| || |
Scaling down production servers
One of my colleague has the following query.
We are working on load testing the Oracle Applications 11i. The test hardware is not of the same scale as the production servers (the machine is of the same type except that the number of CPU's is not the same) and is not able to handle the concurrency levels of production.
We wanted to know from you, how is the user counts to be used on the scaled down hardware calculated? Is there any standard mechanism followed to scale down the user counts to suit the test hardware. Could you please let me know.
Re: Scaling down production servers
Sadly there is no standard calculation that you can let loose on your test setup.
What I have done in the past is run tests on both the test environment and production environment (this was before the production environment was actually brought into production). I put increasing (identical) loads on each of the env. and had full system monitoring switched on. The results of the tests showed me the capacity/capability difference between the 2 environments.
However we also found that certain problems in the test env. simply did not happen in the prod. env. indicating to us that our future conclusions about the the AUT tested in the test env. could potentially be wrong and we could be tempted to dismiss any negative observation in test as something that would not happen in prod.
Anyway, it is a classic problem for performance testers.