Power of the computer that use "saving energy mode"

Jun 27, 2012 at 10:35 AM


 I am not sure that I'm on the right place to post my question, so please tell me where I can ask if here is not the right place.

My question is not directly related to PAL but it concerns the performance of computers.

 We do some capacity analysis of computers so we affect a power to a server depending on the numbers of processor and the frequency of this one. With the percentage of use of the processor we can determine what power is used by a computer and compare it to another.

 It seems to be a good idea, but I discover that some computers have a "saving energy" mode which implies that the frequency of the processor can change.

 If a computer (A) have a frequency of 100% of the maximum frequency and use 50 % processor time

 If a computer (B) have a frequency of 50% of the maximum frequency and use 50 % processor time

 If computer (A) and computer (B) have the same power is that it means that the computer (B) uses half the power that computer (A)?

 Thank you for your help




Jun 28, 2012 at 1:09 AM

My understanding is the same as yours that when the system is at 50% of maximum frequency, then the processors are consuming 50% of the power going to the processors only. Please keep in mind that there are many factors involved such as NUMA nodes and the number of cores per socket. Windows Server 2008 R2 introduced core parking with puts a processor into the lowest (almost off) power state when not needed.

Aug 27, 2012 at 1:01 PM


On this subject i find a article talking about interpreting-cpu-utilization-for-performance-analysis

I think this article can help us to understand more things about cpu performance.


Just for information




Aug 28, 2012 at 12:56 PM

A good article. However, looking at only CPU utilization is a dangerous thing when you are looking at servers :-). The article talks about desktops which is where processor speed may be one of the issues. Although another metric I always look at with cpu is run queue length. If I have a high cpu but a low run queue it may be processor speed is an issue with the code path that is executing, or if i have high cpu and high run queue we definitely have a problem but it may not always have cpu as the root cause. Then you look for other things like i/o wait being high. So it's not as simple as just looking at one metric to ascertain performance.

I live in #7 in the article's list which is hypervisor land :-). And i totally agree on that one.....