ondemand vs conservative

[Daniel Hollocher]

Hey folks,
Anyone know about the difference between the ondemand vs conservative
frequency governors?  I'm on a core2duo dual core intel type chip, and
I am getting much better performance with conservative selected.  I
first heard about the conservative governor in the context of getting
better flash video performance, but it looks better accross the board
for me.  I will post some kraken benchmarks at the bottom so you can
see what I'm talking about.  I've also vaguely tested with some video
compression, and it looks similar in performance to me.

Should I file a bug against ondemand?  or a bug requesting the default
governor be conservative?  I really don't know.  IMHO, the performance
difference and the non-obviousness of how to change the defaults (it
looks like you have to edit the upstart script) warrant some sort of
change.

Thanks,
Dan


TEST                         COMPARISON            FROM
 TO               DETAILS

====================================================================================

** TOTAL **:                 1.25x as fast     33744.5ms +/- 2.5%
27033.1ms +/- 1.4%     significant

====================================================================================

  ai:                        1.194x as fast     1410.6ms +/- 3.9%
1181.2ms +/- 8.1%     significant
    astar:                   1.194x as fast     1410.6ms +/- 3.9%
1181.2ms +/- 8.1%     significant

  audio:                     1.25x as fast     13168.3ms +/- 2.4%
10561.9ms +/- 2.0%     significant
    beat-detection:          1.26x as fast      4669.0ms +/- 3.3%
3704.0ms +/- 3.5%     significant
    dft:                     1.098x as fast     2559.1ms +/- 3.5%
2330.6ms +/- 2.4%     significant
    fft:                     1.30x as fast      4577.2ms +/- 4.8%
3508.8ms +/- 2.3%     significant
    oscillator:              1.34x as fast      1363.0ms +/- 2.0%
1018.5ms +/- 3.3%     significant

  imaging:                   1.25x as fast     13957.9ms +/- 4.8%
11158.0ms +/- 2.2%     significant
    gaussian-blur:           1.29x as fast      6846.8ms +/- 9.6%
5296.9ms +/- 2.6%     significant
    darkroom:                1.22x as fast      3387.3ms +/- 4.0%
2783.7ms +/- 3.0%     significant
    desaturate:              1.21x as fast      3723.8ms +/- 3.6%
3077.4ms +/- 6.6%     significant

  json:                      1.31x as fast      1359.7ms +/- 3.4%
1041.7ms +/- 3.7%     significant
    parse-financial:         1.35x as fast       783.6ms +/- 4.7%
578.6ms +/- 2.9%     significant
    stringify-tinderbox:     1.24x as fast       576.1ms +/- 8.1%
463.1ms +/- 5.7%     significant

  stanford:                  1.25x as fast      3848.0ms +/- 2.6%
3090.3ms +/- 4.9%     significant
    crypto-aes:              1.26x as fast       720.0ms +/- 6.8%
569.5ms +/- 6.6%     significant
    crypto-ccm:              1.21x as fast       610.5ms +/- 8.6%
503.5ms +/- 5.2%     significant
    crypto-pbkdf2:           1.24x as fast      1876.0ms +/- 4.4%
1507.8ms +/- 6.2%     significant
    crypto-sha256-iterative: 1.26x as fast       641.5ms +/- 5.5%
509.5ms +/- 5.0%     significant