A framework for enhancing dependability in self-x systems by Artificial Immune Systems

Abstract

In self-adapting embedded real-time systems, operating systems and software provide mechanisms to self-adapt to changing requirements. Autonomous adaptation decisions introduce novel risks as they may lead to unforeseen system behavior that could not have been specified within a design-time model. However, as part of its functionality the operating system has to ensure the reliability of the entire self-x system during run-time. In this paper, we present our work in progress for an operating system framework which aims to identify anomalous or malicious system states at run-time without a sophisticated specification-time model. Inspired by the Artificial Immune Systems Danger Theory, we propose an anomaly detection mechanism that operates not only on the local system behavior information of the monitored component. Furthermore, to ensure an efficient behavior evaluation, the anomaly detection mechanism implies system-wide input signals that indicate e.g the existence of a potential danger within the overall system or the occurrence of a system adaption. Due to the ability of this framework to cope with dynamically changing behavior and to identify unintended behavioral deviations, it seems to be a promising approach to enhance the run-time dependability of a self-x system.