Two-Level Tightly Concatenating Algorithm for Multiphysics Parallel Numerical Simulations
MO Ze · Chinese Journal of Computers · 2004
Multiphysics parallel numerical simulations are usually essential to simplify researches on complex physical phenomena in which several physics processes are tightly coupled. It is very important on how to concatenate those coupled physics processes for fully scalable parallel simulation. Meanwhile, three objectives should be balanced, the first is efficient data transfer among simulations, the second and the third are efficient parallel executions and simultaneously developments of those simulation codes. This paper presents two concatenating algorithms for multiphysics parallel numerical simulations coupling radiation hydrodynamics with neutron transport on unstructured grid. The first algorithm, Fully Loosely Concatenation (FLC), focuses on the independence of code development and the independence running with optimal performance of code. The second algorithm, Two Level Tightly Concatenation (TLTC), focuses on the optimal tradeoff among above three objectives. Theoretical analyses for communicational complexity and parallel numerical experiments on hundreds of processors on two parallel machines have shown that these two algorithms are efficient and can be generalized to other multiphysics parallel numerical simulations coupling radiation or neutron transport with hydrodynamics. Especially, algorithm TLTC is linearly scalable and has achieved the optimal parallel performance.