linux-kernel - Re: [PATCH v2 6/8] sched/idle: Move busy

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: <5823f298-6fae-5a73-3ab8-f708d90a7e52@linux.intel.com>
Date:   Fri, 14 May 2021 12:11:50 +0800
From:   Aubrey Li <aubrey.li@...ux.intel.com>
To:     Srikar Dronamraju <srikar@...ux.vnet.ibm.com>
Cc:     Ingo Molnar <mingo@...nel.org>,
        Peter Zijlstra <peterz@...radead.org>,
        LKML <linux-kernel@...r.kernel.org>,
        Mel Gorman <mgorman@...hsingularity.net>,
        Rik van Riel <riel@...riel.com>,
        Thomas Gleixner <tglx@...utronix.de>,
        Valentin Schneider <valentin.schneider@....com>,
        Vincent Guittot <vincent.guittot@...aro.org>,
        Dietmar Eggemann <dietmar.eggemann@....com>,
        Gautham R Shenoy <ego@...ux.vnet.ibm.com>,
        Parth Shah <parth@...ux.ibm.com>
Subject: Re: [PATCH v2 6/8] sched/idle: Move busy_cpu accounting to idle
 callback

On 5/13/21 3:31 PM, Srikar Dronamraju wrote:
> * Aubrey Li <aubrey.li@...ux.intel.com> [2021-05-12 16:08:24]:
> 
>> On 5/7/21 12:45 AM, Srikar Dronamraju wrote:
>>> Currently we account nr_busy_cpus in no_hz idle functions.
>>> There is no reason why nr_busy_cpus should updated be in NO_HZ_COMMON
>>> configs only. Also scheduler can mark a CPU as non-busy as soon as an
>>> idle class task starts to run. Scheduler can then mark a CPU as busy
>>> as soon as its woken up from idle or a new task is placed on it's
>>> runqueue.
>>
>> IIRC, we discussed this before, if a SCHED_IDLE task is placed on the
>> CPU's runqueue, this CPU should be still taken as a wakeup target.
>>
> 
> Yes, this CPU is still a wakeup target, its only when this CPU is busy, that
> we look at other CPUs
> 
>> Also, for those frequent context-switching tasks with very short idle,
>> it's expensive for scheduler to mark idle/busy every time, that's why
>> my patch only marks idle every time and marks busy ratelimited in
>> scheduler tick.
>>
> 
> I have tried few tasks with very short idle times and updating nr_busy
> everytime, doesnt seem to be impacting. Infact, it seems to help in picking
> the idler-llc more often.
> 

How many CPUs in your LLC?

This is a system with 192 CPUs, 4 nodes and each node has 48 CPUs in LLC
domain.

It looks like for netperf both TCP and UDP cases have the notable change
under 2 x overcommit, it may be not interesting though.


hackbench(48 tasks per group)
=========
case            	load    	baseline(std%)	compare%( std%)
process-pipe    	group-1 	 1.00 (  6.74)	 -4.61 (  8.97)
process-pipe    	group-2 	 1.00 ( 36.84)	+11.53 ( 26.35)
process-pipe    	group-3 	 1.00 ( 24.97)	+12.21 ( 19.05)
process-pipe    	group-4 	 1.00 ( 18.27)	 -2.62 ( 17.60)
process-pipe    	group-8 	 1.00 (  4.33)	 -2.22 (  3.08)
process-sockets 	group-1 	 1.00 (  7.88)	-20.26 ( 15.97)
process-sockets 	group-2 	 1.00 (  5.38)	-19.41 (  9.25)
process-sockets 	group-3 	 1.00 (  4.22)	 -5.70 (  3.00)
process-sockets 	group-4 	 1.00 (  1.44)	 -1.80 (  0.79)
process-sockets 	group-8 	 1.00 (  0.44)	 -2.86 (  0.06)
threads-pipe    	group-1 	 1.00 (  5.43)	 -3.69 (  3.59)
threads-pipe    	group-2 	 1.00 ( 18.00)	 -2.69 ( 16.79)
threads-pipe    	group-3 	 1.00 ( 21.72)	 -9.01 ( 21.34)
threads-pipe    	group-4 	 1.00 ( 21.58)	 -6.43 ( 16.26)
threads-pipe    	group-8 	 1.00 (  3.05)	 -0.15 (  2.31)
threads-sockets 	group-1 	 1.00 ( 14.51)	 -5.35 ( 13.85)
threads-sockets 	group-2 	 1.00 (  3.97)	-24.15 (  4.40)
threads-sockets 	group-3 	 1.00 (  4.97)	 -9.05 (  2.46)
threads-sockets 	group-4 	 1.00 (  1.98)	 -3.44 (  0.49)
threads-sockets 	group-8 	 1.00 (  0.37)	 -2.13 (  0.20)

netperf
=======
case            	load    	baseline(std%)	compare%( std%)
TCP_RR          	thread-48	 1.00 (  3.84)	 -2.20 (  3.83)
TCP_RR          	thread-96	 1.00 (  5.22)	 -4.97 (  3.90)
TCP_RR          	thread-144	 1.00 (  7.97)	 -0.75 (  4.39)
TCP_RR          	thread-192	 1.00 (  3.03)	 -0.67 (  4.40)
TCP_RR          	thread-384	 1.00 ( 22.27)	-14.15 ( 36.28)
UDP_RR          	thread-48	 1.00 (  2.08)	 -0.39 (  2.29)
UDP_RR          	thread-96	 1.00 (  2.48)	 -4.26 ( 16.06)
UDP_RR          	thread-144	 1.00 ( 49.50)	 -3.28 ( 34.86)
UDP_RR          	thread-192	 1.00 (  6.39)	 +8.07 ( 88.15)
UDP_RR          	thread-384	 1.00 ( 31.54)	-12.76 ( 35.98)