An incremental approach for detecting distributed deadlocks in the generalized model

Abstract

A new algorithm has been presented to detect and resolve generalized deadlock in distributed systems. In this algorithm, the initiator incrementally constructs the Local Wait-For Graph (LWFG) by propagating the probes along the edges of the Wait-For Graph and collects the replies that carry the wait-for relation between the processes directly. It updates the LWFG based on the information possessed to determine a deadlock. Proofs have been presented to authenticate its correctness. The proposed algorithm has the worst case time complexity of d + 2 time units and the message complexity of less than 2e messages, where d is the diameter and e is number of edges in the WFG. This centralized algorithm uses messages of fixed size as compared to the existing centralized algorithms. Simulation results reveal that the performance of the proposed algorithm is better or equal to the earlier generalized deadlock detection algorithms in terms of deadlock duration, message traffic and message length. Furthermore, it resolves deadlock without using additional round of messages and handles simultaneous execution of the algorithm effectively.