Re: Impact of Changed O.S. on Performance
- From: "Dennis Williams" <oracledba.williams@xxxxxxxxx>
- To: VIVEK_SHARMA <VIVEK_SHARMA@xxxxxxxxxxx>
- Date: Fri, 18 Aug 2006 13:39:43 -0500
Vivek,
From your question, it sounds like you want to benchmark two configurations.
Obviously you want both configurations as identical as possible except for
the variable you are testing.
I'm just saying that in the greater scheme of your benchmark, the vendor
brand of the database server isn't the biggest issue. The unfortunate fact
of real life is that you rarely get an opportunity to control all the
possible variables. Obviously you want to control as many variables as
possible, but you also want to clearly note any variables you were not able
to control and try to estimate the affect of this on your results. For
example, if the database server didn't have the desired configuration, that
should be noted along with an estimate of the effect on the benchmark
results.
You can also do a database-only benchmark of the database server over
the network to get a rough idea of the database server performance.
You should also take a look at the pattern of database access that the
application produces. For example, does it submit just a few large queries,
or does it submit a lot of small queries?
Take a look at the great book "Optimizing Oracle Performance" by Cary
Millsap and Jeff Holt. One idea for you would be to run an Oracle 10046
trace during the benchmark. This would allow you to clearly separate the
application and the database components of the response time. This would
allow you to see how the change in application affects the performance. It
would also allow you to understand how the database server would affect the
application response time.
Dennis Williams
Other related posts: