From mboxrd@z Thu Jan 1 00:00:00 1970 From: Dirk Brandewie Subject: Re: powersave governor runs programs faster and uses more power than performance governor Date: Fri, 25 Oct 2013 08:38:58 -0700 Message-ID: <526A9092.5000800@gmail.com> References: <526A82CC.1040709@gmail.com> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: Sender: cpufreq-owner@vger.kernel.org To: Melanie Kambadur Cc: Viresh Kumar , David C Niemi , "cpufreq@vger.kernel.org" , Linux PM list List-Id: linux-pm@vger.kernel.org On 10/25/2013 08:13 AM, Melanie Kambadur wrote: > I appreciate you all taking the time to walk me through this. Let me > see if I understand the new comments. Intel p-states is a HW-based > power manager, Not hardware based but specific to Intel CPUs SandyBridge+ > and strictly an alternative to (i.e., it cannot be > combined with) OS governors and drivers. Correct >If I want to use ondemand with my Dell server I need to: Add intel_pstate=disable on your kernel commond line, this will take intel_pstate out of the picture. For the rest of the config on the dell system I am no help sorry. > > 1) Modify the BIOS to give the OS exclusive power management control > because otherwise an OS driver won't be able to work properly. (I > think I know how to do this now after some more reading, e.g. here > http://en.community.dell.com/techcenter/power-cooling/w/wiki/best-practices-in-power-management.aspx > if anyone is curious.) > 2) Set the O/S cpufreq driver to acpi_cpufreq, and > 3) Set the O/S cpufreq governor to ondemand. > > Is that correct? > > Also, which driver should I try to use if I want to test the > performance & powersave governors again (or if I replicate the > behavior of the performance governor by manually modifying the > min_perf_pct value as Dirk suggested)? Will it still be acpi_cpufreq? If intel_pstate is being used acpi_cpufreq will not be loaded. Setting performance with intel_pstate should work I will look to see where the bug is. > > Finally, the behavior of the C-states is totally independent of > P-states and any kind of OS-based frequency tuning policy, correct? Correct > However, David recommends that leaving C1E on rarely hurts performance > while significantly improving power. > > -Melanie > > On Fri, Oct 25, 2013 at 10:40 AM, Dirk Brandewie > wrote: >> On 10/24/2013 12:42 PM, Melanie Kambadur wrote: >>> >>> >>> From /sys/devices/system/cpu/cpuN/cpufreq/scaling_driver I get that >>> the current p-state driver is called "intel_pstate". David, you >>> mention that the firmware governors are not very efficient, do you >>> suggest replacing the intel_pstate driver with a different driver? >> >> >> I will need to look and see why changing to performance isn't working >> correctly. >> >> To get the behavior of the performance governor you can use >> >> echo 100 > /sys/devices/system/cpu/intel_pstate/min_perf_pct >> >> This will force intel_pstate to select the highest P state and >> leave it there. >> >> Turbostat is useful for collecting frequency (P state) and idle (C state) >> information. >> >> --Dirk >> >>