Re: [RFC/PATCH 01/17] sched: mix tasks and groups

[Srivatsa Vaddagiri <vatsa@linux.vnet.ibm.com>]

On Sun, Mar 09, 2008 at 06:08:51PM +0100, Peter Zijlstra wrote:
> This patch allows tasks and groups to exist in the same cfs_rq. With this
> change the CFS group scheduling follows a 1/(M+N) model from a 1/(1+N)
> fairness model where M tasks and N groups exist at the cfs_rq level.
> 
> [a.p.zijlstra@chello.nl: rt bits]
> Signed-off-by: Dhaval Giani <dhaval@linux.vnet.ibm.com>
> Signed-off-by: Srivatsa Vaddagiri <vatsa@linux.vnet.ibm.com>
> Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
> ---
>  kernel/sched.c      |   54 ++++++++++++++++++++++++++++++++++++++++++++++++++--
>  kernel/sched_fair.c |   48 +++++++++++++++++++++++++++++++++++++++++++---
>  kernel/sched_rt.c   |   15 ++++++++------
>  3 files changed, 106 insertions(+), 11 deletions(-)
> 
> Index: linux-2.6-2/kernel/sched.c
> ===================================================================
> --- linux-2.6-2.orig/kernel/sched.c
> +++ linux-2.6-2/kernel/sched.c
> @@ -273,18 +273,23 @@ struct task_group {
>  };
> 
>  #ifdef CONFIG_FAIR_GROUP_SCHED
> +
> +#ifdef CONFIG_USER_SCHED
>  /* Default task group's sched entity on each cpu */
>  static DEFINE_PER_CPU(struct sched_entity, init_sched_entity);
>  /* Default task group's cfs_rq on each cpu */
>  static DEFINE_PER_CPU(struct cfs_rq, init_cfs_rq) ____cacheline_aligned_in_smp;
> +#endif


I am trying to understand the change this brings for semantics of RT-scheduling.

With this change, /cgroup will be seen as the parent group of all other
groups (say: /cgroup/A, /cgroup/B etc). Is that correct?

If so, the check in __rt_schedulable() needs a change as well, which assumes 
that all task groups form a flat hierarchy.

For example: lets say that init_task_group (/cgroup in this case) had the
default rt_bandwidth of 95% (global_rt_runtime()). A child group under it 
(/cgroup/A) is created. If user tries to assign it a rt-bandwidth of
50%, then AFAICS, it will fail with current code, whereas it shouldn't
(because by giving /cgroup/A 50% bandwidth, we are not really exceeding
the globally allowed RT bandwidth of 95%, since /cgroup/A is a child of
/cgroup).

-- 
Regards,
vatsa