Rapid Bottleneck Identification
Submitted by Alexander Podelko on Sat, 30/07/2005 - 20:28.
exploratory testing | performance testing
A very good white paper Rapid Bottleneck Identification. A Better Way to Load Test from Empirix. Nothing is really Empirix-specific.
It states very important points that almost always are missed in load testing (especially by beginners):
“The first is, in our experience, every application has at least one bottleneck and secondly, no application we’ve ever seen has ever scaled to meet its requirements, at least initially. These broad statements come from years of experience for hundreds of clients through thousands of performance tests.”
I am ready to sign under that. If you admit that, the whole approach to load testing should be different – something like specified in the paper. Starting from simple tests to find system’s bottlenecks, some kind of exploratory testing.
It is really not a completely new word in load testing and I am not sure that it should be named and trademarked (although why not, we are all in business) – more or less I have been using the same approach for a long time as well as, I guess, many other people. Still here it is very well stated. Another good point in the paper – about throughput and concurrency.
Unfortunately is can be quite difficult to implement the approach in practice from the political point of view – quite many people don’t like these assumptions. Quite often you need to have an “official” plan while you are doing something quite different. The wider these assumptions would be accepted, the simple would be work of load testers.
It states very important points that almost always are missed in load testing (especially by beginners):
“The first is, in our experience, every application has at least one bottleneck and secondly, no application we’ve ever seen has ever scaled to meet its requirements, at least initially. These broad statements come from years of experience for hundreds of clients through thousands of performance tests.”
I am ready to sign under that. If you admit that, the whole approach to load testing should be different – something like specified in the paper. Starting from simple tests to find system’s bottlenecks, some kind of exploratory testing.
It is really not a completely new word in load testing and I am not sure that it should be named and trademarked (although why not, we are all in business) – more or less I have been using the same approach for a long time as well as, I guess, many other people. Still here it is very well stated. Another good point in the paper – about throughput and concurrency.
Unfortunately is can be quite difficult to implement the approach in practice from the political point of view – quite many people don’t like these assumptions. Quite often you need to have an “official” plan while you are doing something quite different. The wider these assumptions would be accepted, the simple would be work of load testers.