Scalability analysis and domain decomposition of large eddy simulations of ship wakes

Simulation of turbulent fluctuations in ship wakes is one of the complex applications of large eddy simulations in computational fluid dynamics (CFD). Ship wake simulation requires extensive computations and large amounts of computer resources. The accuracy of ship wake prediction is limited by the memory of the workstation. We present the parallel implementation of large eddy simulations (LES) of a flat plane wake using a data domain decomposition technique for a cluster environment. We present the results of the implementation executed on a cluster of workstations. Also, we show how the implementation scales up with the number of workstations and that it is possible to obtain better accuracy by increasing the number of workstations in the cluster system. Moreover, we show how we can predict the number of workstations required to execute the simulation for given accuracy level based on the number of grid nodes, or estimate the execution time for a certain cluster.