Help needed to implement parameter in SCHED_DEADLINE policy

Help needed to implement parameter in SCHED_DEADLINE policy

[Chanchal Paul]

Hi all,

I am trying to implement my custom scheduling policy in rt patched
linux kernel. I am working with version 3.14.25-rt22. I have
successfully installed the kernel after patching with rt patch of same
version.

Now the installed kernel also has SCHED_DEADLINE i.e. the EDF scheduling policy.

The problem in my hand is to introduce a parameter in the
SCHED_DEADLINE policy which will further dynamically change the
priority according to the supplied value while the task is submitted.

I have already simulated and tested my requirement. I am having
problem with understanding the hierarchy of scheduler code.

Thanks in advance.

Re: Help needed to implement parameter in SCHED_DEADLINE policy

[Henrik Austad]

On Fri, Dec 26, 2014 at 09:14:22PM +0530, Chanchal Paul wrote:
> Hi all,

Hi Paul,

> I am trying to implement my custom scheduling policy in rt patched
> linux kernel. I am working with version 3.14.25-rt22. I have
> successfully installed the kernel after patching with rt patch of same
> version.
> 
> Now the installed kernel also has SCHED_DEADLINE i.e. the EDF scheduling policy.
> 
> The problem in my hand is to introduce a parameter in the
> SCHED_DEADLINE policy which will further dynamically change the
> priority according to the supplied value while the task is submitted.

You do know that SCHED_DEADLINE does not use priorities but _deadlines_, 
right?

care to elaborate a bit?

> I have already simulated and tested my requirement. I am having
> problem with understanding the hierarchy of scheduler code.

You mean the order of the scheduling policies? If you look at the 
sched_class in the different classes, you'll find the following hierarchy:

stop_sched_class
dl_sched_class
rt_sched_class
fair_sched_class
idle_sched_class

So, first the scheduler sees if there's any deadline tasks, then -rt and so 
on.

I (we) need a bit more info about what exactly it is you're struggling with

> 
> Thanks in advance
> --
> To unsubscribe from this list: send the line "unsubscribe linux-rt-users" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

-- 
Henrik Austad

Re: Help needed to implement parameter in SCHED_DEADLINE policy

[Chanchal Paul]

Hi,

Let me explain what I am thinking.

I am only concerned about EDF scheduling policy. The scheduler
determines the order in which it should schedule tasks according to
their DEADLINE. I dont want to disturb that order. If a scheduler
picks up a task which is computationally intensive and lets it run
then due to its thermal characteristic it may heat up processor so
much that eventually a hardware DTM or a frequency scaling occurs and
the system slows down as a effect. Where as if the processor would
have waited a little bit (only that much due to which the task would
not miss its deadline) before letting the task run, and eventually
lets the other computationally non intensive tasks run (from the other
scheduling classes) then may be that DTM would have not occurred and
may be the system would not have slowed down.

So I want to solve this problem. or at least see what are the results
after implementation. I can think that the scheduler need more
information about the EDF tasks about their thermal (or computational)
characteristics. Initially I want to supply that information when I
submit a EDF task.

Thank you.

On Sat, Dec 27, 2014 at 3:44 AM, Henrik Austad <henrik@austad.us> wrote:
> On Fri, Dec 26, 2014 at 09:14:22PM +0530, Chanchal Paul wrote:
>> Hi all,
>
> Hi Paul,
>
>> I am trying to implement my custom scheduling policy in rt patched
>> linux kernel. I am working with version 3.14.25-rt22. I have
>> successfully installed the kernel after patching with rt patch of same
>> version.
>>
>> Now the installed kernel also has SCHED_DEADLINE i.e. the EDF scheduling policy.
>>
>> The problem in my hand is to introduce a parameter in the
>> SCHED_DEADLINE policy which will further dynamically change the
>> priority according to the supplied value while the task is submitted.
>
> You do know that SCHED_DEADLINE does not use priorities but _deadlines_,
> right?
>
> care to elaborate a bit?
>
>> I have already simulated and tested my requirement. I am having
>> problem with understanding the hierarchy of scheduler code.
>
> You mean the order of the scheduling policies? If you look at the
> sched_class in the different classes, you'll find the following hierarchy:
>
> stop_sched_class
> dl_sched_class
> rt_sched_class
> fair_sched_class
> idle_sched_class
>
> So, first the scheduler sees if there's any deadline tasks, then -rt and so
> on.
>
> I (we) need a bit more info about what exactly it is you're struggling with
>
>>
>> Thanks in advance
>> --
>> To unsubscribe from this list: send the line "unsubscribe linux-rt-users" in
>> the body of a message to majordomo@vger.kernel.org
>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>
> --
> Henrik Austad

Re: Help needed to implement parameter in SCHED_DEADLINE policy

[Valdis.Kletnieks]

[-- Attachment #1.1: Type: text/plain, Size: 1018 bytes --]

On Sun, 28 Dec 2014 00:09:36 +0530, Chanchal Paul said:
> their DEADLINE. I dont want to disturb that order. If a scheduler
> picks up a task which is computationally intensive and lets it run
> then due to its thermal characteristic it may heat up processor so
> much that eventually a hardware DTM or a frequency scaling occurs and
> the system slows down as a effect.

This is properly the domain of CPU governors, not schedulers. Modern Intel
chipsets with the Turbo feature will dynamically up-throttle to whatever the
cooling budget allows, and if you have an older chipset that doesn't do that,
you're better off creating a custom governor that doesn't allow selection of
the highest CPU frequenc(y|ies) that exceed the current cooling budget

Note that this gets tricky on multi-core sockets, as activity on other
cores affects the cooling budget for the core you're scheduling onto.

And if your hardware is throwing actual thermal shutdown events, you have
bigger problems that scheduler hacks won't help.



[-- Attachment #1.2: Type: application/pgp-signature, Size: 848 bytes --]

[-- Attachment #2: Type: text/plain, Size: 169 bytes --]

_______________________________________________
Kernelnewbies mailing list
Kernelnewbies@kernelnewbies.org
http://lists.kernelnewbies.org/mailman/listinfo/kernelnewbies