Modeling and regulation of host traffic in ATM networks

For any connection admission control (CAC) algorithm to work correctly and efficiently, accurate information of the traffic flow out of the host systems is required. We develop several approximation approaches for modeling the traffic flows of hard real-time connections. We show that without accurate traffic characterization, the CAC algorithm may do either of the following: (1) admit connections that may cause network congestions which results in violating connection deadline requirements; or (2) pessimistically reject many connections whose QoS can be guaranteed. We propose a traffic approximation model that can characterize the traffic correctly and efficiently, achieving a higher admission probability. From our experimental data we observe that the source traffic from a typical host is bursty. This burstiness may cause congestion within the network. To overcome this problem, we propose and analyze a simple traffic regulation mechanism at the application layer. The performance evaluation data shows that in the regulated system the traffic burstiness is lower and the probability of a connection being admitted is higher than the unregulated system.