Parallelization of a Hybrid UnstructuredGrid Density-based Flow Solver
M. R. Nived, Vinayak Eswaran · 2018
Compressible flow is governed by partial differential equations which are hyperbolic in nature. Density-based solvers using explicit time-stepping are inherently slower to reach convergence to steady state as the time-step size is limited by the CFL stability criterion. The implementation of implicit time-stepping algorithms permits the use of quite large CFL numbers which makes the convergence to steady state much faster. Using the Message Passing Interface(MPI) library to execute the same code on equal sized partitions of a mesh employing multiple cores/ processors, can further accelerate the convergence till steady state by reducing the computational time taken per time-step. This thesis outlines a data-parallel algorithm implemented on an in-house serial implicit laminar solver that produces results nearly identical to those obtained in serial computations. The serial code data structure optimization and the implementation of a mesh partitioning sub-routine are carried out prior to parallelization. A study of its performance and scalability on an Intel Xeon E5-2620 2.20 GHz workstation having 16 processors is carried out using a few benchmark cases. The parallel code gives speed-up close to linear without loss in accuracy