Thanks for the question, John.
Asked: April 28, 2021 - 11:23 pm UTC
Last updated: May 06, 2021 - 4:55 am UTC
Viewed 100+ times
A question from the client that comes up every few years is to predict if the Oracle database server will be able to handle a new application's peak load. Instead of trying to think of all that needs to be considered on the fly, it would be great if there was an Oracle document that had a checklist with all of the questions that we must answer so that we can give the client a definite answer of yes we can predict if x,y and z or performed. I know that in most cases, this will be nearly impossible to answer as it will take too much time to answer and we can't control the variables for other apps that share the same resources like database, network, SAN, etc. For instance, usually the network and SAN are shared with the database server so we will need to get peak loads of all the other applications plus the expected max throughput for the network and SAN.
Thanks for your help,
and we said...
My experience has been the opposite problem, namely, no-one can tell you *what* the new application will do.
People will say: "It will do 500 transactions per second" to which you ask "Is a transaction 1 query plus 1 insert, is it 10 queries? is to 20 inserts?" and you get blank looks back :-)
If the application *exists*, then that app can be measured with (say) AWR to give a total IO bandwidth and average active sessions (a key CPU metric). You can then compare that to the what headroom you have on your existing database to see if it will fit.
If the app does *not* exist (which is the norm), you could look at using something like Swingbench ( http://www.dominicgiles.com/downloads.html
) to perform typical workloads and see how an existing machine copes with extra load at the anticipated level.