Using Markov chains for HTTP benchmarking

blog.andrewvc.com

55 pointsandrewvc4 comments

"What Markov chain generation does is give a nice mix between two different ways of generating load, combining the fuzzing of random request generation with the feel of replaying logs. I think of it as a happy medium."

Why do we need a medium? Replaying access logs at a high-velocity gets the job done.

Well, because there's no one right way to test is why.

Strictly repeating logs doesn't tell you what happens when your access patterns change. Markov chains give you an idea of how your site acts under variations on your current traffic.

It looks interesting, but does the added complexity have any tangible benefits over a benchmark app using a url-frequency table (a 0th-order markov chain)?

Great question. Maybe, maybe not. As I said, it just depends. Every test answers its own question.

Here's the thing about orders, they're like a sliding scale of similarity between source and generated text. Whether that scale is worth toggling for your data is up to you.

Reply on news.ycombinator.com