Data-parallel DNS of turbulent flow

This contribution deals with direct numerical simulation (DNS) of incompressible turbulent flows on parallel computers. We make use of the data-parallel model on shared memory systems as well as on a distributed memory machine. The combination of fast parallel computers and efficient numerical algorithms (keywords: high-order discretization and efficient time-integration) opens the possibility of performing DNS of complex flows at Reynolds numbers in the range 10(4)-10(5). In this paper, we consider two examples of complex flows past bluff bodies: a flow past a long, square cylinder (Re = 22,000) and a flow in a channel with surface mounted cubical obstacles (Re = 13,000). These flows have served as test case for a number of workshops on turbulence modeling for Reynolds-averaged and large-eddy simulations.