NCSI: High-order Statistics of Turbulent Boundary Layers via GPU-oriented Programming

Status	Completed
Mentor Name	Guillermo Araya
Mentor's XSEDE Affiliation	Research Allocation
Mentor Has Been in XSEDE Community	4-5 years
Project Title	High-order Statistics of Turbulent Boundary Layers via GPU-oriented Programming
Summary	Turbulence is a complex multi-scale phenomenon characterized by chaos and randomness. However, turbulent flows are the rule, not the exception. Thus, a deep understanding of the transport phenomena driven by turbulence would permit the development of innovative flow/heat control tools for drag reduction, heat transfer enhancement, aeroacoustic noise control and mixing enhancement. In the last few decades, the discipline of thermal-fluid sciences has been reliant to High-Performance Computing (HPC) as a means of predicting flow behavior and understanding the thin zone around a solid immersed in a viscous fluid flow, the so called âboundary layerâ. Moreover, turbulent boundary layers that evolve along the flow direction are ubiquitous. Computationally speaking, this type of boundary layer (i.e., spatially-developing boundary layer) poses an enormous challenge, due to the need for accurate and time dependent in flow turbulence information. Direct Numerical Simulation (DNS) is a numerical tool that resolves all turbulence scales; thus, it aims to provide high spatial/temporal flow data. In this project, we will utilize a large dataset of direct simulations of turbulent boundary layers in order to compute high order statistics of velocity, pressure and temperature fluctuations, such as two-point correlations and power spectra. The major objectives of the proposed study are three-fold: (i) to develop an efficient GPU-based C++ code as a postprocessing tool, (ii) to understand the effect of compressibility on the boundary layer structure, (iii) to evaluate the analogy between the velocity and thermal field.
Job Description	The intern position is at the High Performance Computing and Visualization Lab (HPCVL) www.uprm.edu/hpcvl/ at the Dept. of Mechanical Engineering in the U. of Puerto Rico-Mayaguez (UPRM). The undergraduate student intern will work on a post-processing code in the area of C++ platform programming with GPU capabilities (CUDA) in order to alleviate computing effort as well as with already developed open-source flow solvers. The main goals/tasks of this internship can be summarized as follows: - Develop a CUDA/C++ code for managing/reading a large database of Direct Numerical Simulation (DNS) related to supersonic/hypersonic turbulent boundary layers. - Write routines for two-point correlation and power spectra computation.
Computational Resources	The DNS database is distributed between Ranch (TACC) and Blue Waters (NCSA). These simulations were performed in Comet (SDSC) and Stampede2 (TACC) under XSEDE computational allocation #TG-CTS170006. The intern will be set as a user of this account in order to get access to the TACC computational resources. In addition, the intern will be able to use a Linux workstation with GPU-capabilities located at the HPCVLab. This intern position is for 15 weeks at 8 hours per week.
Contribution to Community
Position Type	Intern
Training Plan	The intern will initially receive training on fluid mechanics, turbulence, data management, linux programming, data transfer, C++ and Fortran at the HPCVL lab by the PI and other skilled graduate students. The purpose of this position is to encourage students from underrepresented communities to pursue STEM careers at a graduate level.
Student Prerequisites/Conditions/Qualifications	This intern-level position is opened for undergraduate students at the University of Puerto Rico-Mayaguez. Candidates should have basic knowledge on: CUDA, Fortran, C++ programming, Linux and fluid mechanics. In addition, students should demonstrate independence to perform research and willingness to learn GPU computing. A skilled intern has already been identified.
Duration	Semester
Start Date	08/15/2019
End Date	12/15/2019