Re: [patch 0/7] sched: change nohz idle load balancing logic to push model

[Dominik Brodowski <linux@dominikbrodowski.net>]

Hey

On Thu, May 20, 2010 at 09:35:47PM +1000, Nigel Cunningham wrote:
> On 20/05/10 21:17, Dominik Brodowski wrote:
>> Hey,
>>
>> On Thu, May 20, 2010 at 09:07:42PM +1000, Nigel Cunningham wrote:
>>> "Load balancing tick" is still number one in my powertop list of top
>>> causes of wakeups (sitting at ~60 to 80 per second as I type this, with
>>> ~170 wakeups per second total). Comparing this to the numbers I posted
>>> earlier, we seem to have a win.
>>>
>>> I do wonder, though, whether further work could still be done. If I take
>>> one core offline, for example, I'm still getting load balancing ticks.
>>> Intuitively, I'd expect there to be no need for them with only one core
>>> available. But maybe I'm just ignorant of what's going on.
>>
>> Are you using HZ=1000, and is the CPU active ~ 6-8 % ? If so, is it just the
>> regular timer tick while the CPU is active, and so not a real "wakeup"? (Or
>> possibly double the number if both CPUs are active)
>
> HZ is 1000, and the CPU running percentage in Powertop is low (2% when  
> I'm not typing).

Still, that'd count for up to 4 % of the reported ticks. On my system, this
phenomenon accounts pretty much for the events powertop calls "load balancin
ticks". I wonder whether user- or kernelspace should be more smart to not
consider something a wakeup event if it wasn't a wakeup, i.e. to ignore
timer events from timer_stat / powertop if the CPU wasn't actually idle...

Best,
	Dominik