A maximum entropy approach to the analysis of loss systems

The maximum entropy principle allows us to obtain an estimate of the state probability distribution when our knowledge about the system is only the expected value of some functions of the system state. In this paper we explore the maximum entropy approach to evaluate the state probability distribution of loss systems where the call arrival processes are unknown or not specified. We present some numerical results for a simple loss system and compare them with the well-known Erlang loss model. We also generalize the formulation to the case of multiclass traffic, which will have practical applications in the performance analysis of B-ISDN, all-optical networks, and other networks which can be modeled as loss systems.