Re: busted CFS group load balancer?

[Ken Chen <kenchen@google.com>]

On Tue, Nov 18, 2008 at 4:30 AM, Peter Zijlstra <a.p.zijlstra@chello.nl> wrote:
>> I'm also very troubled with this calculation in __update_group_shares_cpu():
>>
>>         shares = (sd_shares * rq_weight) / (sd_rq_weight + 1);
>>
>> Won't you have rounding problem here?  value 'shares' will gradually
>> decrease for each iteration of __update_group_shares_cpu()?
>
> Yes it will, however at the top of the sched-domain tree its reset.
>
>        if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE))
>                shares = tg->shares;

Hmm?  it is only true mostly on flat smp machine, or on numa system
where task is woken up across a sched-domain.

There is this code in try_to_wake_up():

                for_each_domain(this_cpu, sd) {
                        if (cpu_isset(cpu, sd->span)) {
                                update_shares(sd);
                                break;
                        }
                }

It doesn't iterate to the top half of the time on numa machine.

- Ken