Configuring Windows Server 2008 Power Parameters for Increased Power Efficiency

re: Configuring Windows Server 2008 Power Parameters for Increased Power Efficiency

winsrvperf — Tue, 07 Jul 2009 13:11:12 GMT

Indeed, you're right! Thanks for the sharp eyes, Asmus. I corrected this inline in the article.

~Matt Robben

re: Configuring Windows Server 2008 Power Parameters for Increased Power Efficiency

Asmus — Fri, 03 Jul 2009 22:10:52 GMT

Thanks for the excellent article!

If I understand the powercfg syntax correctly, I believe there's an error above under "Changing P-State Parameters with Powercfg.exe"

In the first few sample commands, you're working on the index value "1" -- which I believe stands for a "Power Saver" policy, running on DC/battery. For example:

powercfg /getpossiblevalue sub_processor procperf 1

However, when you apply the changes later, you switch to Index value 2, which is the Balanced policy (AC power) that I believe you were actually intending to change:

>powercfg /setpossiblevalue /sub_processor /procperf 2 binary (and so on)

The above commands do work, but users will have no record of what the value was in their "Balanced" power setting before making the changes, since that value is not output beforehand in your example.

Or did I misunderstand something?

re: Configuring Windows Server 2008 Power Parameters for Increased Power Efficiency

winsrvperf — Wed, 29 Apr 2009 02:35:05 GMT

Good question, new learner - the underlying algorithm and how these percentages are interpreted is a bit confusing. CPU usage does not map directly to the increase and decrease percentages. I want to avoid getting into detailed specifics (see footnote under table 1) as our algorithm for determining P-state changes is proprietary. But a bit more detail is appropriate. :)

When Windows boots, the OS and hardware communicate P-state availability through ACPI. The OS builds a table of performance states as a percentage of maximum processor frequency.

The algorithm uses the current utilization, current p-state, and next highest and lowest P-states (if any) to determine if a state switch should occur. Because we are determining the increase and decrease thresholds using the P-states above and below the current state, the algorithm can be designed so that there is no competition between the increase and decrease thresholds.

So in the scenario you're referring to, a CPU utilization of 40% means nothing by itself. The algorithm would need to evaluate its current P-state and surrounding P-states to determine if a change is necessary using the increase and decrease percentages specified by the user.

Note that 40% utilization != 40% of maximum frequency. This raises another issue, that % utilization is actually dependent on processor state. 40% utilization at P0 means something very different from 40% utilization at Pn (P0 has more cycles per unit time than Pn, so Pn has higher % utilization for the same workload). We are targeting this topic for a future blog post. Stay tuned and thanks for reading!

re: Configuring Windows Server 2008 Power Parameters for Increased Power Efficiency

new learner — Tue, 28 Apr 2009 09:44:22 GMT

The default parameter set increase percent-30% and decrease percent-50%. What will happen if the CPU usage is 40%? If there will be any competition?

Configuring Windows Server 2008 Power Parameters for Increased Power Efficiency

Software Enabled Earth — Wed, 10 Dec 2008 08:59:45 GMT

The Windows Server performance team has posted a great blog on best practices for tuning Windows Server

Windows Server 2008: Configuring Power Parameters for Increased Power Efficiency

Windows Server 2008: Configuring Power Parameters for Increased Power Efficiency — Fri, 05 Dec 2008 10:53:17 GMT

PingBack from http://www.ditii.com/2008/12/04/windows-server-2008-configuring-power-parameters-for-increased-power-efficiency/