[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20230816092033.GB980931@hirez.programming.kicks-ass.net>
Date: Wed, 16 Aug 2023 11:20:33 +0200
From: Peter Zijlstra <peterz@...radead.org>
To: Chen Yu <yu.c.chen@...el.com>
Cc: kernel test robot <oliver.sang@...el.com>, oe-lkp@...ts.linux.dev,
lkp@...el.com, linux-kernel@...r.kernel.org, x86@...nel.org,
Ingo Molnar <mingo@...nel.org>,
Mike Galbraith <umgwanakikbuti@...il.com>
Subject: Re: [tip:sched/eevdf] [sched/fair] e0c2ff903c:
phoronix-test-suite.blogbench.Write.final_score -34.8% regression
On Wed, Aug 16, 2023 at 11:40:33AM +0800, Chen Yu wrote:
> > Are these wakeup preemptions typically on runqueues that have only a
> > single other task?
> >
>
> There are 112 groups of hackbench waker/wakee on a 112 CPUs system.
> Each group has 1 waker and 20 wakees. It seems that there are more than 1
> other task on each runqueue. I collected the statistics below.
Oh, I thought we were still talking about $subject benchmark. Yeah,
hackbench has a ton of tasks.
> According to above data, maybe comparing the deadline does not take much
> effect in this case(per 1).
Yeah, I stuck a trace_printk() in last night and quicly realized the lag
is on the same order of request size, which then totally makes the whole
thing moot.
Powered by blists - more mailing lists