On 27 May 2014 19:32, Peter Zijlstra peterz@infradead.org wrote:
On Fri, May 23, 2014 at 05:53:02PM +0200, Vincent Guittot wrote:
Monitor the activity level of each group of each sched_domain level. The activity is the amount of cpu_power that is currently used on a CPU or group of CPUs. We use the runnable_avg_sum and _period to evaluate this activity level. In the special use case where the CPU is fully loaded by more than 1 task, the activity level is set above the cpu_power in order to reflect the overload of the CPU
+static int get_cpu_activity(int cpu) +{
struct rq *rq = cpu_rq(cpu);u32 sum = rq->avg.runnable_avg_sum;u32 period = rq->avg.runnable_avg_period;if (sum >= period)return power_orig_of(cpu) + rq->nr_running - 1;return (sum * power_orig_of(cpu)) / period;+}
While I appreciate the need to signify the overload situation, I don't think adding nr_running makes sense. The amount of tasks has no bearing on the amount of overload.
I agree that it's not the best way to evaluate overload but it's the only simple one that is available without additional computation. I'm going to try to find a better metric or come back the solution which only adds +1 and compute the overload somewhere else
Also, and note I've not yet seen the use, it strikes me as odd to use the orig power here. I would've thought the current capacity (not the max capacity) is relevant to balancing.
activity also measures the non cfs tasks activity whereas current capacity removes the capacity used by non cfs tasks. So as long as arch_scale_cpu_freq is not used (which is the case for now), original capacity is ok but we might need a 3rd metric so we would have: original capacity (max capacity that can provide a CPU) usable capacity (new value that reflects current capacity available for any kind of processing rt tasks, irq, cfs tasks) current capacity (capacity available for cfs tasks)
Vincent