In general, the traffic characteristics of the individual applications that constitute the aggregate traffic on a network can be important for capacity engineering. In this paper, we demonstrate based on mathematical analysis the value of application specific measurements even when there is no service differentiation. In other words, under certain assumptions, we obtain the result that errors in engineering can occur, and in particular, under-engineering can occur when traffic characteristics of individual applications are ignored. The assumptions are that the individual applications can be modeled adequately as Fractional Brownian Motions and that measurements are available only at relatively coarse time scales. The results in this paper emphasize the value of collecting fine-grained traffic measurements.

A discrete-time queuing model for the performance of a meteor-burst packet communication system is analyzed using matrix analytic methods. Not only is the system subject to interruptions due to the intermittence of the ionization layer, but its analysis is further complicated by the necessity to retransmit packets that have error or that occur at the tail end of a period of availability of the system. The authors' model takes such complexities into account. It is also of independent methodological interest in that it provides an exact analysis of a general queuing model with service interruptions. For the application at hand, the authors demonstrate the feasibility of the algorithms by a selected set of numerical examples. The algorithm can be used to ascertain the effects of the packet size, the bit error rate, the sync acquisition time, and other variables on system performance.>