On the accuracy of admission control tests

In guaranteed performance networks, a connection admission control (CAC) algorithm determines whether or not a new connection can be admitted to the network such that all connections will obtain their required quality of service (QoS) guarantees. However, inaccuracies in a CAC algorithm lead to either under utilization of resources or violations of the promised QoS guarantees. We evaluate the accuracy of a broad range of CAC algorithms that have been proposed in the literature. Our approach is to perform experiments with long traces of MPEG compressed video, considering both heterogeneous traffic streams and priority schedulers. We compare the admissible regions and QoS parameters predicted by our implementations of the CAC algorithms with those obtained from trace driven simulations. We then identify the key aspects of a CAC test for achieving a high degree of accuracy and hence a high statistical multiplexing gain.