Performance of TCP with Multiple Priority Classes

We consider the dimensioning problem for Internet access links carrying TCP traffic with two priority classes. To this end, we study the behaviour of TCP at the flow level described by a multiple-server Processor Sharing (PS) queueing model with two customer classes, where the customers represent flows generated by downloading Internet objects; the sojourn times represent the object transfer times. We present closed-form expressions for the mean sojourn times for high-priority customers and approximate expressions for the mean sojourn times of low-priority customers. The accuracy of the model is demonstrated by comparing results based on the PS model with "real" TCP simulation results obtained by the well-known Network Simulator. The experimental results demonstrate that the model-based results are highly accurate when the mean object size is at least 10 IP-packets, and the loss rate is negligible.