Parallel solution of unsteady incompressible viscous flows using multiblock structured grids

A parallel general-multiblock incompressible viscous solver capable of handling complex geometries with embedded rotating propulsor is developed. The sequential iteration hierarchy is redesigned to incorporate scalable parallelism through a modified Block Jacobi Symmetric Gauss Seidel (BJ-SGS) scheme without significantly diminishing algorithmic performance. The BJ-SGS scheme, along with its variants, is demonstrated to have good convergence properties. An efficient and scalable message-passing framework is used to implement the parallel algorithm and obtain a high performance flow solver. Poly-algorithms keyed to processor and problem characteristics can be used to further improve parallel performance. The code is verified using results from the sequential flow solver and validated using analytical results and experimental data. The verification and validation cases range from an accelerating cylinder and bare submarine hull to fully configured submarine geometries with rotating propulsors. Solutions were obtained on IBM-SP2 and Cray T3E machines for 1.6 million point grids with speeds of up to 2 Gigaflop/s on 51 processors. Scalability studies are presented that indicate that on current hardware parallel efficiencies of 80% and more can be achieved on up to 400 processors using appropriately sized grids. The implementation and use of pre-processors and other utilities to improve solution reliability and speedup problem setup are also discussed.