A. Córdoba, J. Villadangos, F. Fariña, J.R. González de Mendívil, and J.J. Astrain (Spain)
Distributed Systems, Deadlock Resolution, Generalizeddeadlock, P out of Q Request Model, PerformanceMeters, Simulation
The performance of two deadlock-handling techniques for distributed systems under the generalized request model is compared in this work. The two handling deadlock techniques are the time-out method and the distributed deadlock resolution algorithm proposed by Chen et al. [9 10]. The original algorithm has been implemented to work in a dynamic model, that is, a model in which a process may change its status form active to blocked, blocked to active, or may abort as consequence of the resolution of a deadlock. These modifications include multiple concurrent initiations by nodes in order to maintain the liveness property of the algorithm (every deadlock must be resolved), and the inclusion of an efficient mechanism in order to calculate the set of victims once a deadlock is locally detected. The main conclusion obtained from the experimental results is that the throughput provided by the deadlock resolution algorithm is very similar to the one, which would be obtained by the optimal time-out.
Important Links:
Go Back
