Performance analysis of space–time priority queues

Abstract Depending on the application level quality requirements, packets on the Internet can broadly be classified into two classes: Streaming (S) and Elastic (E). Streaming packets require low delays, but can tolerate packet losses without degrading the user level performance. Elastic traffic has less stringent delay demands, but its performance is very sensitive to packet losses, since the latter trigger costly packet recovery mechanisms. In this paper, we analyze a space–time priority queue (STPQ), a queueing discipline that can satisfy the performance criterion of both classes. Streaming packets get time priority, i.e., they are served with priority over elastic packets, while elastic packets get space priority, i.e., in the event the buffer is full, the arrival of an elastic packet will trigger, with probability α , the push-out from the queue of a streaming packet. In our main analytical result, we have developed efficient algorithms to compute the steady-state probabilities and the mean delay of STPQ. We have also analyzed the queue in the light-traffic regime, which allows us to derive a closed-form approximation. Numerical results illustrate that the STPQ can provide a low delay to streaming packets without significant degradation of the blocking probability of elastic packets.