Sharing Servers in Non-Markovian Queueing Systems with Delay Objectives

Experience has shown that efficiency often increases when separate queueing systems are combined into a single system. We define efficiency in terms of the number of servers required to meet the delay objectives of each traffic stream, and we characterize conditions under which the combined system is more efficient. In particular, we consider serving queues with different traffic characteristics and show that it is efficient for one team to serve more than one queue whenever the disparity in the service times is not too large. If more than one traffic stream is served by one team, then two disciplines are considered: the first-come first-served queueing discipline (FCFS) and the delay ratio queueing discipline (DRQ).