An optimal, distributed deadlock detection and resolution algorithm for generalized model in distributed systems

Abstract

We propose a new distributed algorithm for detecting generalized deadlocks in distributed systems. The algorithm records the consistent distributed snapshot of global wait-for-graph (WFG) through propagating the probe messages in the forward phase and reducing the WFG to determine the entire set of deadlocked processes in the backward phase. The reducibility of each process is decided based on the information in replies that have received from its immediate wait-for processes during the reduction. We also formally prove the correctness of our algorithm. It has a worst case time complexity of 2d time units and the message complexity of 2e, where d is the diameter and e is the number of edges of the distributed spanning tree induced by the algorithm. The significant improvement of proposed algorithm over earlier algorithms is that it achieves optimum results using fixed sized messages and minimizes the messages to resolve deadlocks. © 2009 Springer Berlin Heidelberg.