Messages in this thread | | | From | Vladimir Davydov <> | Date | Mon, 4 Jun 2012 17:19:08 +0400 | Subject | Re: [PATCH] cpuidle: menu: use nr_running instead of cpuload for calculating perf mult |
| |
On Jun 4, 2012, at 5:15 PM, "Arjan van de Ven" <arjan@linux.intel.com> wrote:
> On 6/4/2012 3:24 AM, Vladimir Davydov wrote: >> rq->cpuload strongly depends on cgroup hierarchy. For example, if hundreds of >> tasks are running inside cpu:/test cgroup, the sum of cpuload over all cpus >> won't exceed 1024 (by default). That makes the cpuidle menu governor take wrong >> decisions, which can negatively affect overall performance. > > nr_running is the wrong answer... it is instantaneous, not longer term. > cpuidle wants a longer term, per cpu, notion of "busy", to use as proxy > for performance sensitivity. > >
I use not the instantaneous nr_running, but nr_running last seen in update_cpu_load() where cpuload[0], which is currently used by cpuidle, is updated.
| |