Distributed Fuel-Management Computation using RFSP, WIMS-AECL and PRVM

Abstract

The Parallel Virtual Machine (PVM) software package was used to build an interface between RFSP and WIMS-AECL to enable history-based, local-parameter reactor fuel-management simulations in which batches of lattice-cell transport and burnup calculations can be made in parallel. The interface is based on the master/slave crowd-computation model. For slave computers numbering from one to twenty, the overhead spent by the one master preparing input for the slaves and processing their outputs was observed to be small in comparison with the computing time spent by the slaves themselves. Anticipating the availability of a much larger network of slaves in the future, two potential computational bottlenecks that might arise are described, and possible remedies for them are outlined.