Feedback control scheduling in distributed real-time systems

Distributed soft real-time systems are becoming increasingly unpredictable due to several important factors such as the increasing use of commercial-off-the-shelf components, the trend towards open systems, and the proliferation of data-driven applications whose execution parameters vary significantly with input data. Such systems are less amenable to traditional worst-case real-time analysis. Instead, system-wide feedback control is needed to meet performance requirements. In this paper, we extend our previous work on developing software control algorithms based on a theory of feedback control to distributed systems. Our approach makes three important contributions. First, it allows the designer for a distributed real-time application to specify the desired temporal behavior of system adaptation, such as the speed of convergence to desired performance upon load or resource changes. This is in contrast to specifying only steady-state metrics, e.g., deadline miss ratio. Second, unlike QoS optimization approaches, our solution meets performance guarantees without accurate knowledge of task execution parameters - a key advantage in an unpredictable environment. Third, in contrast to ad hoc algorithms based on intuition and testing, our solution has a basis in the theory and practice of feedback control scheduling. Performance evaluation reveals that the solution not only has excellent steady state behavior, but also meets stability, overshoot, and settling time requirements. We also show that the solution outperforms several other algorithms available in the literature.