Feedback Admission Control for Workflow Management Systems

We propose a novel feedback admission control (FAC) algorithm based on control theory as a unified framework to improve the real-time scheduling (RTS) performance in industrial workflow management systems (WMSs). Our FAC algorithm is based on four main principles. First, it does not require the knowledge of RTS parameters of jobs prior to their arrival to the system for scheduling and processing. Second, it does not require a change of the scheduling architecture/policy in the industrial WMS which is a requirement in some industries including the one under consideration in this thesis. Third, we derive dynamic models for computing systems for the purpose of performance control. Finally, we apply established control laws to manage the trade-offs in meeting deadlines and increasing platform utilisation (classical RTS objectives).

The generality and efficiency of our proposed FAC algorithm are demonstrated by its application in three typical scheduling scenarios in industry. First, we tested our algorithm with simple tasks that are periodic and independent. For this application, we developed two FAC versions based on basic and advanced control laws to compare their performance with respect to the RTS objectives. Second, we added task dependencies as a scheduling constraint because they are witnessed in some industrial workloads. We evaluated our FAC algorithm against other baseline algorithms like the completion-ratio admission controller with respect to the RTS objectives. Third, we extended our FAC algorithm to support enterprise resource planning decisions in acquiring additional computing processors in real-time to further achieve the RTS objectives while constrained by industrial projects? financial budgets.