2.5.3 Test environment setup
When performing any performance tests, make sure these criteria are applied throughout all your tests:
- Perform your tests on machines that mirror the production server state.
Be as realistic as possible with the test environment. Try to get the closest hardware configuration possible to the production environment. Make sure you will be able to extrapolate the test results to the production environment.
- Make sure you are using the same data set for all your tests.
This allows for better comparability of your test results. At least one of your test environments should also include production data volumes to ensure that all data access paths are realistically tested.
- Make sure that nobody is using the test machines and that no background processes are running that consume more resources than what you find in production.
As an example, if the intent is to test performance during the database backup, then make sure the backup is running.
This means that running your monitoring software in the background, which also runs in production, like Tivoli or BMC is all right, but having a database backup running may not be valid for your desired test.
- Check CPU, memory and disk utilization before and after each test run to see if there are any unusual patterns.
If the target environment will be using shared infrastructure (messaging servers, authentication providers, for example), try to make sure the shared component is performing under a projected shared load.
- Isolate network traffic as much as possible.
Using switches, there is rarely a circumstance where traffic from one server will overrun the port of another. It is possible, however, to flood ports used for routing off the network to other networks, or even the switch backbone for very heavy traffic. Make sure that your network is designed in such a manner that it isolates the testing environment as much as possible prior to starting, since performance degradation of the network can create unexpected results.