An Automatic Framework for Efficient Software Performance Evaluation and Optimization

Performance evaluation has been an important part of software development. Fast and accurate software performance evaluation can help not only in understanding the behavior of software during development, but also in capacity planning during deployment. In this paper, we propose an automatic framework that can be used to optimize software performance indicators. For optimization, we use response surface methodology (RSM) for its simplicity and its ability to describe the behavior of a system in a whole neighborhood. To efficiently obtain the response data for different parameter settings, a reusable importance sampling (IS) simulation and testing method can be applied. We describe here this novel method based on a combined response surface and importance sampling (RS-IS) framework and we illustrate its usefulness via simulated examples that minimize total cost in a capacity planning problem