Optimized algorithm for solving the nodal diffusion method on shared memory multiprocessors

摘要

Nodal methods play a special role in reactor physics calculations. In recent papers the high computational efficiency of nodal methods has been established and the development of more efficient algorithms tailored to the advanced architectures of modern day computers proposed. The rapidly changing architectures of today's computer influence the way codes have to be programmed so that reasonable speed up and efficiency are attained. We have applied these concepts in solving the one-group neutron diffusion equation in two-dimensional geometry on parallel computers like the Intel iPSC/2 hypercube and the Sequent Balance 8000. The efficiency of the hypercube for the neutron diffusion equation is highly determined by the message passing scheme; on the other hand, on a shared memory processor like the Sequent, it is dependent on the manipulation of variables in shared memory. In this paper, we present a scheme on shared memory processors which produces very high computing efficiencies in agreement with Amdahl's law. 6 refs., 1 fig.