Using TCP models to understand bandwidth assurance in a differentiated services network

In this paper, a comprehensive analytical model to predict the bandwidth achieved by aggregates of TCP flows in a Diffserv network is presented. The model predicts achieved bandwidth in three different cases: an over-provisioned network, an under-provisioned network, and a near-provisioned network. In developing the model, we ensure that all parameters are measurable using standard tools and information available from routers and network management tools in today's networks. Simulation was used to establish the validity of the model and understand its scope of applicability and limitations. Using the model, we explain why achieved excess bandwidth is based on factors such as RTT, packet size, and CIR. Finally, we present a novel extension of the model to predict the bandwidth of TCP flows in a Diffserv network with multiple congested nodes.