Resource allocation for statistical QoS provision in buffered crossbar switches

In this paper, we quantify the resources-bandwidth and credits-required to guarantee that the probability that a prescribed class-based target packet delay will be violated is upper bounded by a prescribed violation probability. We assume that the upper bound on the traffic arrival process for each class can be described statistically via effective envelopes. Our results are derived using a network calculus-based model of the system. We consider both deterministic and probabilistic delay guarantees. We find that the resources required to guarantee a probabilistic delay upper bound are less than those required to guarantee a deterministic delay upper bound of the same value.