Subject: Re: sysbench and ubench benchmarks
To: Jeremy C. Reed <>
From: Chris Wareham <>
List: current-users
Date: 05/02/2007 19:16:28
Jeremy C. Reed said on 1/5/07 00:30:
> Some comparisons of a few operating systems include NetBSD 3.1 and 4.99.17 
> is at
>   Jeremy C. Reed

That's one seriously screwed up benchmark test. Firstly, the author
claims that only companies like Yahoo! bother to tune the OS or database
engine - sorry, but every company I've worked at employs sysadmins and
DBA's to do just that. Then he uses a laptop for the OpenBSD and Nexenta
test runs rather than the desktop machine used for the other systems.
Not exactly the hardware I'd run a production database server on. Then
there's the claim that "disk performance depends on how close the
partition is to the start sector of the disk" - I'm no hardware guru,
but I was under the impression that modern hard drives had multiple
platters for starters. Judging by the comment about wanting to test
NetBSD 4.0, he doesn't appear to understand the versioning either - it
sounds like he thinks 4.99.x is the lead up to version 4.0. And quite
frankly, someone who can't install OpenBSD has got a lot more to learn
than what performance their OS is going to give them...

Anyway, at the end he wishes someone would "do some real-world
performance benchmarks involving production data and applications using
large databases". Well for what it's worth my last company did just
that, comparing Solaris 9 and RedHat EL 4 on dual CPU Sun Opterons and
Dell Xeon servers. The clear winner was Solaris on the 1.8Ghz Opteron
machines (despite the 3.2Ghz Xeons in the Dell machines). Solaris scaled
to more connections, served more rapidly and with more consistent
performance than Linux. The dataset was 20Gb in a MySQL 4.1 database,
and the tests were run from another of the Dells with a test suite
constructed in Java using a rather neat package called JMeter.