Software instrumentation is central to many forms of program analyses. Runtime verification is one such prominent analysis example that imposes stringent constraints on the software behaviour reported by an instrumentation. Reactive components, ubiquitously found in many software systems, dictate further requirements on the instrumentation that is applied to them, such as the need for responsiveness and elasticity. In addition, the heterogeneity of their constituent components often relegates the runtime analysis to outline monitors, which execute separately from system components.

This talk discusses the challenges faced when reactive component systems are instrumented using decentralised outline monitors. We present a novel decentralised instrumentation algorithm based on next-hop routing for this setting, arguing that it satisfies the requirements for monitoring reactive component systems. We also discuss its performance based on an empirical evaluation and show that the runtime behaviour of our algorithm makes it feasible for soft real-time reactive applications. Compared to inline and centralised instrumentation, we conclude that our decentralised approach induces comparable latency to inline monitoring for software that performs long-running, computationally-intensive tasks, such as in Big Data stream processing.

This is joint work with Adrian Francalanza, Luca Aceto and Anna Ingolfsdottir from the University of Malta, Gran Sasso Science Institute and Reykjavik University, respectively.