For comparison, my test slave (dual core), with MySQL configured to use ample memory and the MyIsam folder on a tmpfs reliably runs tests in 5 minutes 30 seconds.

Damien Tournoud

On Sun, Dec 14, 2008 at 2:09 PM, Bart Jansens <bart@motd.be> wrote:
Op zaterdag 13-12-2008 om 17:09 uur [tijdzone -0600], schreef David
Timothy Strauss:
> We have one of the MyISAM databases symlinked to tmpfs in RAM. It's
> fast and somewhat unreliable, but I have no reason to believe that if
> it provides test results that the results are compromised in accuracy.

I imagine it could make a lot more of a difference for SQLite,
especially compared to the performance of an encrypted harddisk.


FYI, A while back I ran some tests to check if I could improve the
performance on my system. These were the results:

- tweaking the Mysql config a little, improved performance by about 8%.
And I'm not a mysql expert - others might be able to improve it even
more.

- placing the mysql db on tmpfs did improve performance slightly (1 or 2
percent). But when running multiple tests in parallel, it actually
degraded performance a little. I'm not quite sure why, but it was
reproducible.

- because modern CPUs have multiple cores, the highest performance boost
was running the tests with higher concurrency. On my quad core, it would
only take about 2 minutes to complete the tests. However, this proved to
be somewhat unreliable, i kept getting database exceptions.



Of course it all depends on the hardware, if you have a system with
slower disk, mysql in tmpfs will be significantly faster. Check what the
bottleneck is and start optimizing there.