Interconnection networks for parallel molecular dynamics simulation based on hamiltonian cubic symmetric topology
A class of interconnection networks for efficient parallel MD simulations based on hamiltonian cubic symmetric graphs is presented. The cubic symmetric graphs have many desirable properties as interconnection networks since they have a low degree and are vertex- and edge-transitive. We present a method for scheduling collective communication routines that are used in parallel MD and are based on the property that the graphs in question have a Hamilton cycle, that is, a cycle going through all vertices of the graph. Analyzing these communication routines shows that hamiltonian cubic symmetric graphs of small diameter are good candidates for a topology that gives rise to an interconnection network with excellent properties, allowing faster communication and thus speeding up parallel MD simulation.