Messages in this thread | | | Date | Tue, 10 Jan 2012 09:08:06 +0100 | Subject | Re: [PATCH] x86,sched: Fix sched_smt_power_savings totally broken | From | Vincent Guittot <> |
| |
On 10 January 2012 02:54, Suresh Siddha <suresh.b.siddha@intel.com> wrote: > On Mon, 2012-01-09 at 15:29 +0100, Vincent Guittot wrote: >> Hi Peter, >> >> I'm also using sched_mc level for doing powersaving load balance on >> ARM platform and we have real benefits. > > hi Vincent, Can you elaborate on your platform topology where you see > the benefits? >
Hi Suresh,
I'm using dual/quad cores ARM platform. When sched_mc is different from 0, the topology is changed from 1 socket/cluster to 2 virtual sockets/clusters with 1 or 2 cores in each cluster. For low cpu load situation, the tasks are gathered in 1 virtual socket and let the cores in the other socket to enter idle state thanks to cpuidle. On ARM platform, cores in a cluster can enter interesting idle state even if the complete cluster can't reach a deep idle state. With dual cores platform, i'm increasing the cpu_power to gather tasks on 1 core when cores are at lowest frequency. In fact i'm detecting a low cpu load (only small tasks are running) and ensure that these small tasks are kept on 1 core.
Regards, Vincent
> on x86, 'sched_mc' was designed for the case of consolidating load into > one socket before using another idle socket. > > Just wondering if this is your use case too or if you are consolidating > at a different topological level. > > thanks, > suresh >
| |