Modeling and optimization of ON/OFF sources transmitting data over an unreliable network

In modern broadband systems several on/off traffic sources transmit over a common telecommunication network. Since the network has a finite capacity, the probability of a breakdown due to congestion increases the more sources simultaneously send data packets. In case of a failure due to congestion a time consuming recovery procedure resolves the overload situation. During this procedure no service can be provided for the traffic sources. In this paper we model this scenario with generalized stochastic Petri nets and investigate the influence of several system parameters on the performability of the system. For these parameters (number of traffic sources in the system, length of on/off period in send modus and length of send/idle period of the sources) we try to find optimal settings which maximize the number of transmitted data packets. Optimization was performed with the tool REMO. Moreover we model and optimize a dynamic decision rule which decides, dependent on the load situation, whether to admit an additional source or not. These investigations could be used to create guidelines for service providers how to optimally operate the system.