New View on Adversarial Queueing on MAC

This work proposes an average-case performance measurement as a meaningful and non-trivial way of evaluating and comparing queueing protocols on a multiple-access channel (MAC). We prove that Little's Law holds for this way of measuring latency and number of packets in the system, also for adversarial packet arrival (which was not the case for the previously considered worst-case measurement), and analyze performance of popular MAC protocols. Interestingly, some of them have asymptotically the same average and worst case queue sizes, but not packet latencies, while others exhibit entirely opposite behavior. We also propose a consistent methodology for classifying queueing protocols, based on protocol features, such as synchronization, collision detection, queue-size awareness etc.