Distributed deadlock detection algorithms using distributed graph reconstruction

Abstract

In a distributed system where processes communicate directly with each other, a deadlock is a situation in which a set of processes in the system are blocked forever because of requirements that can be never satisfied. For a process which suspects that it may be deadlocked, to determine whether it is indeed deadlocked, it is necessary for it to query other processes. Numerous algorithms on distributed deadlock detection in distributed database system have been proposed for the various deadlocked model. The deadlock model is divided by the form of the resource request. for the OR model, socalled communication model, Chandy et al. have proposed the "search" type algorithm which an initiator gets back replies from all of its dependent set and decides whether it is deadlocked or not. Cidon et al. have proposed the algorithm using cycle detection and clustering techniques. For the AND/OR model, Herman et al. suggested the algorithm using the tree computation. In this thesis, a new algorithm, so-called Extrema Finding (EF) algorithm, is proposed for finding the smallest (or largest) of a set of n uniquely numbered processes in a logically circular arrangement. Each process gets the dependency relation of the neighbored process and rearranges the outgoing edge according to that dependency. by the process which cannot be a local minimum does not send the message, the number of messages required can be reduced. EF algorithm requires O(n) message passes, where n is the number of processes involved in the logical circle. EF algorithm can be applied to the problem of the detection of a cycle for the deadlock detection. Based on EF algorithm, new algorithms, so-called Distributed Graph Reconstruction (DGR) algorithm and AND-OR algorithm for the OR model and AND-OR model, the existence of a cycle in WFG is necessary condition of a deadlocked state. For the OR model, DGR algorithm is based on the characteristics that processes in knot are strongly connected and therefore the ...