(mySQL) benchmarks strike back

[Ean Kingston]

> Hi there
>
> I just read
> http://software.newsforge.com/software/04/12/27/1243207.shtml?tid=72&tid=29%20result
> and as in any onther benchmark there is a lot stuff
> that
> can be arguable. I would like to know why is that
> happening?
> the problem is that "we" are pushing
> FreeBSD/postgreSQL as a database solution, and I am
> the guy to blame to, because I  was the one who did
> advocacy for FreeBSD, so I'm sure my boss is going to
> ask me.  And you told us to use FreeBSD instead of
> Linux?
> and I do not want to answer him "beastie is way more
> cool"
>
> I'm doing my own research but some help from here
> would be nice!!!

Others will provide plenty of reasoning for you. Here are some points for
consideration.

That particulary example used a very small database that was cached
entirely in memory. It has no resemblance to a production environment
where databases are always going to be disk-io bound, not cpu-bound (as
was the benchmark).

Linux mounts its filesystems in async mode by default. This is extremely
unsafe and will almost guarantee data loss on a buisy filesystem if the
computer loses power unexpectedly. FreeBSD with softupdate will give much
better performance than Linux with the filsystems mounted syncronously and
you don't risk the data loss if the power fails.

That was a performance test for MySQL. You are using Postgres. These are
different databases. Postgres is a much better choice. It is a much more
complete database server. For example, last time I checked, there was
absolutely no integrity checking in MySQL whereas Postgres does have very
good integrity checking.

The performance metrics from that article were more a test of the current
state of threading than any realistic database system. At the moment, the
newest Linux kernels have much faster threading than FreeBSD. This will
probably change (many times) in the future. The BSDs and Linux go through
cycles of beating eachother in different areas depending on where the
projects are focused at the time.

I, and many others, have found that Linux tends to focus on getting the
next new feature into their systems as quickly as possible. This, I find,
makes for a much less stable system. I have found that the BSDs are much
better at controlling this featurism. The project teams consider the value
of something before they implement it in the production release. The
result is, to me, that you get a much more reliable OS out of FreeBSD than
you do out of Linux.

Your choices of Postgres and FreeBSD indicate that you are looking for
reliability in your database system. For that goal, I think you made the
right choice.

> Jorge Mario Mazo

-- 
Ean Kingston

E-Mail: ean_AT_hedron_DOT_org
URL: http://www.hedron.org/