linux-kernel - Re: [tip:sched/core] sched/fair: Correct unit of load_above

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: <20160523202401.GA18670@intel.com>
Date:	Tue, 24 May 2016 04:24:01 +0800
From:	Yuyang Du <yuyang.du@...el.com>
To:	Morten Rasmussen <morten.rasmussen@....com>
Cc:	Vincent Guittot <vincent.guittot@...aro.org>,
	Peter Zijlstra <peterz@...radead.org>,
	linux-kernel <linux-kernel@...r.kernel.org>,
	Mike Galbraith <efault@....de>, Ingo Molnar <mingo@...nel.org>,
	Dietmar Eggemann <dietmar.eggemann@....com>,
	Wanpeng Li <wanpeng.li@...mail.com>
Subject: Re: [tip:sched/core] sched/fair: Correct unit of load_above_capacity

On Thu, May 19, 2016 at 04:36:38PM +0100, Morten Rasmussen wrote:
> 
> And this is exactly you get with this patch :-) load_above_capacity
> (through max_pull) is multiplied by the group capacity to compute that
> actual amount of [load] to remove:
> 
> env->imbalance 	= load_above_capacity * busiest->group_capacity /
> 		 	SCHED_CAPACITY_SCALE
> 
> 		= 1*NICE_0_LOAD * 3*SCHED_CAPACITY_SCALE /
> 			SCHED_CAPACITY_SCALE
> 
> 		= 3*NICE_0_LOAD
> 
> I don't think we disagree on how it should work :-) Without the capacity
> scaling in this patch you get: 
> 
> env->imbalance = (6*SCHED_CAPACITY_SCALE - 3*SCHED_CAPACITY_SCALE) *
> 			3*SCHED_CAPACITY_SCALE / SCHED_CAPACITY_SCALE
> 
> 		= 9*SCHED_CAPACITY_SCALE
> 
> Coming back to Yuyang's question. I think it should be NICE_0_LOAD to
> ensure that the resulting imbalance has the proper unit [load].

Sorry, I'm still confused. After this patch, the unit is indeed [load], the
load same as the weight visible to the user. However, you literally compared
it with sg_lb_stats's avg_load and load_per_task, which has the unit of
load_avg, which is scaled_load_down(NICE_0_LOAD). Am I missing something?