Re: [PATCH] sched, fair: Allow a small load imbalance between low utilisation SD_NUMA domains v3
From: Mel Gorman
Date: Tue Jan 07 2020 - 04:13:04 EST
On Tue, Jan 07, 2020 at 09:51:11AM +0800, Hillf Danton wrote:
> Hi Folks
> On Mon, 06 Jan 2020 11:44:57 -0500 Rik van Riel wrote:
> > On Mon, 2020-01-06 at 16:33 +0000, Mel Gorman wrote:
> > > On Mon, Jan 06, 2020 at 10:47:18AM -0500, Rik van Riel wrote:
> > > > > + imbalance_adj = (100 / (env->sd->imbalance_pct - 100)) - 1;
> > > > > +
> > > > > + /*
> > > > > + * Allow small imbalances when the busiest group has
> > > > > + * low utilisation.
> > > > > + */
> > > > > + imbalance_max = imbalance_adj << 1;
> > > > > + if (busiest->sum_nr_running < imbalance_max)
> > > > > + env->imbalance -= min(env->imbalance, imbalance_adj);
> > > > > + }
> > > > > +
> > > >
> > > > Wait, so imbalance_max is a function only of
> > > > env->sd->imbalance_pct, and it gets compared
> > > > against busiest->sum_nr_running, which is related
> > > > to the number of CPUs in the node?
> > > >
> > >
> > > It's not directly related to the number of CPUs in the node. Are you
> > > thinking of busiest->group_weight?
> > I am, because as it is right now that if condition
> > looks like it might never be true for imbalance_pct 115.
> > Presumably you put that check there for a reason, and
> > would like it to trigger when the amount by which a node
> > is busy is less than 2 * (imbalance_pct - 100).
> If three per cent can make any sense in helping determine utilisation
> low then the busy load has to meet
> busiest->sum_nr_running < max(3, cpus in the node / 32);
Why 3% and why would the low utilisation cut-off depend on the number of
CPUs in the node? That simply means that the cut-off scales to machine
size and does not take into account any consideration between local memory
latency and memory bandwidth.
> And we can't skip pulling tasks from a numa node without comparing it
> to the local load
> local->sum_nr_running * env->sd->imbalance_pct <
> busiest->sum_nr_running * 100;
> with imbalance_pct taken into account.
Again, why? In this context, an imbalance has already been calculated
and whether based on running tasks or idle CPUs, it's not a negative
number. The imbalance_adj used as already accounted for imbalance_pct
albeit not as a ratio as it's normally used.