Towards Balancing Determinism, Memory Consumption and Throughput for RTSJ-Based Real-Time Applications

Determinism, memory consumption and throughput are three important performance indicators for RTSJ-based real-time applications, but they often interact and conflict with each other. Manually balancing these performance indicators is often time-consuming. Therefore, there is the need to clarify the relationship among the performance metrics and make trade-off automatically. In this paper, we abstract real-time thread properties relating to these performance indicators and analyze their relationship. Based on the analysis, an automatic configuration framework to balance the performance metrics is proposed. In this framework, a stochastic process is developed to represent real-time threads' determinism. And real-time application's throughput and memory consumption is quantified with these thread properties as parameters. Experiment results based on Sweet Factory application show that our approach could optimize memory consumption and throughput while guaranteeing determinism effectively.