High performance computing (HPC) is undergoing a dramatic change in computing architectures. Nextgeneration HPC systems are being based primarily on many-core processing units and general purpose graphics processing units (GPUs). A computing node on a next-generation system can be, and in practice is, heterogeneous in nature, involving multiple memory spaces and multiple execution spaces. This presents a challenge for the development of application codes that wish to compute at the extreme scales afforded by these next-generation HPC technologies and systems - the best parallel programming model for one system is not necessarily the best parallel programming model for another. This inevitably raises the following question: how does an application code achieve high performance on disparate computing architectures without having entirely different, or at least significantly different, code paths, one for each architecture? This question has given rise to the term ‘performance portability’, a notion concerned with porting application code performance from architecture to architecture using a single code base. In this paper, we present the work being done at Sandia National Labs to develop a performance portable compressible CFD code that is targeting the ‘leadership’ class supercomputers the National Nuclear Security Administration (NNSA) is acquiring over the course of the next decade.
High performance computing (HPC) is undergoing a dramatic change in computing architectures. Nextgeneration HPC systems are being based primarily on many-core processing units and general purpose graphics processing units (GPUs). A computing node on a next-generation system can be, and in practice is, heterogeneous in nature, involving multiple memory spaces and multiple execution spaces. This presents a challenge for the development of application codes that wish to compute at the extreme scales afforded by these next-generation HPC technologies and systems - the best parallel programming model for one system is not necessarily the best parallel programming model for another. This inevitably raises the following question: how does an application code achieve high performance on disparate computing architectures without having entirely different, or at least significantly different, code paths, one for each architecture? This question has given rise to the term ‘performance portability’, a notion concerned with porting application code performance from architecture to architecture using a single code base. In this paper, we present the work being done at Sandia National Labs to develop a performance portable compressible CFD code that is targeting the ‘leadership’ class supercomputers the National Nuclear Security Administration (NNSA) is acquiring over the course of the next decade.
Juliano, Thomas J.; Kimmel, Roger L.; Willems, Sebastian; Gulhan, Ali; Wagnild, Ross M.
The HIFiRE-1 is a 7-degree half-angle circular cone with a 2.5-mm nose radius. A successful HIFiRE-1 flight experiment was carried out in March 2010. Due to an anomaly in the exoatmospheric pointing maneuver, the reentry angle of attack was higher than anticipated (5-15 degrees instead of near zero). A test campaign in the H2K hypersonic wind tunnel at DLR Cologne gathered high-frequency pressure fluctuation data and global heat flux via infrared (IR) thermography at the high angles of attack and Reynolds numbers encountered in the as-flown trajectory. This paper presents analysis of data collected at 0° angle of attack at freestream Reynolds numbers from 5.7 to 10.7.106 /m for 1.6-and 2.5-mm-radius nosetips. The transition onset and end locations derived from IR thermography coincide well with the earliest and largest amplification of pressure fluctuations identified by the fast-response surface-mounted pressure transducers. Stability analysis of the astested conditions was done with the Stability and Transition of Boundary Layers (STABL) software suite. An N-factor of 5.5 correlates well with transition location for the 1.6-mmradius nosetip. For the blunter nosetip, N ≈ 5.2 at transition. The peak pressure-fluctuation frequencies predicted by STABL agree within 8% of those measured.
Results for the stability analysis are as follows: maximum N factor trends agree well with previous data; transition N factor difference between Case 2 and Case 3 disagrees with previous data. Requires another look; predicts disturbance frequencies that agree with experiments and VESTA computations; and predicts larger N factors than VESTA
Boundary-layer transition and stability data were obtained at Mach 10 in the Arnold Engineering Development Complex (AEDC) Hypervelocity Wind Tunnel 9 on a 1.5-m long, 7-deg cone at unit Reynolds numbers between 1.8 and 31 million per meter. A total of 24 runs were performed at angles-of-attack between 0 and 10-deg on sharp and blunted cones with nose radii between 5.1 and 50.8-mm. The transition location was determined with coaxial thermocouples and temperature sensitive paint while stability measurements were obtained using high-frequency response pressure sensors. Mean flow and boundary layer-stability computations were also conducted and compared with the experiment. The effect of angle-of-attack and bluntness on the transition location displays similar trends compared to historical hypersonic wind tunnel data at similar Mach and Reynolds numbers. The N factor at start of transition on sharp cones increases with unit Reynolds number. Values between 4 and 7 were observed. The N factor at start of transition significantly decreases as bluntness increases and is successfully correlated with the ratio of transition location to entropy layer swallowing length. Good agreement between the computed and measured spatial amplification rates and most amplified 2nd mode frequencies are obtained for sharp and moderately blunted cones. For large bluntness, where the ratio of transition to entropy swallowing length is below 0.1, 2nd mode waves were not observed before the start of transition on the frustum.
The Quantum-Kinetic (Q-K) chemical reaction model is implemented in a Navier-Stokes solver, US3D, and tested on the Bow Shock UltraViolet flight experiments. The chemical reaction rates predicted by the Q-K model are compared to a commonly used Park model for flows in thermal non-equilibrium. The results show that in thermal equilibrium the reaction rates between these two models are comparable. The Q-K model predicts greater rates for some chemical reactions and lesser rates for other reactions in an five species air chemistry model. In thermal non-equilibrium, the Q-K model maintains comparable rates near thermal equilibrium, while avoiding issues of strong thermal non-equilibrium seen in the Park model. The application of the Q-K model to the Bow Shock UltraViolet flight experiments show that the model remains consistent with previous Navier-Stokes and DSMC computations over altitudes ranging from 53:5 km up to 87:5 km despite the enforcement of translational-rotational equilibrium. The commonly used Park model was unable to match this performance.
High-frequency pressure sensors were used in conjunction with a high-speed schlieren system to study the growth and breakdown of boundary-layer disturbances into turbulent spots on a 7° cone in the Sandia Hypersonic Wind Tunnel. At Mach 5, intermittent low-frequency disturbances were observed in the schlieren videos. High-frequency secondmode wave packets would develop within these low-frequency disturbances and break down into isolated turbulent spots surrounded by an otherwise smooth, laminar boundary layer. Spanwise pressure measurements showed that these packets have a narrow spanwise extent before they break down. The resulting turbulent fluctuations still had a streaky structure reminiscent of the wave packets. At Mach 8, the boundary layer was dominated by secondmode instabilities that extended much further in the spanwise direction before breaking down into regions of turbulence. The amplitude of the turbulent pressure fluctuations was much lower than those within the second-mode waves. These turbulent patches were surrounded by waves as opposed to the smooth laminar flow seen at Mach 5. At Mach 14, second-mode instability wave packets were also observed. Theses waves had a much lower frequency and larger spanwise extent compared to lower Mach numbers. Only low freestream Reynolds numbers could be obtained, so these waves did not break down into turbulence.