[PlanetCCRMA] Performance degradation with kernel-rt-3.14.17-200.rt9.1.fc20.ccrma.x86_64

[Oleg Samarin]

I've localised the problem. It has been raised by the commit
https://github.com/torvalds/linux/commit/c4ee841f602e5eef8eab673295c49c5b49d7732b

Seems the idea of sample time scaling is not good for real timing. So
intel_pstate considers a much smaller value of busy time and it is not
sufficient for increasing the P-state.

We can see it with the test results.

3.14.17
Core [core-id]  :Actual Freq (Mult.)	  C0%   Halt(C1)%  C3 %   C6 %
Core 1 [0]:	  2143.82 (21.44x)      31.1    44.8    33.5    2.09
Core 2 [1]:	  1994.62 (19.95x)      47.9    60.4    10.5       1
Core 3 [2]:	  2077.24 (20.77x)      34.7    33.2    44.7       1
Core 4 [3]:	  2284.95 (22.85x)      45.4    29.5    34.3    5.74

3.12.12
Core [core-id]  :Actual Freq (Mult.)	  C0%   Halt(C1)%  C3 %   C6 %
Core 1 [0]:	  4502.27 (45.02x)	28.8      42    18.8       1
Core 2 [1]:	  4503.14 (45.03x)	24.7    52.3      14       1
Core 3 [2]:	  4524.26 (45.24x)	16.1      20    57.5       1
Core 4 [3]:	  4521.34 (45.21x)	69.8       0    11.9       1

So smaller values of C0% increase pstate with 3.12.12 and do not with 3.14.17.

The small patch attached reverts the change partially.

It fully solves my problem. Here is a test result:

3.14.17 with intel_pstate.3.14-disable_sample_scale.patch applied:
Core [core-id]  :Actual Freq (Mult.)	  C0%   Halt(C1)%  C3 %   C6 %
Core 1 [0]:	  4496.32 (44.98x)      47.8    17.3    18.5       1
Core 2 [1]:	  4522.05 (45.23x)      17.2    36.8    38.2    2.04
Core 3 [2]:	  4503.77 (45.05x)      27.7    27.6      34    1.75
Core 4 [3]:	  4512.65 (45.14x)        63       0    20.5    1.04

When I do not play GO, the pstates are relatively low 

Core [core-id]  :Actual Freq (Mult.)	  C0%   Halt(C1)%  C3 %   C6 %
Core 1 [0]:	  2252.28 (22.52x)         1    0.358      0    99.4
Core 2 [1]:	  1898.29 (18.98x)         1    24.6       0    74.9
Core 3 [2]:	  1983.99 (19.84x)         1    1.26       1    97.3
Core 4 [3]:	  2192.11 (21.92x)      1.83    1.42       1    96.4

that is better tan with min_perf_pct=100.

Could you include this patch in the next release of kernel-rt?

-------------- next part --------------
A non-text attachment was scrubbed...
Name: intel_pstate.3.14-disable_sample_scale.patch
Type: text/x-patch
Size: 1187 bytes
Desc: not available
Url : http://ccrma-mail.stanford.edu/pipermail/planetccrma/attachments/20140912/98ec7ef0/attachment.bin