Search

High-order numerical schemes implemented on high-performance parallel computers are of special interest for contemporary numerical simulations, especially in computational fluid dynamics. In this study, first, a high-order parallel flow solver is presented for some test cases of aerodynamic simulations. The flow solver is based on a discontinuous Galerkin finite element method on arbitrary grids with different orders of polynomial approximation for solving the compressible flow model. Second, the distributed-memory parallel implementation of the flow solver is benchmarked on a shared-memory multicore system. A distributed-memory parallel application can be executed on shared-memory architectures by assuming that each of the parallel processes assumes separate memory address space, although all are present in a common memory bank. This approach can offer an effective measure to address several issues related to limited resources, especially for uninterrupted electric supply. The scalability of the parallel application is analyzed by varying the problem workload per process for the test cases. For some test cases in the present study, over 90% parallel efficiency per process is also observed. The performance of the distributed-memory program on the shared-memory architecture establishes suitability and robustness of the approach for small to medium scale problems, at least. [ABSTRACT FROM AUTHOR]

An efficient parallel scheme based on the nodal discontinuous Galerkin finite element method (nodal-DGFEM) for the numerical solution of the partial differential equations governing fluid flow phenomena is discussed. The flow solver is demonstrated to perform numerical simulation of two-dimensional flow regimes on unstructured triangular grids. The parallel implementation serves to fulfill the requisition of the numerical method regarding high-performance computing resources. The distributed memory programming model with the domain decomposition approach is adopted. The message passing interface library is used for communication among the parallel processes, which are assigned domain-decomposed subproblems. The presented parallelization strategy accurately and efficiently tackles the communication of multi-node data on the element edges between the neighboring parallel processes. The efficacy and efficiency of the parallel solver are demonstrated through solving the well-known problem of non-viscous isentropic convecting vortex flow on parallel systems. The parallelization would extend the scope of the DGFEM by producing solutions in reasonable time frames. [ABSTRACT FROM AUTHOR]