[ltp] results of test ahci vs piix on t60p

[Brad Langhorst]

I read the discussion last september saying that piix was superior to
ahci mode...

Now I'm building 2.6.20.1 and see that sata support is now broken out
from ide and scsi and there are explicit options for AHCI and
intel/piix.

I've just built and installed a version with piix compiled in, which
works and suspend/resumes fine. I also tried an AHCI compiled in version
which also seems to work fine 

I did a few (not very careful) tests to see which was better.

radio off, brightness all the way up, on battery

PIIX, bios compatibility mode
~20W idle ~25W running updatedb
bonnie++
Version  1.03       ------Sequential Output------ --Sequential Input- --Random-
                    -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
Machine        Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP  /sec %CP
upup             4G 29591  70 29831   4 14812   2 35142  82 34557   3 105.4   0
                    ------Sequential Create------ --------Random Create--------
                    -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
              files  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP
                 16   317   1 +++++ +++   334   1   336   1 +++++ +++   224   0

AHCI, bios AHCI mode
~17W idle ~24W running updatedb
bonnie++
Version  1.03       ------Sequential Output------ --Sequential Input- --Random-
                    -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
Machine        Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP  /sec %CP
upup             4G 30080  71 31065   4 15109   4 36760  88 38148   7 121.5   0
                    ------Sequential Create------ --------Random Create--------
                    -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
              files  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP
                 16   297   2 +++++ +++   339   2   334   3 +++++ +++   261   1


Looks like AHCI is a marginally better choice to me, but a bit more cpu
intensive (which does not jibe with the power measurements).

Does somebody else have some observation?
Maybe a critical flaw in my methodology?

Brad