In this bimonthly feature, HPCwire will highlight newly published research in the high-performance computing community and related domains. From exascale to quantum computing, the details are here. Check back on the first and third Mondays of each month for more!
Experimenting with runtime and energy tradeoffs in HPC
Energy efficiency is a core concern for HPC systems as they scale up, both for cost reasons and environmental reasons. In this paper, published in Progress in Industrial Ecology, the authors examine the tradeoffs between energy, memory, and runtime in a series of HPC algorithms. The authors show the influence of aggregate transmissions on the execution of parallel calculations and describe a better method for streamlining runtime, energy, and memory tradeoffs.
Authors: G. Uma Maheswari and S. Subha
Classifying faults in HPC systems online using ML
As HPC enters the exascale era, researchers expect excessive failure rates due to extensive parallelism – as often as every few minutes. These researchers – a team from Germany and Italy – set out to detect, classify, and correct such failures as they occur. They propose a method for HPC systems based on machine learning and designed for streaming data. The researchers say that they achieved “almost perfect” classification accuracy with low computational overhead.
Authors: Alessio Netti, Zeynep Kiziltan, Ozalp Babaoglu, Alina Sirbu, Andrea Bartolini, and Andrea Borghesi
Highlighting practical challenges for biomedical modeling using HPC
Computational biomedicine has the potential to drastically improve diagnosis and treatment by processing (highly varied and vast) patient characteristic datasets. In this paper, a team from the University of London examines the spectrum of computational requirements that are required to achieve those goals. They highlight experiences from two biomedical modeling scenarios (brain blood flow and small molecule drug selection) to pinpoint current limitations and suggest solutions.
Authors: David W. Wright, Robin A. Richardson, and Peter V. Coveney
Understanding communication problems in HPCG
Conjugate gradient algorithms are a significant element of HPC applications ranging from weather modeling to bioinformatics, allowing numerical solutions to complex systems. A team of researchers from the University of Warwick set out to characterize communication patterns in the High Performance Conjugate Gradient benchmark, focusing on the implications for the HPC system interconnect and exploring potential bottlenecks. The researchers conclude that there are patterns and features that may be useful in improving the performance of CG algorithms and the applications that depend on them.
Authors: Dean G. Chester, Steven A. Wright, and Stephen A. Jarvis
Understanding I/O behavior in HPC workflows
Scientific discovery increasingly relies on multi-phase workflows with up to millions of parallelizable tasks. In this paper, a team from the University of Reading claim that most workflow models ignore the impacts that these tasks have on storage resources. The researchers go on to address those impacts, presenting an approach to improve the I/O efficiency of tasks and introducing a conceptual framework for HPC datacenter deployments. They conclude by discussing future challenges and outlining how real-world workflows could be improved by their approach.
Authors: J. Lüttgau, S. Snyder, P. Carns, J.M. Wozniak, J. Kunkel, and T. Ludwig.
Examining how China fosters competition in the race to exascale
China, the U.S., Japan, and the EU are all participating in the race to an exascale supercomputer – but as this author points out, China is taking an unusual approach. In China, three competing teams raced to develop prototypes that were tested and given trial software runs. The author outlines the competitive Chinese strategy and how it impacts their timeline for exascale supercomputer production. He makes the case that China’s target date of 2020 is likely to slip.
Author: Dennis Normile
Learning lessons from the EU’s DISRUPT project
Much attention is being paid to modernization of industrial processes and manufacturing through HPC – a vision called “Industry 4.0.” In this paper, a team from the University of Manchester takes a deep dive into the an EU-funded project called DISRUPT, identifying key challenges in the relationship between Industry 4.0 and HPC technologies.
Authors: Rizos Sakellariou, Jorge Buenabad-Chávez, Evangelia Kavakli, Ilias Spais, Vasilios Tountopoulos
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.