In this section we present the issues involved in parallelizing Monte Carlo eigenvalue (using source iteration and fission matrix) and perturbations (using correlated sampling applied to the fission matrix approach) calculation algorithms [Maj95b]. Monte Carlo eigenvalue and perturbation algorithms differ from fixed source algorithms in the sense that both require an iteration procedure to determine the source distribution and the eigenvalue. At the end of each iteration, all the processors (master and slaves) need to be synchronized and the slave processors need to exchange information with the master processor. These synchronization and communication requirements increase parallelization overhead compared to fixed source Monte Carlo algorithms. The basic parallel algorithm used for eigenvalue and reactivity type simulation is shown in figure 5.5. We present in this section observed speedup performances from the IBM-SP2 parallel computer and fit predicted theoretical curves to these observed speedup results.
for the IBM-SP2