Re: [PATCH V2] sched: Improve load balancing in the presence of idle CPUs

[Srikar Dronamraju <srikar@linux.vnet.ibm.com>]

* Jason Low <jason.low2@hp.com> [2015-03-26 22:07:21]:

> On Fri, 2015-03-27 at 10:12 +0800, Wanpeng Li wrote:
> > Hi Preeti,
> > On Thu, Mar 26, 2015 at 06:32:44PM +0530, Preeti U Murthy wrote:
> > >
> > >1. An ILB CPU was chosen from the first numa domain to trigger nohz idle
> > >load balancing [Given the experiment, upto 6 CPUs per core could be
> > >potentially idle in this domain.]
> > >
> > >2. However the ILB CPU would call load_balance() on itself before
> > >initiating nohz idle load balancing.
> > >
> > >3. Given cores are SMT8, the ILB CPU had enough opportunities to pull
> > >tasks from its sibling cores to even out load.
> > >
> > >4. Now that the ILB CPU was no longer idle, it would abort nohz idle
> > >load balancing
> > 
> > I don't see abort nohz idle load balancing when ILB CPU was no longer idle 
> > in nohz_idle_balance(), could you explain more in details?
> 
> Hi Wanpeng,
> 
> In nohz_idle_balance(), there is a check for need_resched() so if the
> cpu has something to run, it should exit nohz_idle_balance(), which may
> cause it to not do the idle balancing on the other CPUs.
> 

Yes, the need_resched() in nohz_idle_balance() would exit the
nohz_idle_balance if it has something to run. However I wonder if we
should move the need_resched check out of the for loop. i.e the
need_resched check should probably be there with the idle check.

With the current code when the ilb cpus are not free: 
- We would be updating the nohz.next_balance even through we havent done
  any load balance.
- We might iterate thro the nohz.idle_cpus_mask()s to find balance_cpus.

-- 
Thanks and Regards
Srikar Dronamraju