The capability of multidestination wormhole allows a message to be propagated along any valid path in a wormhole-routed network conforming to the underlying base routing scheme. The multicast on the path-based routing model is highly dependent on the spatial locality of destinations participating in multicasting. In this paper, we propose two proximity grouping schemes for efficient multicast in wormhole-routed mesh networks with multidestination capability by exploiting the spatial locality of the destination set. The first grouping scheme, graph-based proximity grouping, is proposed to group the destinations together with locality to construct several disjoint sub-meshes. This is achieved by modeling the proximity grouping problem to graph partitioning problem. The second one, pattern-based proximity grouping, is proposed by the pattern classification schemes to achieve the goal of the proximity grouping. By simulation results, we show the routing performance gains over the traditional Hamiltonian-path routing scheme.
Journal of systems architecture 46(10), pp.919-930