A Toolset for Predicting Performance of Legacy Real-Time Software Based on the RAST Approach
该文介绍了一个基于RAST方法的工具集，用于预测在资源共享环境中运行的遗留实时分布式软件系统的性能，利用系统请求日志替代APM数据，并通过TeaStore基准系统验证了其仿真准确性。
Simulating and predicting the performance of a distributed software system that works under stringent real-time constraints poses significant challenges, particularly when dealing with legacy systems being in production use, where any disruption is intolerable. This challenge is exacerbated in the context of a System Under Evaluation (SUE) that operates within a resource-sharing environment, running concurrently with numerous other software components. In this article, we introduce an innovative toolset designed for predicting the performance of such complex and time-critical software systems. Our toolset builds upon the RAST ( R egression A nalysis, S imulation, and load T esting) approach, significantly enhanced in this article compared with its initial version. While current state-of-the-art methods for performance prediction often rely on data collected by Application Performance Monitoring (APM), the unavailability of APM tools for existing systems and the complexities associated with integrating them into legacy software necessitate alternative approaches. Our toolset, therefore, utilizes readily accessible system request logs as a substitute for APM data. We describe the enhancements made to the original RAST approach, we outline the design and implementation of our RAST-based toolset, and we showcase its simulation accuracy and effectiveness using the publicly available TeaStore benchmarking system. To ensure the reproducibility of our experiments, we provide open access to our toolset’s implementation and the utilized TeaStore model.