Evolutionary Design of Collective Communications on Wormhole Networks

This thesis describes the technique of the evolutionary design aimed at scheduling of collective communications on interconnection networks of parallel computers. In order to avoid contention for links and associated delays, collective communications proceed in synchronized steps. A minimum number of steps is sought for the given network topology, wormhole (pipelined) switching, minimum routing and given sets of sender and/or receiver nodes. The proposed technique is not only able to re-invent optimum schedules for known symmetric topologies like hypercubes, but it can find schedules even for any asymmetric, irregular, multistage and fat topologies in case of general many-to-many collective communications. In most cases, the number of steps reaches the theoretical lower bound for the given communication pattern; if it does not, non-minimum routing can provide further improvement. Optimal schedules may serve for writing high-performance communication routines for application-specific networks on chip or for the development of communication libraries in the case of general-purpose interconnection networks.