In this bimonthly feature, HPCwire highlights newly published research in the high-performance computing community and related domains. From parallel programming to exascale to quantum computing, the details are here.
Exascale modeling of plasma particle accelerators
Particle accelerators provide benefits to a wide range of fields ranging from medicine to energy and environmental matters – however, they are both large and expensive. This paper, written by a team from Berkeley Lab, Lawrence Livermore National Laboratory (LLNL) and the Stanford Linear Accelerator Center (SLAC), discusses their development of a new plasma acceleration simulation tool for use on HPC. Specifically, they intend the tool to run efficiently at scale on exascale supercomputers.
Authors: Maxence Thévenet, Jean-Luc Vay, Ann Almgren, Diana Amorim, John Bell, Axel Huebl, Revathi Jambunathan, Rémi Lehe, Andrew Myers, Jaehong Park, et al.
Scheduling jobs on HPC systems with simultaneous fair-share
Usually, HPC job schedulers schedule based on a linear sum of weighted terms – but that can make scheduling difficult when a system has a wide range of job types. This paper, written by Craig Steffen of the National Center for Supercomputing Applications (NCSA), proposes a new scheme for HPC job scheduling calls “simultaneous fair-share” (SFS). SFS works by “considering the jobs already committed to run in a given time slice and adjusting which jobs are selected to run accordingly.”
Author: Craig P. Steffen
Improving the Supercomputing Institute
Los Alamos National Laboratory (LANL) has been hosting an internship program called “the Supercomputing Institute,” providing a basis in cluster computing for both undergraduate and graduate students. In this paper, two researchers from LANL outline the ten-week program and describe how the program has changed (and improved) over time, including how it has become an important recruitment tool for LANL.
Authors: J. Lowell Wofford and Cory Lueninghoener
Driving brain science with big data and HPC
“Big data and HPC,” these authors (a team from four institutes in China) write, “play essential roles in studying brain function, brain diseases, and large-scale brain models or connectomes.” In this paper, they review the intersections of brain science, big data and HPC, highlighting the roles of deep learning and improved performance. They conclude that big data and HPC will continue to play crucial roles in the field and provide new neuromorphic insights.
Authors: Shanyu Chen, Zhipeng He, Xinyin Han, Xiaoyu He, Ruilin Li, Haidong Zhu, Dan Zhao, Chuangchuang Dai, Yu Zhang, Zhonghua Lu, et al.
Enabling machine learning-ready HPC ensembles
Increasingly, researchers are combating computational complexity with machine learning techniques for analyzing large-scale ensemble data. In this paper, a team from LLNL presents “Merlin,” a workflow framework that enables large, ML-friendly ensembles of scientific HPC simulations and which was deployed on the Sierra supercomputer. “By augmenting traditional HPC with distributed compute technologies,” they write, “Merlin aims to lower the barrier for scientific subject matter experts to incorporate ML into their analysis.”
Authors: J. Luc Peterson, Rushil Anirudh, Kevin Athey, Benjamin Bay, Peer-Timo Bremer, Vic Castillo, Francesco Di Natale, David Fox, Jim A. Gaffney, David Hysom, et al.
Using supercomputer technologies for solving computational physics problems
In this paper, a team of mathematical physicists from the Institute of Computational Mathematics and Mathematical Geophysics in Russia discuss their use of supercomputer technologies to solve compute-intensive mathematical physics problems. They describe in detail the process of initial collaboration, algorithm and technology selection, and energy efficiency evaluations when tackling a new problem on a supercomputer. They propose improving the process in several ways.
Authors: Boris Glinsky, Yury Zagorulko, Igor Kulikov and Anna Sapetina.
Introducing a new fault analysis tool for HPC researchers
Reliability issues are expected to increase as systems grow in size and complexity. These authors, a team from the University of Washington Seattle, Clemson University and the University of Illinois at Urbana-Champaign, introduce “FaultSight,” a fault injection analysis tool that the authors claim can efficiently assist in analyzing HPC application reliability and resilience scheme effectiveness.
Authors: Einar Horn, Dakota Fulp, Jon C. Calhoun and Luke N. Olson.
Do you know about research that should be included in next month’s list? If so, send us an email at [email protected]. We look forward to hearing from you.