A new framework for developing unstructured grid based Petascale CFD codes has been developed. The framework incorporates very fast parallel domain decomposition and partition topological information construction methodologies and is suited for massively parallel computing. The framework handles any general simplex; but for the present study only tetrahedra, prisms, pyramids and hexahedra are considered. Very large hybrid grids are efficiently handled due to a novel distributed storage scheme. Representative tests confirm the benefits of the developed framework. The framework holds good potential for implementation of high accuracy unstructured grid based Petascale CFD codes for high fidelity simulations.

Keywords: Framework, Mesh partitioning, Parallel domain decomposition, Unstructured grids, High order methods, CFD, Petascale computing, DNS.