Event-driven dynamic workload scaling for uniprocessor real-time embedded systems

Abstract

Many embedded systems are designed to take timely reactions to the occurrences of particular scenarios. Such systems could sometimes experience transient overloads because of workload bursts or hardware malfunctions. Thus a mechanism to focus limited resources on the processing of urgent events is a key to retain system validity under stressing workloads. In this paper, we propose a new approach for workload Scaling in uniprocessor real-time embedded systems. The idea is to view the system as a black box, and workload scaling for overload management can be done via very intuitive primitives, i.e., how hardware events are selectively fed into the system. Such a new approach removes the need for the adjustments of task periods and task phasing, which is important for many workload-scaling techniques. The proposed approach is implemented in a real-time surveillance system. Experimental results show that the system still delivers good accuracy and high responsiveness for visual-object tracking under the presence of overloads.