Reliability in Distributed Queuing Systems in the Presence of Random Delays

Service reliability in a distributed queuing system can be assessed by means of the probability of successfully serving queued tasks in a heterogeneous network of nodes. Moreover, randomness in delays imposed by the communication medium adds to the uncertainties in the shared information among these faulty nodes, which complicates the dynamics of the queuing system. A novel approach based on stochastic regeneration is used to obtain renewal equations that characterize the service reliability. The theory is applied to develop optimal load-balancing policies for a two-node distributed-computing system, which is then systematically extended to a multi-node system yielding a computationally efficient suboptimal solution. Index Terms renewal theory, queuing theory, reliability, distributed computing, load balancing