Isabel Beichl and
Jim Blue, ACMD
Francis Sullivan, Supercomputing Research Center
It is not known how to deal with truly parallel problems efficiently, that is, problems where processors communicate with other processors randomly and the entire machine looks to the user like one large flat memory space. We address these questions in our development and implementation of a parallel algorithm to simulate molecular beam epitaxy (MBE) using a domain decomposition approach with a MIMD model of a parallel machine. We have implemented the parallel version on a CM-5 using the node language, CMMD. Communication between processors must occur on the boundary of each domain where atoms may move among the domains. Communication is done with message passing. A method of conflict resolution along the boundary has been developed and synchronization of the concurrent Monte Carlo simulations has been studied. MBE simulations using extremely large domains are possible with this method. We are in the process of moving the simulation to the SP-2. We are also extending this work to more practical cases of face centered cubic and body centered cubic crystals.