Abstract: Loss networks have long been used to model various types of communication systems, including circuit-switched networks, local area networks, multiprocessing architectures, mobile/cellular networks and broadband pattern networks. Such networks often use admission controls, such as trunk reservation, to optimize the performance (often revenue). Trunk reservation is a threshold policy designed to protect the more valuable traffic, especially under high usages.
Due to the computational complexity of calculating exact solutions, much attention has been devoted to efficient approximation of performance measures such as the percentage of calls lost. One class of these is known as reduced load approximations, the simplest of which is the Erlang Fixed Point approximation. This has been extended to the case of trunk reservation using a birth and death approximation technique. However, a key assumption of the Erlang Fixed Point method is that links are assumed to block independently. Certain methods of specifically accounting for dependence in this blocking have been developed for an uncontrolled setting, and this paper explores possible extensions to the controlled case, with the aim of gaining insight into the essential elements of an effective approximation. In order to isolate the dependency factor, we restrict our attention to the case of a highly linear network.
Acknowledgement: This worked was funded by the Australian Research Council.
Last modified: 6th November 1999