Path Selection Techniques for SCTP Multihoming

This paper proposes a new innovative concept for using machine learning techniques to optimize the overall data throughput over low bandwidth networks with several paths for a parallel transmission of data. The multihoming feature of the stream control transmission protocol is used to transfer user data parallel over several end-to-end paths. The work focuses on the selection of the actual best path, where the capacities of the present paths are not necessarily known in advance. The path selection is adapted in an autonomic and flexible way by a new network component, the so called intelligent network placed at the network layer. Using this adaptive mechanism it is possible to implement applications like long-distance medicine in mobile scenarios.