A safe distributed deadlock resolution algorithm for the OR request model

Abstract

This paper presents an algorithm for resolving OR deadlocks in distributed systems. The algorithm works in two concurrent phases. Firstly, it collects the paths of the WFG (at each initiator process). A termination detection mechanism is used to know the ending of this phase. Collected paths are then analyzed to discover whether the process belongs to a deadlock. The proposed algorithm has two primary advantages. First, it assures that only true deadlocks are detected. Second, since only one process detects each deadlock, it simplifies the task of deadlock resolution. The algorithm resolves all deadlocks with a communication cost of O(2e) messages in the worst case (being e the number of wait-for relations between nodes of the knot), so their complexity is also better than or equal to the existing algorithms.