Re: [GIT PULL] Locking changes for v4.2

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

On 06/26/2015 05:11 AM, Eric Dumazet wrote:
> On Mon, 2015-06-22 at 09:30 +0200, Ingo Molnar wrote:
>> Linus,
>>
>> Please pull the latest locking-core-for-linus git tree from:
>>
>>     git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git locking-core-for-linus
>>
>>     # HEAD: 68722101ec3a0e179408a13708dd020e04f54aab locking/lockdep: Remove hard coded array size dependency
>>
>> The main changes are:
>>
>>   - 'qspinlock' support, enabled on x86: queued spinlocks - these are now the
>>     spinlock variant used by x86 as they outperform ticket spinlocks in every
>>     category. (Waiman Long)
> Very interesting.
>
> While rebasing http://www.kernelhub.org/?msg=737214&p=2
>
> I found that indeed qspinlock improved the scalability issue by a big
> factor.
>
> linux-4.1
>
> lpaa23:~# pipebench -n 16
> threads:16 sleep:30 pinned:1
> total operations: 6977349, ops/sec 232578
> 6977349
>
> lpaa23:~# pipebench -n 16
> threads:16 sleep:30 pinned:1
> total operations: 7326280, ops/sec 244209
> 7326280
>
>
> Current Linus tree :
>
> lpaa23:~# pipebench -n 16
> threads:16 sleep:30 pinned:1
> total operations: 15640802, ops/sec 521360
> 15640802
>
> lpaa23:~# pipebench -n 16
> threads:16 sleep:30 pinned:1
> total operations: 15045022, ops/sec 501500
> 15045022
>
>
> Adding fd_install() patch then :
>
> lpaa23:~# pipebench -n 16
> threads:16 sleep:30 pinned:1
> total operations: 21471043, ops/sec 715701
> 21471043
>
> lpaa23:~# pipebench -n 16
> threads:16 sleep:30 pinned:1
> total operations: 21068501, ops/sec 702283
> 21068501

I am glad that the qspinlock patch helps. May I know what kind of 
machine (cpu type, # of sockets, etc) that you are running the test on? 
Usually the more sockets the machine has, the bigger the improvement you 
will see.

Cheers,
Longman