Visualization projects by HLRS at SC17
Computer visualizations have for years enabled researchers to demonstrate scientific and industrial achievements in ways that are more eye-catching and understandable than pure bits and bytes. Visualization is one of the key components of high-performance computing (HPC), serving as an essential tool in helping researchers better understand their simulations, ultimately leading to better products reaching the market faster. “Visualization is only one method for analyzing HPC simulations, but plays an indispensable role in moving forward cutting-edge research and development in industry,” said Uwe Wössner, Head of the HLRS Visualization Department.
At the HLRS booth at SC17, for example, Wössner and colleagues demonstrated a use case commissioned by Dräger, a company that develops medical technology. Using augmented reality technology and with the help of airflow simulations running on HLRS resources, Dräger optimized product development of medical instruments such as respirators and operating lights, which can affect the airflow in surgery rooms. More precisely, the company aims to design instruments in such a way as to reduce the risk of return flows from unclean areas of the surgery room, thus reducing the risk of bacterial infections in patients’ wounds.
In Denver HLRS also displayed a remote visualization of large-scale computational fluid dynamics (CFD) data located in Stuttgart. Using Vistle, a software package developed at HLRS that integrates simulations on supercomputers, post-processing, and parallel, interactive visualization, researchers simulated complex flow properties inside a water turbine on HPC resources in Stuttgart and presented the visualized simulation data in a 3D virtual reality environment on the SC17 exhibition floor. In addition, visitors could wear a virtual reality headset and take a virtual test ride on the MULTI Elevator System, a new concept from thyssenkrupp Elevator AG in which cabins move through shafts horizontally as well as vertically. By using HLRS resources for both mechanical and airflow simulations, the company increased comfort and safety for passengers using the new elevator system, ultimately helping facilitate the development process. Read more about HLRS's contribution to the MULTI project here.
With the exponential growth of data and the challenges of timely, goal-oriented data analysis, and effective storage and transfer capabilities that go with it, Big Data and High Performance Data Analytics (HPDA) were again hot topics at SC. “Before the HPC community decides how to use Big Data for HPC applications,” Bastian Koller, managing director of HLRS, points out, “there must be a common understanding of what Big Data means and how it can be differentiated from machine learning and artificial intelligence. Otherwise it will simply remain a marketing buzzword and not live up to its potential.”
HLRS managing director Bastian Koller discusses the convergence
of HPC and data analytics at the NEC booth. (Photo: HLRS)
At SC17, Koller participated in the panel "HPC and Big Data Convergence" alongside other representatives of industry and science. He also spoke on possibilities of ‘Maximizing Synergies of HPC and HPDA’ at the NEC booth. “Instead of trying to produce a one-for-all IT tool in hardware and software that solves all potential problems,” Koller stated, “it is time to show how we can already use today’s technology to create success stories that combine HPC and Big Data. This will foster further uptake and thus further investments in the necessary technologies.”
At HLRS, these synergies are currently being investigated in the national and international research projects CATALYST—which is researching possible uses of HPDA in science and industry—and EOPEN, which aims to develop an open platform for unified access to and analysis of Earth observation data.
Hazel Hen stable on Top500, HPCG, and HPGMG benchmarks
Twice a year the HPC community eagerly awaits the benchmarking results of Top500’s LINPACK and High Performance Geometric Multigrid (HPGMG). The latest Top500 ranking of the world's fastest HPC systems was announced during SC17, with the High-Performance Computing Center Stuttgart's Hazel Hen ranked number 19. With its maximal LINPACK performance of 5,640 Tflops (LINPACK tests each HPC system on solving a dense set of linear equations) Hazel Hen held its position as the fastest supercomputer in Germany.
In the High Performance Conjugate Gradients (HPCG) Benchmark, Hazel Hen is currently ranked number 14 and remains in third place in Europe. HPCG is a complete, standalone code that measures the performance of basic operations. Its goal is to complement the LINPACK benchmark with a less dense set of equations.
For the recently published HPGMG benchmark, Hazel Hen is ranked number 4 in the world. As compared with the Top500 LINPACK benchmark, HPGMG does not measure how fast linear equations are solved, but how state-of-the-art multigrid methods perform while solving real-world computational problems in scientific and industrial fields such as engineering. This makes the HPGMG benchmark highly relevant for HPC users, providing an impression of how their real-life operations will perform on a specific system.
On the HPGMG benchmark Hazel Hen outperformed systems like Piz Daint (Swiss Center for Supercomputing, ranked number 3 in the Top500) and Titan (Oak Ridge National Laboratory, ranked number 5 in the Top500). “Since Hazel Hen was first ranked two years ago we can look back on a very satisfying history of results for all relevant benchmarks,” reminisces Michael Resch, HLRS director. “For us, it is most important to support our users with a reliable and efficient HPC infrastructure, which is achieved through a combination of hardware, software, and technical skills.”