Re: [RFC][Patch 02/18] sched: increase SCHED_LOAD_SCALE resolution

[Nikhil Rao]

On Thu, Apr 28, 2011 at 2:54 AM, Nikunj A. Dadhania
<nikunj@xxxxxxxxxxxxxxxxxx> wrote:
> On Wed, 20 Apr 2011 13:51:21 -0700, Nikhil Rao <ncrao@xxxxxxxxxx> wrote:
>> Introduce SCHED_LOAD_RESOLUTION, which scales is added to SCHED_LOAD_SHIFT and
>> increases the resolution of SCHED_LOAD_SCALE. This patch sets the value of
>> SCHED_LOAD_RESOLUTION to 10, scaling up the weights for all sched entities by
>> a factor of 1024. With this extra resolution, we can handle deeper cgroup
>> hiearchies and the scheduler can do better shares distribution and load
>> load balancing on larger systems (especially for low weight task groups).
>>
>> This does not change the existing user interface, the scaled weights are only
>> used internally. We do not modify prio_to_weight values or inverses, but use
>> the original weights when calculating the inverse which is used to scale
>> execution time delta in calc_delta_mine(). This ensures we do not lose accuracy
>> when accounting time to the sched entities.
>>
>> Signed-off-by: Nikhil Rao <ncrao@xxxxxxxxxx>
>> ---
>> Âinclude/linux/sched.h | Â Â3 ++-
>> Âkernel/sched.c    Â|  18 ++++++++++--------
>> Â2 files changed, 12 insertions(+), 9 deletions(-)
>>
>> diff --git a/include/linux/sched.h b/include/linux/sched.h
>> index 8d1ff2b..d2c3bab 100644
>> --- a/include/linux/sched.h
>> +++ b/include/linux/sched.h
>> @@ -794,7 +794,8 @@ enum cpu_idle_type {
>> Â/*
>> Â * Increase resolution of nice-level calculations:
>> Â */
>> -#define SCHED_LOAD_SHIFT Â Â 10
>> +#define SCHED_LOAD_RESOLUTION Â Â Â Â10
>> +#define SCHED_LOAD_SHIFT Â Â (10 + SCHED_LOAD_RESOLUTION)
>> Â#define SCHED_LOAD_SCALE Â Â (1L << SCHED_LOAD_SHIFT)
>>
>> Â/*
>> diff --git a/kernel/sched.c b/kernel/sched.c
>> index 50f97cc..bfee8ff 100644
>> --- a/kernel/sched.c
>> +++ b/kernel/sched.c
>> @@ -293,7 +293,7 @@ static DEFINE_SPINLOCK(task_group_lock);
>> Â * Âlimitation from this.)
>> Â */
>> Â#define MIN_SHARES Â 2
>> -#define MAX_SHARES Â (1UL << 18)
>> +#define MAX_SHARES Â (1UL << 28)
> Shouldn't this change depend on SCHED_LOAD_RESOLUTION?
>
> #define MAX_SHARES Â Â Â(1UL << (18 + SCHED_LOAD_RESOLUTION))
>

Right, and the other 18 bits come from SCHED_LOAD_SCALE + some slack,
which is also implicitly defined. I can make this more explicit in the
next rev as you suggest. I don't expect SCHED_LOAD_RESOLUTION to
change often though.

> Nikunj
>
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/