An efficient multilevel master-slave model for distributed parallel computation

The master-slave (MS) parallel computing model is one of the most widely used model in a networked computing environment due to its ease of implementation. This model, however, suffers from the disadvantages of the sequential generation of slave processes and heavy communication overheads imposed on the master processor. To overcome this problem, we present in this paper an efficient multilevel master-slave (MMS) scheme which is especially useful for solving decomposable large-scale problems such as structure mechanics or dynamics problems with rotational symmetry, on networked workstations. Our MMS model implements the MS model at multiple levels and generates processes using a special class of tree structures, allowing parallel creation of slave processes. It also improves performance in the distribution of initial data and merging of computed results to and from slave processes. We shall describe the generation of processes using different MMS structures to generate a prescribed number of processes and to broadcast global data to all processes. We then present the implementation of the optimal MMS model via PVM on a networked computer system consisting of workstations for a plate-bending problem that is discretized using the finite strip method The performance of our numerical experiments employing this MMS model is reported to demonstrate its efficiency. This scheme can be applied equally well to other types of problems that can be decomposed using the Fourier decomposition or circular decomposition, no matter whether the physical problem is discretized by the boundary element method or finite element method. Transactions on Modelling and Simulation vol 22, © 1999 WIT Press, www.witpress.com, ISSN 1743-355X