The design principle of the Internet has been changed from best-effort service to QoS (Quality of Service)-differentiated service paradigm. In best-effort service networks, all application streams are served identically through a single FIFO (First-In First-Out) queue at each router. In contrast, in QoS-differentiated service networks, application streams are mapped into a given number of QoS queues, and served differently to meet individual QoS requirements. For an optimal design of such a mapping between application streams and a given number of QoS queues, we need individual application traffic models which not only represent the stochastic nature of individual application streams respectively but also can be used as building blocks for the mapping.
In this thesis, we analyze the characteristics of individual application streams in the existing best-effort Internet, and define some traffic types with respect to their QoS requirements. By showing stochastic independence among application streams, each application stream is modeled, being separated from other application streams. For an application stream model, two independent Markov Chains are used to represent interarrival-time and packet-length processes separately. Such an independent modeling of interarrival-time process and packet-length process for an application stream is justified by showing empirically the independence between the two processes. Through extensive queueing simulations, we show that the proposed traffic models can predict queueing performance accurately enough in the design of QoS-differentiated networks.
