Performance tuning of Java EE application servers with multi-objective differential evolution

摘要

This paper presents an empirical approach for the performance tuning of Java EE application servers (ASs) using a multi-objective differential evolution algorithm. It features multi-objective black-box optimization of selected AS's configuration parameters. The proposed approach is used for performance tuning of the AS GlassFish and Java EE test application DayTrader. The obtained results improve the objectives' values of referenced configuration (AS's default settings) individually, as well as a whole. We also find a Pareto front approximation which entirely dominates the referenced configuration objectives' values. Comparison with existing approaches is made based on previously found relations between certain configuration parameters' values. Results from multi-objective performance tuning offer a choice of alternative configurations and thus enable the attainment of business goals even within an environment where the objectives' priorities may change.