Posts Tagged stress testing
For the last couple of months I’m involved in Informix Centaurus EVP, and being a part of it is a very nice thing because (among other benefits) you have the opportunity to test the cool new features of the version before it is generally available. Now, having singed the non-disclosure agreement, I can’t really say much, but there is one thing I believe IBM won’t mind me mentioning.
Some time ago I heard a story how each new version, just out of the box, is faster than the previous one. Using our stress test feature with Apache Jmeter that I previously explained in this post, I produced the same bunch of tests on the two instances – 11.70 and 12.10. Both versions were set up on the same machine, using the exactly same hardware and same configuration. Data chunks (cooked files) were on the same partition, and the databases were, of course, identical for the tests. To sum it up, in the identical environment, our test have shown that 12.10 was up to 3% faster than the 11.70. With no extra configuration or tune-up whatsoever.
Here’s a good way to test performance of your database and the machine it’s running on, especially if there is lot of business logic in the database itself – use Apache JMeter. It’s an open source Java desktop application that allows you to simulate load on server.
What you need to do is prepare a test system, a stack of SQL statements each JMeter thread (session) will execute and a set of variable values to use in those statements, when needed. After that, you should download and install the JMeter, create a new test plan and start testing. There is a good explanation with screen shots on how to do that at this blog post, so I won’t explain this in more detail. We followed the instructions and set up the tests with no problem, just replaced MySQL with Informix.
Remember, there’s no rollback implemented whatsoever, so if your test changes data, you should execute some code manually that’ll take the database in original state, or restore it from previous backup.