Comparative performance of scheduling strategies for switching and multiplexing in a hub based ATM network: a simulation study

Abstract

We model an ATM network comprising an input queuing cell switching hub. Customer access lines are multiplexed into the relatively faster input links of this hub. Each access multiplexer is fed by on/off sources. Motivated by the fact that an ATM network should support both bursty as well as smooth traffic, we consider the scenario in which some multiplexers are fed by sources with long bursts of cells, and others by sources with short bursts. Here we report the results of a detailed simulation of this hub-based ATM network. Our objective is to compare the performance of various strategies for scheduling cell service in the access multiplexers, and in the ATM switch. The simulation results confirm what might be expected from the results of our earlier analytical modelling of the multiplexer and the switch in isolation (this analysis assumed a particular Markovian model for the aggregate cell arrival processes into the ATM switch). In particular, we find that, if mean burst delay is the performance criterion then, for a small ratio of ATM-link to customer-access-line speed (≤3 for the models and parameters we use), the more bursty traffic should undergo burst level multiplexing at the access multiplexer, and should be given lower priority during output contention resolution in the input queueing hub. If worst case delay performance (e.g., the 99.9 percentile of the end-to-end delay of a burst) is a consideration, however, we find that, even for small ATM-link to access-line speed ratio, the best combination of strategies is that the access multiplexers should multiplex the packets from the access lines in a round-robin fashion, and the switch should still give lower priority to the burstier traffic.