"As it does for many people today, the Linux movement
enthralls me. I'm interested not only because I'm more of a
UNIX-based DBA but also because of the amazing speed with which
major vendors, such as HP, Compaq, Dell, IBM and Oracle, have
embraced this open-source operating system. Last year Linux server
sales accounted for approximately 30% of Compaq's, 13.7% of Dell's
and 13.5% of IBM's total server sales, according to eWeek.
Moreover, IBM spent a billion dollars on Linux development in 2001,
after having ported Linux to all their hardware platforms in 2000.
Furthermore, Intel's new 64-bit Itanium CPU lists only four
supported operating systems: Windows, Linux, AIX and HP-UX. And
let's not forget that Oracle released 9i on Linux months ahead of
the Windows port. Then again, maybe I just like the underdog--I
mean I'm writing this article on my AMD Athlon-based PC.
But no matter how fashionable Linux may be, that popularity does
not automatically translate into nor does it guarantee performance.
Even though Linux runs on everything from IBM 3/90s to Sun
SPARC-based boxes, most people at this point are still probably
running Linux on Intel-based server platforms. Now without sounding
condescending, let me state that the PC architecture was never
really intended to scale to the heights Linux makes possible. Thus
we need to make sure that we squeeze every last drop of blood out
of the turnip when we deploy an Intel based Linux
server--especially for enterprise databases like DB2 and Oracle.
Believe it or not, it's quite easy to get upwards of 1000% database
improvement through proper Linux tuning and database configuration
As with any scientific endeavor, in this article we will attempt
to evaluate different tuning techniques by establishing a
controlled environment where we can ascertain a baseline, identify
all the changeable relevant variables, modify one variable at a
time and obtain a reliable measurement of the effects for that one
change. Wow, I haven't written techno-babble like this since I was
a Physics student at Ohio State. In plain English, we must test one
tuning concept at a time in order to accurately measure the
observable effects of only that change."