Parallel Implementation of the PHOENIX Generalized Stellar Atmosphere Program. III: A parallel algorithm for direct opacity sampling

We describe two parallel algorithms for line opacity calculations based on a local file and on a global file approach. The performance and scalability of both approaches is discussed for different test cases and very different parallel computing systems. The results show that a global file approach is more efficient on high-performance parallel supercomputers with dedicated parallel I/O subsystem whereas the local file approach is very useful on farms of workstations, e.g., cheap PC clusters.