Performance analysis of packet networks

We develop analytical models for packet network performance which consist of state equations linking nodes input parameters to output parameters and state equation linking neighboring nodes. These state equations are then solved iteratively until convergence is reached. From the input and output parameters, steady-state performance parameters such as delay, throughput, and blocking are derived.; Within the context of Multihop networks subject to geometric traffic sources we generalized Brassil-Cruz performance model for packet arrivals subject to the independence and memoryless assumptions for uniform traffic and deflection routing. Our generalized model apply to arbitrary network topologies, with or without buffers, and with an improved computational efficiency. Moreover, we extend our model for the store-and-forward routing strategy.; Then, we developed a computationally efficient novel model for Manhattan Street Networks subject to packet train traffic uniformly distributed with a variation of the wormhole routing algorithm.; Lastly, via simulation models (modified version of ns-2), we investigated self-similarity impacts on data network performance when mixing voice and data traffic within the internet. We accounted for the emergence of QoS standards by allowing for packet classification and bandwidth access arbitration.; Our research most significant contributions include: (1) Extension of Brassil-Cruz model for geometric traffic sources to arbitrary topologies, with or without buffers, and with computational efficiency. (2) Analytical model for arbitrary topologies subject to geometric traffic sources with store-and-forward routing. (3) State probability expressions for degree two switching and queueing of fixed size, geometric inter-arrival packet trains. (4) Analytical model for symmetrical Manhattan Street Networks subject to uniformly distributed fixed size, geometric inter-arrival packet trains with a modified wormwhole routing algorithm. (5) Simulation model to quantify performance issues associated with mixing voice and data traffic within an internet network, together with simple models for engineering such networks.