System tick rate

System tick rate

[Jan Engelhardt]

Hi,


is there an "official" way of obtaining the current tick rate when 
CONFIG_NO_HZ? This is for a simple statistics display tool. Counting the 
interrupts from /proc/interrupts (calculating the difference between two 
timepoints) seems to work, though I have seen at least one x86 machine 
where both IRQ0 and the LOC counters increase, making it not entirely 
obvious which number to take as a base. Suggestions?

Re: System tick rate

[Arjan van de Ven]

On Thu, 12 Mar 2009 14:38:19 +0100 (CET)
Jan Engelhardt <jengelh@medozas.de> wrote:

> Hi,
> 
> 
> is there an "official" way of obtaining the current tick rate when 
> CONFIG_NO_HZ? 

what does "current tick rate" mean for you in a no-hz situation ?
is it the HZ value (which is supposed to be invisible from userspace,
so good luck) or the wakeup count ?



-- 
Arjan van de Ven 	Intel Open Source Technology Centre
For development, discussion and tips for power savings, 
visit http://www.lesswatts.org

Re: System tick rate

[Jan Engelhardt]

On Thursday 2009-03-12 15:21, Arjan van de Ven wrote:
>On Thu, 12 Mar 2009 14:38:19 +0100 (CET) Jan Engelhardt wrote:
>> 
>> is there an "official" way of obtaining the current tick rate when 
>> CONFIG_NO_HZ? 
>
>what does "current tick rate" mean for you in a no-hz situation ?
>is it the HZ value (which is supposed to be invisible from userspace,
>so good luck) or the wakeup count ?

Hm, I guess what I want is the number of timer interrupts which
occurred within the observation period that called the scheduler
code. (So that excludes RTC on usual x86en.)

Re: System tick rate

[Peter Zijlstra]

On Thu, 2009-03-12 at 15:45 +0100, Jan Engelhardt wrote:
> On Thursday 2009-03-12 15:21, Arjan van de Ven wrote:
> >On Thu, 12 Mar 2009 14:38:19 +0100 (CET) Jan Engelhardt wrote:
> >> 
> >> is there an "official" way of obtaining the current tick rate when 
> >> CONFIG_NO_HZ? 
> >
> >what does "current tick rate" mean for you in a no-hz situation ?
> >is it the HZ value (which is supposed to be invisible from userspace,
> >so good luck) or the wakeup count ?
> 
> Hm, I guess what I want is the number of timer interrupts which
> occurred within the observation period that called the scheduler
> code. (So that excludes RTC on usual x86en.)

There is no tick rate with NO_HZ mode, only a max tick rate, but as
Arjan said, not even that is exposed to user-space.

I think Dave's Niagra has the crown here, because he ran into some bug
in the NO_HZ code some while back because he idled longer than we ever
seen before. I think it was in the order of 30 minutes or something
without a single wakeup of the CPU.

Re: System tick rate

[Jan Engelhardt]

On Friday 2009-03-13 09:49, Peter Zijlstra wrote:
>On Thu, 2009-03-12 at 15:45 +0100, Jan Engelhardt wrote:
>> On Thursday 2009-03-12 15:21, Arjan van de Ven wrote:
>> >On Thu, 12 Mar 2009 14:38:19 +0100 (CET) Jan Engelhardt wrote:
>> >> 
>> >> is there an "official" way of obtaining the current tick rate when 
>> >> CONFIG_NO_HZ? 
>> >
>> >what does "current tick rate" mean for you in a no-hz situation ?
>> >is it the HZ value (which is supposed to be invisible from userspace,
>> >so good luck) or the wakeup count ?
>> 
>> Hm, I guess what I want is the number of timer interrupts which
>> occurred within the observation period that called the scheduler
>> code. (So that excludes RTC on usual x86en.)
>
>There is no tick rate with NO_HZ mode, only a max tick rate, but as
>Arjan said, not even that is exposed to user-space.
>
>I think Dave's Niagra has the crown here, because he ran into some bug
>in the NO_HZ code some while back because he idled longer than we ever
>seen before. I think it was in the order of 30 minutes or something
>without a single wakeup of the CPU.
>
Well if it idled that long, it surely did not run any processes, did it?