[Xenomai] question: scheduling latency

[Xenomai] question: scheduling latency

[ali hagigat]

What is the meaning of the scheduling latency in Xenomai and Xenomai
tests like the latency test by the latency utility?

Is it the time after finishing an interrupt and executing the first
instruction of a real time task? Or it is the time that each real time
task is executed below that?

Re: [Xenomai] question: scheduling latency

[Gilles Chanteperdrix]

On 06/25/2012 11:17 AM, ali hagigat wrote:
> What is the meaning of the scheduling latency in Xenomai and Xenomai
> tests like the latency test by the latency utility?
> 
> Is it the time after finishing an interrupt and executing the first
> instruction of a real time task? Or it is the time that each real time
> task is executed below that?

Read the source, what the test does it obvious, it is the time from the
timer interrupt, to the time we return to the measurement thread.

-- 
					    Gilles.

[Xenomai] question: scheduling latency

[ali hagigat]

Does the scheduling latency is variable for a Xenomai system and
depends on the amount of load of a system and the number of the system
tasks and Linux applications?

Re: [Xenomai] question: scheduling latency

[Gilles Chanteperdrix]

On 06/30/2012 12:48 PM, ali hagigat wrote:
> Does the scheduling latency is variable for a Xenomai system and
> depends on the amount of load of a system and the number of the system
> tasks and Linux applications?

The scheduling latency is variable, but the worst case scheduling
latency and best case scheduling latencies are not, by definition.

Running the latency test under high load for a long time gives you a
good idea of the worst case and best case scheduling latencies. But you
should be sure to try and cover as much as possible of the whole code of
the linux kernel, which is why we use LTP as part of the load we put on
the system.

-- 
                                                                Gilles.

Re: [Xenomai] question: scheduling latency

[ali hagigat]

Does the worst/best case scheduling latencies depend on the number of
the real time tasks? How about the number of Linux non real-time
tasks?
Does the number of the interrupts effect the latency? How about the
type of the interrupt?

Much appreciate for the reply.

On Sat, Jun 30, 2012 at 3:24 PM, Gilles Chanteperdrix
<gilles.chanteperdrix@xenomai.org> wrote:
> On 06/30/2012 12:48 PM, ali hagigat wrote:
>> Does the scheduling latency is variable for a Xenomai system and
>> depends on the amount of load of a system and the number of the system
>> tasks and Linux applications?
>
> The scheduling latency is variable, but the worst case scheduling
> latency and best case scheduling latencies are not, by definition.

Re: [Xenomai] question: scheduling latency

[Gilles Chanteperdrix]

On 06/30/2012 01:12 PM, ali hagigat wrote:
> Does the worst/best case scheduling latencies depend on the number of
> the real time tasks? How about the number of Linux non real-time
> tasks?

Again: the worst case/best case scheduling latencies do not depend on
anything, by definition.

-- 
                                                                Gilles.

Re: [Xenomai] question: scheduling latency

[ali hagigat]

Thank you so much for the reply. So how the worst/best case latencies
are computed?

Regards

On Sat, Jun 30, 2012 at 3:43 PM, Gilles Chanteperdrix
<gilles.chanteperdrix@xenomai.org> wrote:
> On 06/30/2012 01:12 PM, ali hagigat wrote:
>> Does the worst/best case scheduling latencies depend on the number of
>> the real time tasks? How about the number of Linux non real-time
>> tasks?
>
> Again: the worst case/best case scheduling latencies do not depend on
> anything, by definition.
>
> --
>                                                                Gilles.

Re: [Xenomai] question: scheduling latency

[Gilles Chanteperdrix]

On 06/30/2012 02:01 PM, ali hagigat wrote:
> Thank you so much for the reply. So how the worst/best case latencies
> are computed?

I already answered that, you run the latency test under heavy load, for
a long time, and the measured minimum and maximum latencies should be a
good approximation of the worst and best case latencies.

-- 
                                                                Gilles.