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.

Parallel streaming between heterogeneous HPC resources for real-time analysis

In order to reduce processing time and eliminate the need to write and read large volumes of data at once, some researchers are performing data analysis concurrently with the HPC simulation producing the data. In this paper, written by a team of researchers from University of St. Thomas, Georgia Institute of Technology, Oak Ridge National Laboratory, and Argonne National Laboratory, the authors discuss the “in transit” method for concurrent simulation and analysis. The “in transit” method streams data from the computing resource to the analysis resource. The authors discuss the efficiencies of this method.

Authors: Thomas Marrinan, Greg Eisenhauer, Matthew Wolf, Joseph A. Insley, Silvio Rizzi and Michael E. Papka

Managing resources for extreme-scale HPC systems in the presence of failures

As the scale of HPC systems grows toward exascale, the number of failures occurring in each system grows in tandem. In this dissertation, written by a researcher from Colorado State University, the author proposes a framework for “intelligently characterizing and managing extreme-scale HPC system resources.” The author then explores how to better model the negative effects of system failures.

Author: Daniel Dauwe

Comparing performance between quantum and classical machine learning

In this paper, the authors – a team from Southern Methodist University and KPMG – present a performance comparison of ML algorithms run on quantum and classical computing systems, with the aim of exploring the potential benefits of applying quantum computing to machine learning. The authors conclude that in some cases, the quantum computers achieved higher accuracy than their classical counterparts.

Authors: Christopher L. Havenstein, Damarcus T. Thomas and Swami Chandrasekaran.

Comparative benchmarking of HPC systems for global systems science

In this paper, a team of German researchers set out to support global systems science applications – applications that aim to leverage HPC power to answer “extremely complex societal and scientific problems,” developed under the European Commission’s Centre of Excellence for Global Systems Science (CoeGSS). Specifically, they investigate which HPC architectures and systems are best-suited for running GSS applications. They define their challenges, conduct tests, and make recommendations for a final benchmarking tool.

Authors: Damian Kaliszan, Steffen Fürst, Michael Gienger, Sergiy Gogolenko, Norbert Meyer and Sebastian Petruczynik.

Improving support of MPI+OpenMP applications

Hybrid applications are seen as a key option for achieving exascale computing — however, these options have typically been developed by separate communities, muddling the process of deploying the applications. In this paper, written by two researchers from Oak Ridge National Laboratory, the authors propose a “design for capabilities that would enable the precise definition and deployment of application layouts on compute node[.]” They outline a new notation scheme, a new runtime library for MPI and OpenMP runtimes, and a set of new components for the MPI runtime.

Authors: Geoffroy R. Vallée and David Bernholdt

Achieving energy efficiency in HPC datacenters

Energy efficiency, these researchers say, is an “evergreen problem” for HPC. In this paper, the authors (a team from Pakistan and Malaysia) survey the “issues, challenges […] and solutions” surrounding energy efficiency in HPC systems from 2010-2016. From this background, the authors propose a seven-pillar framework for energy efficiency in HPC systems and datacenters.

Authors: Sardar Mehboob Hussain, Abdul Wahid, Munam Ali Shah, Adnan Akhunzada, Faheem Khan, Noor ul Amin, Saba Arshad and Ihsan Ali.

Using high-performance algorithms for counting collisions

Counting collisions or interactions is a common use of computational power – for scientific simulations, computer graphics and so on. In this paper, the authors – a team from the University of Sao Paulo – focus on how interaction calculation can be conducted more efficiently. They assess two algorithms: a sequential algorithm with high computational intensity and a parallel algorithm with more efficient GPU utilization. They find performance increases under both algorithms and report the results.

Authors: Matheus Henrique Junqueira Saldanha and Paulo Sergio Lopes de Souza.

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.