linux-kernel - Re: [External] : Re: [REGRESSION] sched/fair: Add lag based placement

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <04208f27-44a6-4e64-a2a7-fc1233ebf736@oracle.com>
Date: Thu, 14 Nov 2024 14:50:49 -0500
From: Joseph Salisbury <joseph.salisbury@...cle.com>
To: Phil Auld <pauld@...hat.com>
Cc: peterz@...radead.org, mingo@...nel.org, juri.lelli@...hat.com,
        vincent.guittot@...aro.org, dietmar.eggemann@....com,
        Steven Rostedt <rostedt@...dmis.org>, bsegall@...gle.com,
        mgorman@...e.de, vschneid@...hat.com, linux-kernel@...r.kernel.org
Subject: Re: [External] : Re: [REGRESSION] sched/fair: Add lag based placement




On 11/13/24 13:33, Joseph Salisbury wrote:
>
>
>
> On 11/13/24 13:22, Joseph Salisbury wrote:
>>
>>
>>
>> On 11/13/24 13:19, Phil Auld wrote:
>>> Hi,
>>>
>>> On Wed, Nov 13, 2024 at 01:03:00PM -0500 Joseph Salisbury wrote:
>>>> Hello,
>>>>
>>>> During performance testing, we found a regression of ~9% 
>>>> performance with
>>>> the TPCC benchmark.   This performance regression was introduced in
>>>> v6.6-rc1.  After a bisect, the following commit was identified as 
>>>> the cause
>>>> of the regression:
>>>>
>>>> 86bfbb7ce4f6 ("sched/fair: Add lag based placement")
>>>>
>>>> I was hoping to get some feedback from the scheduler folks. Do you 
>>>> think
>>>> gathering any additional data will help diagnose this issue? Are 
>>>> there any
>>>> tunable options that can changed to see how performance is affected?
>>>>
>>> You can try turning off the PLACE_LAG sched feature:
>>>
>>>      echo NO_PLACE_LAG > /sys/kernel/debug/sched/features
>>>
>>> It's not what I'd call a tunable but it would allow you to test w/o 
>>> it and
>>> see what it does.  It should allow you to switch back and forth 
>>> easily for
>>> testing.
>>>
>>>
>>> Cheers,
>>> Phil
>> Thanks so much for the suggestion, Phil!  I will give that a try and 
>> report the results.
We can confirm that using NO_PLACE_LAG adds back 5% of the performance 
that was lost.  However, we have not yet measured what effect this will 
have on other benchmarks.

We will continue testing and can help test the patches that add 
PLACE_LAG and RUN_TO_PARITY as sysctl options.
>>>
>>>> Thanks,
>>>>
>>>> Joe
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>
>>
> I just noticed this thread, which is probably related:
> https://urldefense.com/v3/__https://lore.kernel.org/lkml/ZxuujhhrJcoYOdMJ@BLRRASHENOY1.amd.com/T/__;!!ACWV5N9M2RV99hQ!MhxYsyXTgwxk1HIWrxUHGSEZcJyBENlm5apMv2TEqf6Tn2uoi14-V8YSTymPDvjax78DSQR4m6zdQiJwxJ89K8iTmWl4hvUQ$ 
>