Security-aware Scheduling of Mixed-Trust Tasks in Multiprocessor Real-Time Systems

Abstract

Real-time systems are a fundamental component of safety-critical cyber-physical systems, such as autonomous vehicles, robotics, and avionics, where missed deadlines can have severe consequences. As these systems become increasingly interconnected, security has emerged as a critical concern, yet enhancing security typically increases computational demands, risking deadline misses. While security mechanisms for uniprocessor real-time systems are well researched, globally scheduled fixed-priority multiprocessor systems have received comparatively little attention. This thesis addresses that gap by introducing the Mixed-Trust task model for globally scheduled fixed-priority multiprocessor systems to enhance security, along with a corresponding schedulability analysis. The proposed model distinguishes between trusted and untrusted tasks, and upon detection of a security threat, transitions the system to a higher security level by introducing new security enhancing tasks while suspending untrusted ones. Inspired by the work on mixed-criticality scheduling the model additionally supports changing periods for tasks across security levels with the objective to increase the surveillance of the system. Finally to mitigate timing-based attacks, a randomization algorithm extending a uniprocessor algorithm, known as TaskShuffler, to globally scheduled fixed-priority multiprocessor systems is presented, diversifying task execution order to increase unpredictability for a potential attacker. The proposed mechanisms are evaluated empirically: the schedulability analysis is assessed using synthetically generated task sets, revealing how the task model and the number of available processors interact in ways that may help the system designer in selecting the hardware platform. The randomization algorithm is evaluated through simulation, demonstrating a meaningful increase in execution order unpredictability.