A scalable multicast service for mesh networks

The authors investigate the scalability of a multicast algorithm designed for wormhole-routed mesh networks. The algorithm, known as the U-mesh algorithm, is shown to scale well in four ways: with the dimension of the mesh, with the number of destinations, with the system size, and with the problem size. It is demonstrated that the only factor that affects the multicast latency is the number of destinations and that, for a given number of destinations, the number of time steps required to perform a multicast operation is minimal. Performance measurements of implementations on a 64-node nCUBE-2 and a 168-node Symult 2010 are given