Re: [PATCH 0/2] locking/mutex: Enable optimistic spinning of lock waiter

[Waiman Long <waiman.long@hpe.com>]

On 02/09/2016 04:44 PM, Jason Low wrote:
> On Tue, 2016-02-09 at 14:47 -0500, Waiman Long wrote:
>> This patchset is a variant of PeterZ's "locking/mutex: Avoid spinner
>> vs waiter starvation" patch. The major difference is that the
>> waiter-spinner won't enter into the OSQ used by the spinners. Instead,
>> it will spin directly on the lock in parallel with the queue head
>> of the OSQ. So there will be a bit more cacheline contention on the
>> lock cacheline, but that shouldn't cause noticeable impact on system
>> performance.
>>
>> This patchset tries to address 2 issues with Peter's patch:
>>
>>   1) Ding Tianhong still find that hanging task could happen in some cases.
>>   2) Jason Low found that there was performance regression for some AIM7
>>      workloads.
> This might help address the hang that Ding reported.
>
> Performance wise, this patchset reduced AIM7 fserver throughput on the 8
> socket machine by -70%+ at 1000+ users.
>
>                  | fserver JPM
> -----------------------------
> baseline	| ~450000
> Peter's patch	| ~410000
> This patchset	| ~100000
>
> My guess is that waiters spinning/acquiring the lock is less efficient,
> and this patchset further increases the chance for waiters to
> spin/acquire the lock over the fastpath optimistic spinners.
>
> Jason
>

That was just a configuration error as the CPU scaling governor wasn't 
set to performance. With the performance scaling governor, the 
patchset's performance was comparable to Peter's patch.

Cheers,
Longman