Dynamic reconfiguration of software objects using Petri nets and network unfolding

This paper demonstrates the use of supervisory control theory in synthesizing plug-ins for distributed software. The plug-ins are software objects that supervise an existing distributed system so that certain properties such as fairness and deadlock freedom are guaranteed. The distributed application is modeled as a bounded ordinary Petri net and system analysis is accomplished through a partial order method known as unfolding. The unfolding constructs an event structure that provides a natural encapsulation of concurrent threads of execution whose selective disablement by the supervisory plug-in assures the desired application property. The synthesis of the plug-in is based on results from supervisory control theory and the synthesized plug-ins are optimal in that they are maximally permissive. We demonstrate our approach on a distributed cache system