Re: [PATCH v2] mm/slub: improve performance by skipping checked node in get_any_partial()

[Wei Yang <richard.weiyang@gmail.com>]

On Fri, Dec 21, 2018 at 01:37:38AM +0000, Christopher Lameter wrote:
>On Thu, 20 Dec 2018, Andrew Morton wrote:
>
>>   The result of (get_partial_count / get_partial_try_count):
>>
>>    +----------+----------------+------------+-------------+
>>    |          |       Base     |    Patched |  Improvement|
>>    +----------+----------------+------------+-------------+
>>    |One Node  |       1:3      |    1:0     |      - 100% |
>
>If you have one node then you already searched all your slabs. So we could
>completely skip the get_any_partial() functionality in the non NUMA case
>(if nr_node_ids == 1)
>

Yes, agree.

>
>>    +----------+----------------+------------+-------------+
>>    |Four Nodes|       1:5.8    |    1:2.5   |      -  56% |
>>    +----------+----------------+------------+-------------+
>
>Hmm.... Ok but that is the extreme slowpath.
>
>>    Each version/system configuration combination has four round kernel
>>    build tests. Take the average result of real to compare.
>>
>>    +----------+----------------+------------+-------------+
>>    |          |       Base     |   Patched  |  Improvement|
>>    +----------+----------------+------------+-------------+
>>    |One Node  |      4m41s     |   4m32s    |     - 4.47% |
>>    +----------+----------------+------------+-------------+
>>    |Four Nodes|      4m45s     |   4m39s    |     - 2.92% |
>>    +----------+----------------+------------+-------------+
>
>3% on the four node case? That means that the slowpath is taken
>frequently. Wonder why?

Hmm... not sure.

>
>Can we also see the variability? Since this is a NUMA system there is
>bound to be some indeterminism in those numbers.

Oops, I have deleted those raw data. I need to retest this.

-- 
Wei Yang
Help you, Help me