Message-ID: <1338128955.7400.43.camel@marge.simpson.net>
Subject: Re: [rfc][patch] select_idle_sibling() inducing bouncing on westmere
From: Mike Galbraith <efault@gmx.de>
To: Arjan van de Ven <arjan@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>, lkml <linux-kernel@vger.kernel.org>,
        Suresh Siddha <suresh.b.siddha@intel.com>,
        Paul Turner <pjt@google.com>,
        Andreas Herrmann <andreas.herrmann3@amd.com>
Date: Sun, 27 May 2012 16:29:15 +0200
In-Reply-To: <4FC2362A.8080503@linux.intel.com>
References: <1337857490.7300.19.camel@marge.simpson.net>
	 <1337865431.9783.148.camel@laptop> <1337865641.9783.149.camel@laptop>
	 <1337926468.5415.48.camel@marge.simpson.net>
	 <1338014259.7302.26.camel@marge.simpson.net>
	 <1338017364.14636.9.camel@twins>
	 <1338020834.7747.8.camel@marge.simpson.net>
	 <1338110259.7678.77.camel@marge.simpson.net>
	 <4FC2362A.8080503@linux.intel.com>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: 7bit
Mime-Version: 1.0
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 1835
Lines: 44

On Sun, 2012-05-27 at 07:11 -0700, Arjan van de Ven wrote: 
> On 5/27/2012 2:17 AM, Mike Galbraith wrote:
> > On Sat, 2012-05-26 at 10:27 +0200, Mike Galbraith wrote: 
> >> Hohum, back to finding out what happened to cpufreq.
> > 
> > Answer: nothing.. in mainline.
> > 
> > I test performance habitually, so just never noticed how bad ondemand
> > sucks.  In enterprise, I found the below, explaining why cores crank up
> > fine there, but not in mainline.  Somebody thumped ondemand properly on
> > it's pointy head.
> > 
> > But, check out the numbers below this, and you can see just how horrible
> > bouncing is when you add governor latency _on top_ of it. 
> 
> part of it is not ondemand, but cpufreq.
> cpufreq forces you to schedule a kernel thread to change cpu
> frequency... on the cpu that's already busy.
> God knows what the scehduler then does in terms of load balancing.

Well, it'll take a spot that could have been used to authorize an affine
wakeup for one, switch freqs a tad too late if it doesn't preempt, not
to mention munching valuable cycles.

> (yes this is one of the things that will be fixed in the code that we
> now have working internally, and we're now making sure does not regress)

Cool.

> btw, on modern Intel CPUs, where in Idle, you have a frequency of Zero,
> regardless of what you ask for when you're running, and where an idle
> cpu is clock gated.... the performance governor behaves almost the same
> as ondemand in terms of power (due to the suckitude of ondemand)... but
> much better performance.

That sounds like it should rock.

-Mike

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/