Message-ID: <1503118674.5112.20.camel@gmx.de>
Subject: Re: [PATCH v2] sched/fair: Make PELT signal more accurate
From: Mike Galbraith <efault@gmx.de>
To: Joel Fernandes <joelaf@google.com>, linux-kernel@vger.kernel.org
Cc: Vincent Guittot <vincent.guittot@linaro.org>,
        Peter Zijlstra <peterz@infradead.org>, Juri Lelli <juri.lelli@arm.com>,
        Brendan Jackman <brendan.jackman@arm.com>,
        Dietmar Eggeman <dietmar.eggemann@arm.com>, kernel-team@android.com,
        Ingo Molnar <mingo@redhat.com>
Date: Sat, 19 Aug 2017 06:57:54 +0200
In-Reply-To: <20170818235026.618-1-joelaf@google.com>
References: <20170818235026.618-1-joelaf@google.com>
Content-Type: text/plain; charset="UTF-8"
Mime-Version: 1.0
Content-Transfer-Encoding: 8BIT
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 1170
Lines: 21

On Fri, 2017-08-18 at 16:50 -0700, Joel Fernandes wrote:
> The PELT signal (sa->load_avg and sa->util_avg) are not updated if the amount
> accumulated during a single update doesn't cross a period boundary. This is
> fine in cases where the amount accrued is much smaller than the size of a
> single PELT window (1ms) however if the amount accrued is high then the
> relative error (calculated against what the actual signal would be had we
> updated the averages) can be high - as much 2% in my testing. On plotting
> signals, I found that there are errors especially high when we update just
> before the period boundary is hit. These errors can be significantly reduced if
> we update the averages more often.
> 
> Inorder to fix this, this patch does the average update by also checking how
> much time has elapsed since the last update and update the averages if it has
> been long enough (as a threshold I chose 512us).

Ok, I gotta ask:  In order to fix what?  What exactly does the small
but existent overhead increase buy us other than an ever so slightly
different chart?  What is your motivation to care about a microscopic
change in signal shape?

	-Mike