[procps] Re: C-States handling - new switch?
- From: Jaromir Capik <jcapik@xxxxxxxxxx>
- To: procps@xxxxxxxxxxxxx
- Date: Thu, 09 Feb 2012 10:26:22 -0500 (EST)
I'm attaching an example of the dynamic frequency scaling (CPU throttling)
collected on my workstation using the "ondemand" policy (cpupower tools).
t[s] f[GHz]
0.0 1.73
0.2 1.73
0.4 1.73
0.6 2.66
0.8 1.73
1.0 1.73
1.2 1.73
1.4 1.73
1.6 1.73
1.8 1.73
2.0 1.73
2.2 1.73
2.4 1.73
2.6 1.73
2.8 1.73
3.0 1.73
3.2 1.73
3.4 2.66
3.6 2.66
3.8 1.73
4.0 1.73
4.2 1.73
4.4 2.66
4.6 2.66
4.8 1.73
5.0 1.73
As You can see, the CPU frequency is changing so fast,
that it must affect the stats negatively.
It results in something like a dynamic compression
of the load peaks, because everytime the CPU gets
under a higher load, it's trying to do a fast regulation
by changing it's frequency and that could look like there
are absolutely no load peaks whilst they are present
(but users can't see them, because they are hidden
under the scaling).
J.
Other related posts:
