Analysis of Software Aging Impacts on Plant Anomaly Detection with Edge Computing

Abstract

Edge computing allows data-intensive analytic processes running on a cloud server to migrate to a local computing environment near the data sources, and as a result the architecture can benefit from a significant bandwidth reduction as well as improved security. However, due to the scarcity of resources on an edge device, a problem of software aging is likely to occur after longtime execution. The consequence of software aging should be significant, whereas such a risk has not been studied in the previous literature. In this paper, we propose a deterministic and stochastic Petri Net (DSPN) to quantitatively analyze the impacts of software aging phenomenon on a cyber-physical system using edge computing. As a data analytic application using real-world sensor data, we consider anomaly detection for a water treatment plant. We introduce two application-oriented performance measures, namely detection loss probability and the expected number of false alarms in terms of plant anomaly detection. Our numerical and simulation experiments on the proposed DSPN show that software aging clearly has negative impacts on the defined performance measures that reveal the necessity of greater attention to software aging issues in edge computing systems.