A key challenge in reliable multicast is to reduce the retransmission cost such as bandwidth usage and delay. Local recovery approach employing replicated servers has recently been shown to be effective in reducing this cost. The basic idea in the local recovery approach is to perform data replication on a subset of routers inside the network known as replicated servers. Each replicated server is attached to a router and responsible for the retransmission requests from receivers within its local area. The replicated server itself can recover lost packets from the sender or another replicated server. Clearly, the placement of replicated servers is crucial in determining system performance. In a multicast network, packets are forwarded from a source (sender) to group receivers along a dis...[ Read more ]

A key challenge in reliable multicast is to reduce the retransmission cost such as bandwidth usage and delay. Local recovery approach employing replicated servers has recently been shown to be effective in reducing this cost. The basic idea in the local recovery approach is to perform data replication on a subset of routers inside the network known as replicated servers. Each replicated server is attached to a router and responsible for the retransmission requests from receivers within its local area. The replicated server itself can recover lost packets from the sender or another replicated server. Clearly, the placement of replicated servers is crucial in determining system performance. In a multicast network, packets are forwarded from a source (sender) to group receivers along a distribution tree where the source is the root, the receivers are the leaves and the multicast-capable routers are the internal nodes. In this thesis, we consider the problem of placing multiple replicated servers within the multicast-capable routers. This problem can be formulated as an optimization problem by taking into consideration the characteristics of the network topology. We first present the problem formulation for the multicast tree rooted at a single source and then the general graph with multiple sources, where there is a distribution tree for each source. We then show that this is a NP-hard problem and present a number of heuristics-based algorithms for server replication. Finally, in order to investigate various tradeoffs in terms of cost and algorithm complexity, we carry out comparison studies among different heuristic algorithms.