A Semi-centralized Algorithm to Detect and Resolve Distributed Deadlocks in the Generalized Model

Abstract

To detect and resolve distributed deadlocks in the generalized model, a few algorithms have been proposed. Most of them are distributed algorithms which are based on the diffusing computation technique where propagation of probes and backward propagation of replies are required. On the contrary, centralized algorithms send the dependency information directly to the initiator. However, existing algorithms do not distinguish local deadlocks from global deadlocks. In this paper, we present a semi-centralized algorithm which effectively detects and resolves local deadlocks prior to global deadlocks. To be specific, a consistent snapshot of dependency information of involved nodes is recorded in an information table at each site. Therefore, the detection and resolution of local deadlocks can be performed at each site before the dependency information is sent to the initiator to further detect global deadlocks. The proposed algorithm also handles the problem of phantom edges and concurrent executions. Performance analysis shows that the performance of our algorithm is close to or better than that of the existing algorithms in terms of execution time, message traffic and message size.