Efficient Parallelization of the Monte Carlo N Particle (MCNP) K-code
Nathan Carstens, Massachusetts Institute of Technology
Deterministic nuclear reactor design codes can no longer be easily used to design intricate next generation reactors. Monte Carlo based codes are rapidly replacing deterministic codes with the Monte Carlo N Particle (MCNP) code dominant in this category.
Unfortunately, MCNP k-code calculations currently require significant amounts of communication during parallel calculation, fundamentally limiting massively parallel computation. This poster will present an algorithm and modifications to the MCNPX code that greatly minimize k-code communication in parallel. It is hoped that by allowing massively parallel MCNP k-code calculations, significant new nuclear reactor design problems can be addressed in the near future.
Abstract Author(s): Nathan Carstens