Official URL: http://arxiv.org/pdf/1405.5753v1.pdf

In random access protocols, the service rate depends on the number of stations with a packet buffered for transmission. We demonstrate via numerical analysis that this state-dependent rate along with the consideration of Poisson traffic and infinite (or large enough to be considered infinite) buffer size may cause a high-throughput and extremely long (in the order of hours) transitory phase when traffic arrivals are right above the stability limit. We also perform an experimental evaluation to provide further insight into the characterisation of this transitory phase of the network by analysing statistical properties of its duration. The identification of the presence as well as the characterisation of this behaviour is crucial to avoid misprediction, which has a significant potential impact on network performance and optimisation. Furthermore, we discuss practical implications of this finding and propose a distributed and low-complexity mechanism to keep the network operating in the high-throughput phase.