On 9 October 2014 14:16, Peter Zijlstra [off-list ref] wrote:

On Tue, Oct 07, 2014 at 02:13:36PM +0200, Vincent Guittot wrote:

quoted

+static inline bool
+group_has_capacity(struct lb_env *env, struct sg_lb_stats *sgs)
 {
+     if ((sgs->group_capacity * 100) >
+                     (sgs->group_usage * env->sd->imbalance_pct))
+             return true;

Why the imb_pct there? We're looking for 100% utilization, not 130 or
whatnot, right?

Having exactly 100% is quite difficult because of various rounding.
So i have added a margin/threshold to prevent any excessive change of the state.
I have just to use the same margin/threshold than in other place in
load balance.

so the current threshold depends of the sched_level. it's around 14% at MC level

quoted

+     if (sgs->sum_nr_running < sgs->group_weight)
+             return true;

With the code as it stands, this is the cheaper test (no mults) so why
is it second?

quoted

+     return false;
+}

+static inline bool
+group_is_overloaded(struct lb_env *env, struct sg_lb_stats *sgs)
+{
+     if (sgs->sum_nr_running <= sgs->group_weight)
+             return false;
+
+     if ((sgs->group_capacity * 100) <
+                     (sgs->group_usage * env->sd->imbalance_pct))
+             return true;

+     return false;
 }

Same thing here wrt imb_pct