datastore vs. t.lock2

[Matthias Andree]

Shawn Michael <blkmajik at monkeyspunk.net> writes:

> On our database servers we had some bad luck with ext3 in RedHat's
> default install.  The performance just wasn't anything like what we had
> hoped it would be.  The fix for us was to modify one of the mount
> options.
>
> Try the following options to mount:
>
> defaults,noatime,data=writeback
>
> The data=writeback helped tremendously under heavy IO and the noatime
> helps to reduce needless IO.  How many people acutally need atime?

On my machine that runs the k_athlon-2.4.19-167 kernel RPM SuSE ship for
their SuSE Linux 8.1 distribution, ext3 + SCSI is fast, as is reiserfs +
IDE, but ext3 + IDE creeps. I just posted test code to simulate
bogofilter or bogoutil -l creating a 20 MB data base in 500,000 writes
in random order to 5,000 pages -- which is closely resemblant to Greg's
access pattern. Mounting writeback will not scrub away a whole order of
magnitude, but we have almost two orders of magnitude difference between
fast and slow. There must be something wrong with the Linux Kernel Greg
and I are using.

-- 
Matthias Andree