Pattern-driven performance optimization at runtime: Experiment on JEE systems

Poor-performance solutions need to be refactored for optimizing the performance in distributed computing systems. This paper presents a case study of pattern-driven online performance optimization by automatically detecting poor-performance solutions and refactoring them into great-performance ones. Poor-performance solutions are abstracted as negative patterns and the corresponding great-performance ones are abstracted as positive patterns. Both negative and positive patterns are manually defined by experts using a meta-model in order to make the patterns understandable for middleware and reusable. For a running system, poor-performance solutions are detected automatically by discovering the negative patterns from the runtime context. For a detected poor-performance solution, the refactoring operations which change the running system from the poor-performance solution to the great-performance one are automatically generated by comparing the negative and positive pattern. The refactoring operations will be mapped to the actual management operations provided by the middleware which then executes the refactoring without stopping the running system. ? 2010 ACM.EI