Re: [PATCH] thp: reduce usage of huge zero page's atomic counter

[Aaron Lu]

On 08/30/2016 11:39 AM, Andrew Morton wrote:
> On Tue, 30 Aug 2016 11:09:15 +0800 Aaron Lu <aaron.lu@xxxxxxxxx> wrote:
> 
>>>> Case used for test on Haswell EP:
>>>> usemem -n 72 --readonly -j 0x200000 100G
>>>> Which spawns 72 processes and each will mmap 100G anonymous space and
>>>> then do read only access to that space sequentially with a step of 2MB.
>>>>
>>>> perf report for base commit:
>>>>     54.03%  usemem   [kernel.kallsyms]   [k] get_huge_zero_page
>>>> perf report for this commit:
>>>>      0.11%  usemem   [kernel.kallsyms]   [k] mm_get_huge_zero_page
>>>
>>> Does this mean that overall usemem runtime halved?
>>
>> Sorry for the confusion, the above line is extracted from perf report.
>> It shows the percent of CPU cycles executed in a specific function.
>>
>> The above two perf lines are used to show get_huge_zero_page doesn't
>> consume that much CPU cycles after applying the patch.
>>
>>>
>>> Do we have any numbers for something which is more real-wordly?
>>
>> Unfortunately, no real world numbers.
>>
>> We think the global atomic counter could be an issue for performance
>> so I'm trying to solve the problem.
> 
> So, umm, we don't actually know if the patch is useful to anyone?

It should help when multiple processes are doing read only anonymous
page faults with THP enabled.

> 
> Some more measurements would help things along, please.
 
In addition to the perf cycles drop in the get_huge_zero_page function,
the throughput for the above workload also increased a lot.

usemem -n 72 --readonly -j 0x200000 100G

base commit
$ cat 7289420fc8e98999c8b7c1c2c888549ccc9aa96f/0/vm-scalability.json 
{
  "vm-scalability.throughput": [
    1784430792
  ],
}

this patch
$ cat a57acb91d1a29efc4cf34ffee09e1cebe93dcd24/0/vm-scalability.json 
{
  "vm-scalability.throughput": [
    4726928591
  ],
}

Throughput wise, it's a 164% gain.
Runtime wise, it's reduced from 707592 usecs to 303970 usecs, 50%+ drop.

Granted, real world use case may not encounter such an extreme case so
the gain would be much smaller.

Thanks,
Aaron