Date: Fri, 18 Nov 2016 08:45:05 +0530
From: Viresh Kumar <viresh.kumar@linaro.org>
To: Stratos Karafotis <stratosk@semaphore.gr>
Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net>,
        "Srivatsa S. Bhat" <srivatsa.bhat@linux.vnet.ibm.com>,
        "linux-pm@vger.kernel.org" <linux-pm@vger.kernel.org>,
        LKML <linux-kernel@vger.kernel.org>
Subject: Re: [RFC][PATCH] cpufreq: governor: Change the calculation of load
 for deferred updates
Message-ID: <20161118031504.GC3110@vireshk-i7>
References: <4f7f7d9e-4b6b-a9f8-bf54-a6d99bd952c6@semaphore.gr>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <4f7f7d9e-4b6b-a9f8-bf54-a6d99bd952c6@semaphore.gr>
User-Agent: Mutt/1.5.24 (2015-08-30)
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 929
Lines: 27

On 17-11-16, 21:54, Stratos Karafotis wrote:
> Commit 18b46abd0009 ("cpufreq: governor: Be friendly towards latency-
> sensitive bursty workloads"), introduced a method to copy the calculated
> load from the previous sampling period in case of a deferred timer
> (update).
> 
> This helps on bursty workloads but generally coping the load for the
> previous measurement could be arbitrary, because of the possibly different
> nature of the new workload.
> 
> Instead of coping the load from the previous period we can calculate the
> load considering that between the two samples, the busy time is comparable
> to one sampling period. Thus:
> 
>  busy = time_elapsed - idle_time
> 
> and
> 
>  load = 100 * busy / sampling_rate;
> 
> Also, remove the 'unlikely' hint because it seems that a deferred update
> is a very common case on most modern systems.

You have any numbers to prove that this improves something ?

-- 
viresh