Analysis of the distribution of consecutive cell losses in an ATM switch using stochastic activity networks

Advancements in fast packet switching technology have made possible ATM-based B-ISDNs and integration of diverse telecommunication services. High throughput requirements and diverse services place stringent quality of service (QoS) demands on the associated switches. The often computed average cell-loss probability (clp) is an interesting but not a sufficient measure, since it is averaged over both time and all switch inputs. There exist many applications that are sensitive to the pattern of cell loss, where despite a low average dp, consecutive cell loss implies insufficient QoS. Further, the cell loss pattern as seen by the switch and a specific port can differ even if the average clp for the two is the same. It is therefore important to distinguish between the loss behavior at the switch and at a port, especially when examining QoS as perceived by the users of a specific switch port. In this paper, we use stochastic activity networks (SANs) to analyze the distribution of consecutive cell loss, both with respect to the ATM switch as well as a specific port. To do this, we use UltraSAN, a SAN-based performance modeling and analysis tool to construct and solve the detailed Markov processes associated with the switch and a bursty workload. Our results provide useful information, both about the usefulness of SANs and UltraSAN, as well as the importance of sophisticated measures, such as the distribution of consecutive cell losses, when evaluating ATM switch designs.