Mailinglist Archive: opensuse-amd64 (40 mails)

[Siegbert Baude <siegbert.baude@xxxxxx>]

Darren Davis schrieb:
> http://biz.yahoo.com/bw/061003/20061003005999.html?.v=1

Just a small remark. They write
"The test database was copied onto the test drive."
in http://www.worlds-fastest.com/disk2.pdf

And they claim correctly:

"The percentage of a disk that is used will affect the performance that
the drive delivers. A drive where the working set of data almost fills
the drive will have many seeks across the entire surface of the drive.
These long seeks are very slow in comparison to tightly clustered seeks."

So to get rid of disk effects and really test 32bit vs. 64bit, they
should have checked disk performance over all cylinders and make sure,
that the test database is always in the fastest region of the disk, e.g.
by creating a small partition there, which just fits the size of the
database. Maybe they did this, but they do not write it. So if they just
installed the 64bit parallel to the 32bit SLES and the system and
database were not located on the same disk space (and therefore disk
performance was not the same) the tests are including more than just the
32/64bit difference.

Ciao
Siegbert