linux-kernel - Re: [locking/rwsem] 4f23dbc1e6: reaim.jobs_per

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]

Date:   Mon, 24 Jun 2019 15:45:03 -0400
From:   Waiman Long <longman@...hat.com>
To:     kernel test robot <rong.a.chen@...el.com>
Cc:     Ingo Molnar <mingo@...nel.org>,
        Peter Zijlstra <peterz@...radead.org>,
        Borislav Petkov <bp@...en8.de>,
        Davidlohr Bueso <dave@...olabs.net>,
        "H. Peter Anvin" <hpa@...or.com>,
        Linus Torvalds <torvalds@...ux-foundation.org>,
        Thomas Gleixner <tglx@...utronix.de>,
        Tim Chen <tim.c.chen@...ux.intel.com>,
        Will Deacon <will.deacon@....com>,
        huang ying <huang.ying.caritas@...il.com>,
        LKML <linux-kernel@...r.kernel.org>, tipbuild@...or.com,
        lkp@...org
Subject: Re: [locking/rwsem] 4f23dbc1e6: reaim.jobs_per_min -37.0% regression

On 6/24/19 1:45 AM, kernel test robot wrote:
> Greeting,
>
> FYI, we noticed a -37.0% regression of reaim.jobs_per_min due to commit:
>
>
> commit: 4f23dbc1e657951e5d94c60369bc1db065961fb3 ("locking/rwsem: Implement lock handoff to prevent lock starvation")
> https://git.kernel.org/cgit/linux/kernel/git/tip/tip.git locking/core
>
> in testcase: reaim
> on test machine: 48 threads Intel(R) Xeon(R) CPU E5-2697 v2 @ 2.70GHz with 64G memory
> with following parameters:
>
> 	runtime: 300s
> 	nr_task: 7000t
> 	test: shared_memory
> 	cpufreq_governor: performance
> 	ucode: 0x42e
>
> test-description: REAIM is an updated and improved version of AIM 7 benchmark.
> test-url: https://sourceforge.net/projects/re-aim-7/

With 7000 users, there will be extreme contention on the rwsems. The
lock handoff patch is known to reduce throughput with such level of lock
contention. This is a tradeoff between throughput and fairness. I don't
think this is a problem unless some real world workloads also have
significant regression in performance. I will try to reproduce the
regression and see if we can do something to reduce the performance
regression.

Cheers,
Longman