Analysis and optimization of hierarchical reliable transport protocols

Several schemes have been proposed that address the problem of scalable reliable multicast. The Tree-based Acknowledgement protocols family achieve scalable reliable transport in one-to-many communications based on a hierarchical approach. It has been recently selected as a standard-track possible protocol architecture by IETF. In these protocols, dedicated servers are used to perform error recovery and statistical aggregations. Nevertheless a number of dimensioning issues remain open, such as the number of required servers, and the optimal number of hierarchical levels. In order to evaluate and dimension such protocols, we consider an analytical model using point processes, where points are scattered in the plane in order to represent receiver locations in the Internet, and a loss model which takes correlated losses into account. We define cost functions based on the number of retransmissions in the control tree for a packet sent by the source. We proposed random geometric model allowing us to find explicit expressions for the cost functions, w.r.t. to a limited set of parameters: number of receivers and number of repair servers. We are then able to deduce optimal control tree structures (optimal number of repair servers), and generic dimensioning rules.