Inferring Performance from Code: A Review

Performance is an important non-functional property of software that has a direct impact on the end-user’s perception of quality of service since it is related to metrics such as response time, throughput, and utilization. Performance-by-construction can be defined as a development paradigm where executable code carries some kind of guarantee on its performance, as opposed to the current practice in software engineering where performance concerns are left to the later stages of the development process by means of profiling or testing. In this paper we argue that performance-by-construction techniques need to be probabilistic in nature, leveraging accurate models for the analysis. In support of this idea, here we carry out a literature review on methods that can be used as the basis of performance-by-construction development approaches. There has been significant research—reviewed elsewhere—on performance models derived from high-level software specifications such as UML diagrams or other domain-specific languages. This review, instead, focuses on methods where performance information is extracted directly from the code, a line of research that has arguably been less explored by the software performance engineering community.