Challenges in particle tracking in turbulence on a massive scale

An important but somewhat under-investigated issue in turbulence as a challenge in high-performance computing is the problem of interpolating, from a set of grid points, the velocity of many millions of fluid particles that wander in the flow field, which itself is divided into a larger number of sub-domains according to a chosen domain decomposition scheme. We present below the main elements of the algoithmic strategies that have led to reasonably good performance on two major Petascale computers, namely Stampede and Blue Waters. Performance data are presented at up to 16384 CPU cores for 64 million fluid particles.