Fault-tolerant distributed deadlock detection/resolution

Abstract

In an unreliable distributed system, faulty processors may prevent a deadlock detection algorithm from properly detecting deadlocks. However, few of the algorithms proposed in the literature address the issue of handling process failures in a distributed system. This paper proposes a fault-tolerant distributed deadlock detection algorithm which integrates a priority-based probe algorithm with a PMC-based diagnosis model. This algorithm detects deadlock cycles as well as identifies process failures under a bounded number of failures in a deadlock cycle by using extended probe messages that contain additional information about faulty processors.<>