
FYI, this week ARM opensourced the Workload Automation suite that is useful for EAS evaluation License is Apache ver 2.0
Good inputs for three excellent cases for cpu workload.
Now kernel has not enabled cpuidle, so that this will introduce static power leakage; so even the cpu is in idle state, it still will consume power. If enable the cpuidle, we can get power benefit from cpu idle state.
In the cases which cpus run with 100% busy state, if the thermal is high enough, there mainly have two ways to cool down cpu: - Limit the cpu frequency, so that decrease the dynamic power leakage; Thermal framework has provided cpu cooling device driver for that; - Limit the cpu numbers (or using hotplug governor), the specially case is to hotplug out all cpus in the second cluster so that can easily decrease the static power leakage; I'm not sure EAS/IPA can support for this or not?
So far we need firstly enable PSCI's hotplug/cpuidle functionality, so that it can support furthmore's power profiling/optimization.
Thanks, Leo Yan