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.

Optimizing visualization performance on power-constrained supercomputers

Power consumption is a major hurdle on the road the exascale. In this dissertation, Stephanie Labasan of the University of Oregon focuses on power consumption by visualization and analysis applications, which tend to be more data-intensive than traditional HPC applications. She examines power/performance tradeoffs for popular algorithms under different configurations, demonstrating that additional performance can  be gained by redistributing power based on predicted performance.

Author: Stephanie Alyssa Labasan

Enabling HPC-as-a-Service using ‘HEAppE’

HPC-as-a-service allows users to access supercomputing power in a simple and intuitive way without purchasing their own hardware. In this paper, written by a team from Germany and the Czech Republic, the authors present their own in-house application framework for HPC-as-a-Service: “High-End Application Execution Middleware,” or “HEAppE Middleware.” The authors describe how their architecture “enables unified access to different HPC systems through simple object-oriented web-based APIs” and outline the results of several pilot use cases.

Authors: Vaclav Svaton, Jan Martinovic, Jan Krenek, Thomas Esch and Pavel Tomancak.

Designing an FPGA-based multi-chip module for HPC 

Current hardware technologies are largely not suited for the density and efficiency requirements of the exascale era. In this paper – which was supported by the EU’s Horizon 2020 programme – a team from Canada, France and Greece presents design, fabrication and characterization details for an “ExaNoDe” multi-chip module prototype. Using the prototype, the authors aim to help enable exascale computing through a highly integrated, heterogeneous design.

Authors: Yann Beilliard, Maxime Godard, Aggelos Ioannou, Astrinos Damianakis, Michael Ligerakis, Iakovos Mavroidis, Pierre-Yves Martinez, David Danovitch, Julien Sylvestre and Dominique Drouin.

Scheduling beyond CPUs for HPC

To bridge the gap between I/O and processing capabilities, many systems employ “burst buffers,” but job schedules remain primarily CPU-centric. In this paper – written by a team from the Illinois Institute of Technology, Argonne National Laboratory and Lawrence Berkeley National Laboratory – the authors present “BBSched,” a multi-resource scheduling scheme that schedules user jobs in a less CPU-centric manner. In test scenarios, the authors find that BBSched improved scheduling performance by up to 41% compared to existing methods.

Authors: Yuping Fan, Zhiling Lan, Paul Rich, William E. Allcock, Michael E. Papka, Brian Austin and David Paul.

Developing an HPC performance anomaly suite for reproducing performance variations

Performance variations routinely afflict modern HPC systems – sometimes, incurring over 100 percent variation under the same input conditions. In this paper, written by a team from Boston University and Sandia National Laboratories, the authors address the need for a standardized way to create performance variability-inducing synthetic anomalies. To do this, they introduce HPAS – an “HPC Performance Anomaly Suite” – which consists of anomaly generators for HPC systems. The authors describe several use cases, including performance anomaly diagnosis, evaluation of resource management policies and more.

Authors: Emre Ates, Yijia Zhang, Burak Aksar, Jim Brandt, vitus J. Leung, Manuel Egele and Ayse K. Coskun.

Performing large-scale urban traffic simulation with Scala and HPC

HPC systems enable large-scale simulations of systems, including traffic. In this paper, researchers from the AGH University of Science and Technology in Poland focus on the “scalable implementation of a traffic simulation system in an asynchronous and notably desynchronized way.” They describe the concept of the system as well as a series of experiments on clusters up to 1,000 nodes, emphasizing the ease of implementation for programmers.

Authors: Michal Janczykowski, Wojciech Turek, Maciej Malawski and Aleksanders Byrski.

Conducting high-performance overlay analysis of massive geographic polygons

Overlay analysis (which combines characteristics from several geographic datasets into one – is a common task in geographic computing) – and, with the emergence of high-resolution satellite technology, it is growing increasingly computationally intensive. In this paper, written by a team of researchers from China, the authors discuss the design and implementation of a parallel processing algorithm designed to allow for easier overlay analysis of complex polygonal geographic data in high-performance cloud computing.

Authors: Kang Zhao, Baoxuan Jin, Hong Fan, Weiwei Song, Sunyu Zhou and Yuanyi Jiang.

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.