Towards Parameter-aware Benchmarking of Bytecode and API for Predicting Component Performance

Abstract. Performance prediction of component-based software systems is nee-ded for systematic evaluation of design decisions, but also when an application’s execution system is changed. Often, the entire application cannot be benchmarked in advance on its new execution system due to high costs or because some re-quired services cannot be provided there. In this case, performance of bytecode instructions or other atomic building blocks of components can be used for per-formance prediction. However, the performance of bytecode instructions depends not only on the execution system they use, but also on their parameters, which are not considered by most existing research. In this paper, we demonstrate that pa-rameters cannot be ignored when considering Java bytecode. Consequently, we outline a suitable benchmarking approach and the accompanying challenges.