Messages in this thread | | | Subject | Re: [patch 3/6] sched, nohz: sched group, domain aware nohz idle load balancing | From | Suresh Siddha <> | Date | Wed, 30 Nov 2011 17:03:34 -0800 |
| |
On Tue, 2011-11-29 at 01:44 -0800, Peter Zijlstra wrote: > On Mon, 2011-11-28 at 15:51 -0800, Suresh Siddha wrote: > > On Thu, 2011-11-24 at 03:47 -0800, Peter Zijlstra wrote: > > > On Fri, 2011-11-18 at 15:03 -0800, Suresh Siddha wrote: > > > > + for_each_domain(cpu, sd) { > > > > + struct sched_group *sg = sd->groups; > > > > + struct sched_group_power *sgp = sg->sgp; > > > > + int nr_busy = atomic_read(&sgp->nr_busy_cpus); > > > > + > > > > + if (nr_busy > 1 && (nr_busy * SCHED_LOAD_SCALE > sgp->power)) > > > > + goto need_kick; > > > > > > This looks wrong, its basically always true for a box with HT. > > > > In the presence of two busy HT siblings, we need to do the idle load > > balance to figure out if the load from the busy core can be migrated to > > any other idle core/sibling in the platform. And at this point, we > > already know there are idle cpu's in the platform. > > might have to, this nr_busy doesn't mean its actually busy, just that > its not nohz, it might very well be idle.
correct. But we can change that.
We can track nr_busy_cpus separately and can be updated when ever the rq goes into idle and during the first busy tick after idle. Whereas the nohz.idle_cpus_mask can be updated only during tickless entry.
> > I will modify the above check to: > > > > if (sd->flags & SD_SHARE_PKG_RESOURCES && nr_busy > 1) > > goto need_kick; > > > > This way, if there is a SMT/MC domain with more than one busy cpu in the > > group, then we will request for the idle load balancing. > > Potentially 1 more than 1 busy, right? And we do the balancing just in > case there are indeed busy cpus. > > I think its useful to mention that somewhere near, that this nr_busy > measure we use is an upper bound on actual busy.
The above should cover this.
I will send the updated version shortly.
thanks, suresh
| |