Re: [RFC V2 2/2] sched/fair: Fallback to sched-idle CPU if idle CPU isn't found

[Subhra Mazumdar <subhra.mazumdar@oracle.com>]

On 5/14/19 10:27 AM, Subhra Mazumdar wrote:
>
> On 5/14/19 9:03 AM, Steven Sistare wrote:
>> On 5/13/2019 7:35 AM, Peter Zijlstra wrote:
>>> On Mon, May 13, 2019 at 03:04:18PM +0530, Viresh Kumar wrote:
>>>> On 10-05-19, 09:21, Peter Zijlstra wrote:
>>>>> I don't hate his per se; but the whole select_idle_sibling() thing is
>>>>> something that needs looking at.
>>>>>
>>>>> There was the task stealing thing from Steve that looked 
>>>>> interesting and
>>>>> that would render your apporach unfeasible.
>>>> I am surely missing something as I don't see how that patchset will
>>>> make this patchset perform badly, than what it already does.
>>> Nah; I just misremembered. I know Oracle has a patch set poking at
>>> select_idle_siblings() _somewhere_ (as do I), and I just found the 
>>> wrong
>>> one.
>>>
>>> Basically everybody is complaining select_idle_sibling() is too
>>> expensive for checking the entire LLC domain, except for FB (and thus
>>> likely some other workloads too) that depend on it to kill their tail
>>> latency.
>>>
>>> But I suppose we could still do this, even if we scan only a subset of
>>> the LLC, just keep track of the last !idle CPU running only SCHED_IDLE
>>> tasks and pick that if you do not (in your limited scan) find a better
>>> candidate.
>> Subhra posted a patch that incrementally searches for an idle CPU in 
>> the LLC,
>> remembering the last CPU examined, and searching a fixed number of 
>> CPUs from there.
>> That technique is compatible with the one that Viresh suggests; the 
>> incremental
>> search would stop if a SCHED_IDLE cpu was found.
> This was the last version of patchset I sent:
> https://lkml.org/lkml/2018/6/28/810
>
> Also select_idle_core is a net -ve for certain workloads like OLTP. So I
> had put a SCHED_FEAT to be able to disable it.
Forgot to add, the cpumask_weight computation may not be O(1) with large
number of CPUs, so needs to be precomputed in a per-cpu variable to further
optimize. That part is missing from the above patchset.
>
> Thanks,
> Subhra
>>
>> I also fiddled with select_idle_sibling, maintaining a per-LLC bitmap 
>> of idle CPUs,
>> updated with atomic operations.  Performance was basically unchanged 
>> for the
>> workloads I tested, and I inserted timers around the idle search 
>> showing it was
>> a very small fraction of time both before and after my changes.  That 
>> led me to
>> ignore the push side and optimize the pull side with task stealing.
>>
>> I would be very interested in hearing from folks that have workloads 
>> that demonstrate
>> that select_idle_sibling is too expensive.
>>
>> - Steve