linux-kernel - Re: [PATCH v3] sched/fair: Introduce SIS_UTIL to search idle CPU based on sum of util

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

Message-ID: <20220512081414.GA31450@chenyu5-mobl1>
Date:   Thu, 12 May 2022 16:14:14 +0800
From:   Chen Yu <yu.c.chen@...el.com>
To:     Yicong Yang <yangyicong@...wei.com>
Cc:     Peter Zijlstra <peterz@...radead.org>,
        Vincent Guittot <vincent.guittot@...aro.org>,
        Mel Gorman <mgorman@...e.de>,
        Yicong Yang <yangyicong@...ilicon.com>,
        K Prateek Nayak <kprateek.nayak@....com>,
        Tim Chen <tim.c.chen@...el.com>,
        Chen Yu <yu.chen.surf@...il.com>,
        Ingo Molnar <mingo@...hat.com>,
        Juri Lelli <juri.lelli@...hat.com>,
        Dietmar Eggemann <dietmar.eggemann@....com>,
        Steven Rostedt <rostedt@...dmis.org>,
        Barry Song <21cnbao@...il.com>,
        Srikar Dronamraju <srikar@...ux.vnet.ibm.com>,
        Len Brown <len.brown@...el.com>,
        Ben Segall <bsegall@...gle.com>,
        Aubrey Li <aubrey.li@...el.com>,
        Abel Wu <wuyun.abel@...edance.com>,
        Zhang Rui <rui.zhang@...el.com>, linux-kernel@...r.kernel.org,
        Daniel Bristot de Oliveira <bristot@...hat.com>
Subject: Re: [PATCH v3] sched/fair: Introduce SIS_UTIL to search idle CPU
 based on sum of util_avg

On Tue, May 10, 2022 at 08:41:57PM +0800, Yicong Yang wrote:
> On 2022/4/29 2:24, Chen Yu wrote:
> > @@ -61,6 +61,7 @@ SCHED_FEAT(TTWU_QUEUE, true)
> >   * When doing wakeups, attempt to limit superfluous scans of the LLC domain.
> >   */
> >  SCHED_FEAT(SIS_PROP, true)
> > +SCHED_FEAT(SIS_UTIL, false)
> >  
> 
> I see you mentioned they're mutually exclusive in the commit, worth a comment here?
>
Yes, previously I thought it could be made mutually exclusive, and Peter has
suggested that we should make SIS_UTIL enabled by default, so later we could
remove SIS_PROP if SIS_UTIL behaves stable. So I assume there is no need to
add comment in the next version now. 
> One minor question: nr is updated in load balance so there maybe a delay because of
> interval of load balancing.
Yes, this is a good question. The default interval between two load balance is sd_weight ms,
which is 112ms in my case. This interval was a trade off to reduce cache contention. Besides,
every 1st idle CPU or the balanced CPU in one sched group within the LLC domain has the chance
to launch a periodic load balance, for example, although CPU0 and CPU1's periodic load balance
are both triggered every 112ms, CPU1 could help launch the load balance when CPU0 is not in
load balance work. Consider there are many CPUs in a LLC domain, the 'internal' to launch
the periodic load balance becomes smaller.
> Furthermore, the LLC domain may not be balanced everytime
> if the lowest domain is not LLC, like CLS->LLC. So maybe a bit more delay included.
> 
I thought every domain has its chance to launch a load balance, the difference is different
domains have different interval. No?
> The test results is fine and as expected. The improvement of netperf at a heavy load
> condition, compared to your v2 version.
>
Thanks for the test, would you mind if I add Tested-by tag?

thanks,
Chenyu 
> Thanks,
> Yicong
> 
> TCP_RR node 0-1
> threads
> 16	57559.56667	57930.03333 (+0.64%)
> 32	56373		57754.53333 (+2.45%)
> 64	18831.4		46234.76667 (+145.52%)
> 128	15658.9		19620.26667 (+25.30%)
> 256	7959.896667	8869.013333 (+11.42%)
> 
> TCP_RR node 0
> threads
> 16	58389.43333	59026.03333 (+1.09%)
> 32	23779.6		51563.33333 (+116.84%)
> 64	20514.56667	23485.63333 (+14.48%)
> 128	8202.49		9205.483333 (+12.23%)
> 256	3843.163333	4304.8      (+12.01%)
> 
> tbench4 node 0-1
>                            5.18-rc1                patched
> Hmean     1        299.02 (   0.00%)      307.73 *   2.91%*
> Hmean     2        597.88 (   0.00%)      619.10 *   3.55%*
> Hmean     4       1207.11 (   0.00%)     1239.57 *   2.69%*
> Hmean     8       2406.67 (   0.00%)     2463.63 *   2.37%*
> Hmean     16      4755.52 (   0.00%)     4979.46 *   4.71%*
> Hmean     32      9449.01 (   0.00%)     9709.59 *   2.76%*
> Hmean     64     10538.89 (   0.00%)    10727.86 *   1.79%*
> Hmean     128    13333.84 (   0.00%)    14580.63 *   9.35%*
> Hmean     256    11735.24 (   0.00%)    11737.16 (   0.02%)
> 
> tbench4 node 0
>                            5.18-rc1                patched
> Hmean     1        302.26 (   0.00%)      313.43 *   3.70%*
> Hmean     2        603.87 (   0.00%)      618.56 *   2.43%*
> Hmean     4       1213.91 (   0.00%)     1249.63 *   2.94%*
> Hmean     8       2469.72 (   0.00%)     2527.48 *   2.34%*
> Hmean     16      4980.70 (   0.00%)     5099.62 *   2.39%*
> Hmean     32      9001.88 (   0.00%)     9730.27 *   8.09%*
> Hmean     64      7032.07 (   0.00%)     7691.56 *   9.38%*
> Hmean     128     6037.76 (   0.00%)     6712.86 *  11.18%*
> Hmean     256     8513.83 (   0.00%)     9117.79 *   7.09%*
>