[lkml]   [2020]   [Feb]   [27]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRE: [RFC 0/1] Weighted approach to gather and use history in TEO governor
On 2020.02.24 21:13 Gautham R Shenoy wrote:


> Could you also provide power measurements for the duration when the
> system is completely idle for each of the variants of TEO governor ?
> Is it the case that the benefits that we are seeing above are only due
> to Wt. TEO being more conservative than TEO governor by always
> choosing a shallower state ?

For what it's worth:

CPU: Intel: i7-2600K
Kernel: 5.6-rc2 (teo) and + this patch set (wtteo)
Note: in general, "idle" on this system is considerably more "idle" than most systems.
Sample period: 5 minutes.
CPU scaling driver: intel_cpufreq
Governor: performance
Deepest idle state: 4 (C6)

Test duration 740 minutes (12.33 hours).
Average processor package power: 3.84 watts
Idle state 0: 4.19 / minute
Idle state 1: 29.26 / minute
Idle state 2: 46.71 / minute
Idle state 3: 7.42 / minute
Idle state 4: 1124.55 / minute
Total: 2.525 idle entries per cpu per second

Test duration 1095 minutes (18.25 hours).
Average processor package power: 3.84 watts
Idle state 0: 7.98 / minute
Idle state 1: 30.49 / minute
Idle state 2: 52.51 / minute
Idle state 3: 8.65 / minute
Idle state 4: 1125.33 / minute
Total: 2.552 idle entries per cpu per second

The above/below data for this test is incomplete because my program
doesn't process it if there are not enough state entries per sample period.
(I need to fix that for this type of test.)

I have done a couple of other tests with this patch set,
but nothing to report yet, as the differences have been minor so far.

... Doug

 \ /
  Last update: 2020-02-27 17:25    [W:0.133 / U:0.500 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site