Re: [RFC PATCH V3 1/1] sched/numa: Fix disjoint set vma scan regression

[Raghavendra K T <raghavendra.kt@amd.com>]

On 7/16/2023 7:47 PM, Oliver Sang wrote:
> hi, Raghavendra K T,
> 
> On Wed, Jul 05, 2023 at 11:18:37AM +0530, Raghavendra K T wrote:
>> On 5/31/2023 9:55 AM, Raghavendra K T wrote:
>>>    With the numa scan enhancements [1], only the threads which had previously
>>> accessed vma are allowed to scan.
>>>
>>> While this had improved significant system time overhead, there were corner
>>> cases, which genuinely need some relaxation. For e.g.,
>>>
>>> 1) Concern raised by PeterZ, where if there are N partition sets of vmas
>>> belonging to tasks, then unfairness in allowing these threads to scan could
>>> potentially amplify the side effect of some of the vmas being left
>>> unscanned.
>>>
>>> 2) Below reports of LKP numa01 benchmark regression.
>>>
>>> Currently this is handled by allowing first two scanning unconditional
>>> as indicated by mm->numa_scan_seq. This is imprecise since for some
>>> benchmark vma scanning might itself start at numa_scan_seq > 2.
>>>
>>> Solution:
>>> Allow unconditional scanning of vmas of tasks depending on vma size. This
>>> is achieved by maintaining a per vma scan counter, where
>>>
>>> f(allowed_to_scan) = f(scan_counter <  vma_size / scan_size)
>>>
>>> Result:
>>> numa01_THREAD_ALLOC result on 6.4.0-rc2 (that has numascan enhancement)
>>>                   	base-numascan	base		base+fix
>>> real    		1m1.507s	1m23.259s	1m2.632s
>>> user    		213m51.336s	251m46.363s	220m35.528s
>>> sys     		3m3.397s	0m12.492s	2m41.393s
>>>
>>> numa_hit 		5615517		4560123		4963875
>>> numa_local 		5615505		4560024		4963700
>>> numa_other 		12		99		175
>>> numa_pte_updates 	1822797		493		1559111
>>> numa_hint_faults 	1307113		523		1469031
>>> numa_hint_faults_local 	612617		488		884829
>>> numa_pages_migrated 	694370		35		584202
>>>
>>> Summary: Regression in base is recovered by allowing scanning as required.
>>>
>>> [1] https://lore.kernel.org/lkml/cover.1677672277.git.raghavendra.kt@amd.com/T/#t
>>>
>>> Fixes: fc137c0ddab2 ("sched/numa: enhance vma scanning logic")
>>> regression.
>>> Reported-by: Aithal Srikanth <sraithal@amd.com>
>>> Reported-by: kernel test robot <oliver.sang@intel.com>
>>> Closes: https://lore.kernel.org/lkml/db995c11-08ba-9abf-812f-01407f70a5d4@amd.com/T/
>>> Signed-off-by: Raghavendra K T <raghavendra.kt@amd.com>
>>
>> Hello kernel test robot,
>>
>> Gentle ping to check if the patch has helped your regression report.
> 
> sorry for late.
> 
> previously we found a 118.9% regression of autonuma-benchmark.numa01.seconds
> on a Cascade Lake, which happened to be converted for other test purposes, so
> we cannot test your patch on it again.
> 
> however, we also found a 39.3% regression on a Sapphire Rapids test machine:
> 
> =========================================================================================
> compiler/cpufreq_governor/iterations/kconfig/rootfs/tbox_group/test/testcase:
>    gcc-11/performance/4x/x86_64-rhel-8.3/debian-11.1-x86_64-20220510.cgz/lkp-spr-r02/numa02_SMT/autonuma-benchmark
> 
> ef6a22b70f6d9044 fc137c0ddab29b591db6a091dc6
> ---------------- ---------------------------
>           %stddev     %change         %stddev
>               \          |                \
>      193.14           +39.2%     268.84        autonuma-benchmark.numa01.seconds
>        8.14            -0.7%       8.09        autonuma-benchmark.numa02.seconds
> 
> 
> now we tested your v3 patch on it, found regression mostly recovered
> (55fd15913b18d6a790c17d947df is just [RFC PATCH V3 1/1] sched/numa: Fix disjoint set vma scan regression)
> 
> =========================================================================================
> compiler/cpufreq_governor/iterations/kconfig/rootfs/tbox_group/test/testcase:
>    gcc-11/performance/4x/x86_64-rhel-8.3/debian-11.1-x86_64-20220510.cgz/lkp-spr-r02/numa01_THREAD_ALLOC/autonuma-benchmark
> 
> ef6a22b70f6d9044 55fd15913b18d6a790c17d947df
> ---------------- ---------------------------
>           %stddev     %change         %stddev
>               \          |                \
>      193.14            +5.8%     204.37 ±  3%  autonuma-benchmark.numa01.seconds
>        8.14            -0.9%       8.06        autonuma-benchmark.numa02.seconds
> 
> detail comparison as below:
> 

Thank you for the confirmation. So looks like we got back most of the
regression with the patch for numa01_THREAD_ALLOC case.

Andrew, could you please help by picking up this patch unless
Mel , PeterZ do not have any concern about the patch / direction.

(as we note it brings back little bit system time overhead by allowing
some scanning..)

Thanks and Regards
- Raghu