HIGH-PERFORMANCE SPECTRAL SIMULATION OF TURBULENT FLOWS IN MASSIVELY-PARALLEL MACHINES WITH DISTRIBUTED-MEMORY

Here we have demonstrated the possibility of very high performance in the implementation of a global spectral methodology on a massively par allel architecture with distributed memory. Spectral simulations of ch annel flow and thermal convection in a three-dimensional Cartesian geo metry have yielded a very high performance-up to 26 Gflops/s on a 512- node CM5. In general, implementation of spectral methodology in parall el processors with distributed memory requires nonlocal interprocessor data transfer that is not restricted to being between nearest neighbo rs. In spite of their increased communication overhead, better perform ance is possible in global methodologies owing to their dense matrix o perations and organized data communication. In this paper we outline a general methodology for the data-parallel implementation of spectral methods on massively parallel machines with distributed memory. Follow ing the steps presented here, very high performance can be obtained on a wide variety of massively parallel architectures.