In distributed systems, certain control processes run concurrently with application processes. Examples include deadlock and termination detectors, garbage collectors, and consistent global state coordinators. These types of processes sometimes inhibit actions of the application process; in other words, they disable a subset of actions that the application would otherwise perform. For example, consistent global state coordinators sometimes prevent the sending of application messages temporarily. Such inhibition is often undesirable for a computation, particularly if performance is a primary issue. It has been formally demonstrated that inhibition can be necessary in order to develop certain control algorithms, and that there are sometimes tradeoffs between inhibition and increased message traffic. In particular, this is the case for consistent global state coordinators. This project investigates the role of inhibition in a broader range of distributed control algorithms. Potential targets include termination and deadlock detection, garbage collection, and multicasting. The goal is to understand precisely when inhibition is necessary, how to minimize inhibition, and how it affects other factors such as execution time, message complexity and fault tolerance.
