Cgroups RT scheduling

Cgroups RT scheduling

[Rolando Martins]

Hi,

I would like to confirm the following:
cpuset.sched_load_balance doesn't work with RT, right?
You cannot have tasks for sub-domain 2 to utilize bandwidth of
sub-domain 3, right?

                              __1__
                             /        \
                            2         3
                      (50% rt)  (50% rt )

For my application domain;) it would be interesting to have
rt_runtime_ns as a min. of allocated rt and not a max.
Ex. If an application of domain 2 needs to go up to 100% and domain 3
is idle, then it would be cool to let it utilize the full bandwidth.
(we also could have a hard upper limit in each sub-domain, like
hard_up=0.8, i.e. even if we could get 100%, we will only utilize
80%).

Thanks,
Rolando Martins

Re: Cgroups RT scheduling

[Dhaval Giani]

[Adding peterz to the cc]

On Wed, Sep 09, 2009 at 04:49:52PM +0100, Rolando Martins wrote:
> Hi,
> 
> I would like to confirm the following:
> cpuset.sched_load_balance doesn't work with RT, right?
> You cannot have tasks for sub-domain 2 to utilize bandwidth of
> sub-domain 3, right?
> 
>                               __1__
>                              /        \
>                             2         3
>                       (50% rt)  (50% rt )
> 
> For my application domain;) it would be interesting to have
> rt_runtime_ns as a min. of allocated rt and not a max.
> Ex. If an application of domain 2 needs to go up to 100% and domain 3
> is idle, then it would be cool to let it utilize the full bandwidth.
> (we also could have a hard upper limit in each sub-domain, like
> hard_up=0.8, i.e. even if we could get 100%, we will only utilize
> 80%).
> 

As far as I understand, all RT group scheduling assures is a certain
bandwidth which you cannot exceed in a given time period. It doesn't
provide the guarantee (that I think you are looking for) that you want.

Peter should be able to tell more about future plans.

Thanks,
-- 
regards,
Dhaval

Re: Cgroups RT scheduling

[Peter Zijlstra]

On Fri, 2009-09-11 at 16:53 +0530, Dhaval Giani wrote:
> [Adding peterz to the cc]
> 
> On Wed, Sep 09, 2009 at 04:49:52PM +0100, Rolando Martins wrote:
> > Hi,
> > 
> > I would like to confirm the following:
> > cpuset.sched_load_balance doesn't work with RT, right?

It does.

> > You cannot have tasks for sub-domain 2 to utilize bandwidth of
> > sub-domain 3, right?

Depends on what you do, for some contorted reason we have 2 cpu control
groups, cpusets and this other thing.

I'm still wanting to fold them into one, this confusion is in part
because of that train-wreck.

You can place a task in one cpuset cgroup and in another scheduler
cgroup - but yeah, that does interact funny with the load-balancer.

> > 
> >                               __1__
> >                              /        \
> >                             2         3
> >                       (50% rt)  (50% rt )
> > 
> > For my application domain;) it would be interesting to have
> > rt_runtime_ns as a min. of allocated rt and not a max.
> > Ex. If an application of domain 2 needs to go up to 100% and domain 3
> > is idle, then it would be cool to let it utilize the full bandwidth.
> > (we also could have a hard upper limit in each sub-domain, like
> > hard_up=0.8, i.e. even if we could get 100%, we will only utilize
> > 80%).

This doesn't sound a like a real-time application.

Variable bandwidth is useless for determinism.