Re: [PATCH v1 5/6] mm: make vmstat_update periodic run conditional

[Gilad Ben-Yossef]

On Mon, May 7, 2012 at 10:40 PM, Christoph Lameter <cl@xxxxxxxxx> wrote:
> On Mon, 7 May 2012, KOSAKI Motohiro wrote:
>
>> > > @@ -1204,8 +1265,14 @@ static int __init setup_vmstat(void)
>> > >
>> > >           register_cpu_notifier(&vmstat_notifier);
>> > >
>> > > + INIT_DELAYED_WORK_DEFERRABLE(&vmstat_monitor_work,
>> > > +                         vmstat_update_monitor);
>> > > + queue_delayed_work(system_unbound_wq,
>> > > +                         &vmstat_monitor_work,
>> > > +                         round_jiffies_relative(HZ));
>> > > +
>> > >           for_each_online_cpu(cpu)
>> > > -         start_cpu_timer(cpu);
>> > > +         setup_cpu_timer(cpu);
>> > >   #endif
>> > >   #ifdef CONFIG_PROC_FS
>> > >           proc_create("buddyinfo", S_IRUGO,
>> > > NULL,&fragmentation_file_operations);
>> >
>> > So the monitoring thread just bounces around the system? Hope that the
>> > scheduler does the right thing to keep it on processors that do some other
>> > work.
>>
>> Good point. Usually, all cpus have update items and monitor worker only makes
>> new noise. I think this feature is only useful some hpc case.  So I wonder if
>> this vmstat improvemnt can integrate Frederic's Nohz cpusets activity. I.e.
>> vmstat-update integrate timer house keeping and automatically stop when
>> stopping
>> hz house keeping.
>
> Right. We could do the same processing in vmstat update and the
> thread could check if it is the last vmstat update thread. If so simply
> continue and do not terminate.
>
> But this would still mean that the vmstat update thread would run on an
> arbitrary cpu. If I have a sacrificial lamb processor for OS processing
> then I would expect the vmstat update thread to stick to that processor
> and avoid to run on the other processor that I would like to be as free
> from OS noise as possible.
>

OK, what about -

- We pick a scapegoat cpu (the first to come up gets the job).
- We add a knob to let user designate another cpu for the job.
- If scapegoat cpus goes offline, the cpu processing the off lining is
the new scapegoat.

Does this makes better sense?

Gilad


-- 
Gilad Ben-Yossef
Chief Coffee Drinker
gilad@xxxxxxxxxxxxx
Israel Cell: +972-52-8260388
US Cell: +1-973-8260388
http://benyossef.com

"If you take a class in large-scale robotics, can you end up in a
situation where the homework eats your dog?"
 -- Jean-Baptiste Queru
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/