EAGER: Reaching Higher in Numerical Simulations of Turbulence

Georgia Tech Research Corporation, Atlanta GA

Investigators

Abstract

1139037 Yeung Turbulent fluid flows occurring in many fields of science and engineering are characterized by the complexities of fluctuations over a wide range of scales, to the extent that their detailed computation is a premier challenge worthy of the use of world-class computational resources. The core objectives of this project are to pursue intensive efforts both computationally and scientifically to ensure successful conduct of a massive investigation of mixing and dispersion in turbulent flow, at a new world-record level of resolution on a new machine of true sustained Petascale capability. The targeted simulation will reach a resolution of 8192-cube (more than half a trillion) grid points and a Reynolds number which exceeds most laboratory experiments. In order to succeed, there are intellectual challenges to be overcome in pursuing the intricacies of several emerging new computational paradigms, including hybrid multi-cored programming and overlap between computation and interprocessor communication subject to the logic of the numerical methods employed as well as characteristics of the new machine hardware. The conduct of the simulations and subsequent data analyses at Reynolds number higher than ever achieved before will enable important progress in the study of intermittency of turbulent fluctuations and the effect of these fluctuations on many aspects of turbulent mixing and dispersion. Strategies for code development and exploration of new programming models for the full simulation code on the future NSF Track 1 Petascale machine carry inherent risk, because of the massive size of the system whose characteristics are not fully known or determined at this time. However the urgency for supporting code development is great since the machine is scheduled to become available in 2012 and CPU hours will be in great demand from a number of research teams pre-approved by NSF. Results from the simulations and consequent unprecedented massive database are expected to draw interest from the turbulence community worldwide and be of interest in many problems of societal concern, including the efficiency of combustion devices and the dispersion of pollutants in the environment. To maximize the scientific impact of this work the challenges of making large datasets better maintained and more readily accessible will be be addressed via services provided by major supercomputer centers. This work will also provide a unique training opportunity for a graduate student of high caliber, echoing the success of prior joint mentoring activities by both investigators.