linux-kernel - [Regression][stable] Major hackbench pipes regression due to commit 'sched/fair: Search a task from the tail of the queue'

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>] [day] [month] [year] [list]

Message-ID: <f3c31802-3276-fbf4-208f-a72b5d8c9170@linux.alibaba.com>
Date:   Fri, 26 Apr 2019 16:09:34 +0800
From:   Shile Zhang <shile.zhang@...ux.alibaba.com>
To:     urezki@...il.com
Cc:     mingo@...hat.com, peterz@...radead.org,
        linux-kernel@...r.kernel.org
Subject: [Regression][stable] Major hackbench pipes regression due to commit
 'sched/fair: Search a task from the tail of the queue'

Hi, Uladzislau

We found a major regression (more than -40%) of hackbench on 2VCPU 
(skylake) virtual machine, with LTS kernel v4.19.36. Following commit be 
bisect by test 'hackbench --thread --pipe --group 1 --loop 60000'

---

commit 93824900a2e242766f5fe6ae7697e3d7171aa234
Author: Uladzislau Rezki <urezki@...il.com>
Date:   Wed Sep 13 12:24:30 2017 +0200

     sched/fair: Search a task from the tail of the queue

---

-default: 21.048

-revert: 14.385

It said that the hackbench has slightly better performance with 40 
groups in the commit log, but it seems more performance dropped with 
less groups, such as less than 32 group. I also tried this benchmark 
with newest v5.1-rc6 kernel, same regression was found.

Could you please also have a check?

Thanks!

BRs/Shile