Skip site navigation (1)Skip section navigation (2)
Date:      Wed, 9 Feb 2005 23:20:42 -0800
From:      "Ted Mittelstaedt" <tedm@toybox.placo.com>
To:        "Joe Kelsey" <joe@zircon.seattle.wa.us>, "Matt Olander" <matt@offmyserver.com>
Cc:        freebsd-questions@freebsd.org
Subject:   RE: FreeBSD 5.3 MySQL Performance
Message-ID:  <LOBBIFDAGNMAMLGJJCKNGEFEFAAA.tedm@toybox.placo.com>
In-Reply-To: <1108005975.683.47.camel@zircon.zircon.seattle.wa.us>

next in thread | previous in thread | raw e-mail | index | archive | help
owner-freebsd-questions@freebsd.org wrote:
> On Wed, 2005-02-09 at 16:44 -0800, Matt Olander wrote:
>> On Wed, Feb 09, 2005 at 07:58:56PM -0500, Alec Berryman wrote:
>>>> Also, does anybody have any FreeBSD 5.3/MySQL benchmarks? I
>>>> searched the mailing lists but didn't turn up anything.
>>>
>>> There was an article posted to Newsforge today about benchmarking
>>> MySQL on different operating systems.
>>
>> oh! thanks...but according to this article, Linux outperformed
>> FreeBSD in every metric shown :-(
>>

No, not true, reread the article.  Performance was on par for
uniprocessor
versions of Linux and FreeBSD for most tests.  Also, the author said the
following
about the testing:

"highest performer in one category for a limited set of tests does not a
"best" operating system make."

Please keep in mind that editors of publications love benchmarking
articles, because they always get someone's tit in a wringer, and
attract a lot of attention.  And attention sells newspapers.

But you shouldn't take these things too seriously.  The article
points out a few things and the accompanying reader responses point
out a few more things that are educational if you are choosing
to run a database on a UNIX system, but by no means should
the article be used as the sole basis for choosing one OS over another.

People that do benchmarking and publish the results are generally
hoping to help point out problems.  Sometimes this is because they
have an axe to grind and want to see their favorite OS or program
or whatever get some attention, sometimes just because it's nice to
see some of your work in print.  But regardless of why they do it,
the results are valuable, because if problems that benchmarking
reveals wern't pointed out, they wouldn't ever get fixed.

My take on the article is the most surprising thing in it was that
Sun's own support staff couldn't answer the authors query about why
Solaris was so slow, and the author finally figured it out by himself
(The filesystem wasn't mounted with the forcedirect option) and
set the needed option, whereupon performance dramatically improved.
We always hear from commercial OS vendors how their products are
so much better because they are supported - well it seems to me that
if Sun's support was this bad for their own OS, well that throws
the entire argument out the window, don't it?

>> is that accurate?
>
> LinuxThreads is the WORST implementation of threading that anyone can
> imagine.  Do not ever use Linux or the horrid LinuxThreads for
> anything that you want to save.
>
> Any so-called "benchmark" comparing Linux to anything else (especially
> windoze) has been polluted by the tradition in the linux/windoze world
> of running their disks in the completely unsafe "asynchronous" mode so
> popular with the ATA disk drive manufacturers.

The author of the article avoided this by using a test method that in
his words:

"I performed one test run to prime the system, almost all of the data was
cached by MySQL, so there was little or no disk access."

>
> Many companies have used FreeBSD and MySQL for years and years.  There
> is no reason to not jump to FreeBSD and start using MySQL.

Exactly, we use MySQL and FreeBSD quite a lot and have no problem with
it.

> At my last
> job, we ran very large MySQL databases on FreeBSD.  For speed we used
> 15,000 RPM SCSI-3 disk drives.  This gives you all the speed you need
> with the guaranteed safety of FreeBSD.  Of course, SCSI-3 15,000 RPM
> drives are more expensive than those wimpy ATA drives.
>

You see, this here points out the crux of the problem.

Boiled down the article essentially said that Linux performed better
because it's SMP implementation allowed mysql to take advantage of
both CPU's while FreeBSD's SMP implementation didn't.

But you see the problem with this is that in a real life situation,
it is not often that you have such a small database and such a large
amount
of system memory that the OS can load the entire database into a
disk cache in ram.  As you can no doubt understand, if the database
is on disk all the additional CPU's in the world won't make the
database run any faster once the disk channel gets saturated, which
is easy to do.

And even if you can load the entire database in ram, if you make a
lot of writes to it, the system has to push these to the disk channel
eventually, unless of course you like for your entire database to
vanish if there's a power interruption or system crash of some
kind.  So for a situation of a steady stream of writes, you
end up I/O bound again.  SMP on a database is no help if the system
is I/O bound.

And if your database is going to I/O bind, because of how it's used
and setup and how big it is, then this benchmark article is completely
useless to you.

And even if your database isn't going to I/O bind then read the
following comment one of the readers posted regarding OpenBSD and
FreeBSD:

"They both use userland only threading, and therefore mysql is only
ever running on a single CPU, no matter how many are in the system
...Different processes will run on different CPUs just fine,
postgresql for instance would see an increase in perfomance on these
platforms"

One last thing about SMP - rarely is mysql used in a vacuum, many
times it's paired up with apache.  In those cases if there's multiple
accesses to the server at the same time then even though mysql only
runs on 1 cpu, an instance of apache may be on the other cpu at the
same time, answering some other query.  So when you take in all the
other stuff on a typical production mysql server, the multithreading
in linux isn't going to make any difference for just 2 CPU's.

Ted



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?LOBBIFDAGNMAMLGJJCKNGEFEFAAA.tedm>