Logo des Repositoriums
 

Energy Efficiency of a Low Power Hardware Cluster for High Performance Computing

dc.contributor.authorGörtz, Michael Dominik
dc.contributor.authorKühn, Roland
dc.contributor.authorZietek, Oliver
dc.contributor.authorBernhard, Roman
dc.contributor.authorBulinski, Michael
dc.contributor.authorDuman, Dennis
dc.contributor.authorFreisen, Benedikt
dc.contributor.authorJentsch, Uwe
dc.contributor.authorKlöppner, Tobias
dc.contributor.authorPopovic, Dragana
dc.contributor.authorXu, Lili
dc.contributor.editorEibl, Maximilian
dc.contributor.editorGaedke, Martin
dc.date.accessioned2017-08-28T23:48:09Z
dc.date.available2017-08-28T23:48:09Z
dc.date.issued2017
dc.description.abstractHigh performance computing has become more and more limited by the hardware’s energy consumption, rendering it increasingly difficult to build even faster compute clusters, while modern low power hardware is making great improvements, regarding its performance. In order to overcome the limited energy density, we propose to use low power System on a Chip (SoC) devices, instead of high performance CPUs, exploiting the energy efficiency of modern low power processors. We evaluated our suggestion by building a low power cluster based on 40 single board computers with ARM Cortex-A53 quad-core SoCs and measuring its performance, energy consumption and efficiency using synthetic and application benchmarks with different workload types. Our tests demonstrated that our cluster could perform the given benchmarks, using up to 70% less energy than an Intel-based reference server system, which lead to an increase in efficiency of up to 425%. Our evaluation showed that modern low power processors have become a good alternative for high performance computing, with large workloads profiting from massive parallelization and that we can expect further improvements in this field, regarding the hardware’s performance and efficiency.en
dc.identifier.doi10.18420/in2017_256
dc.identifier.isbn978-3-88579-669-5
dc.identifier.pissn1617-5468
dc.language.isoen
dc.publisherGesellschaft für Informatik, Bonn
dc.relation.ispartofINFORMATIK 2017
dc.relation.ispartofseriesLecture Notes in Informatics (LNI) - Proceedings, Volume P-275
dc.subjectenergy efficiency
dc.subjectlow power hardware
dc.subjecthigh performance computing
dc.subjectARM processors
dc.titleEnergy Efficiency of a Low Power Hardware Cluster for High Performance Computingen
gi.citation.endPage2548
gi.citation.startPage2537
gi.conference.date25.-29. September 2017
gi.conference.locationChemnitz
gi.conference.sessiontitleStudierendenkonferenz Informatik 2017 (SKILL 2017)

Dateien

Originalbündel
1 - 1 von 1
Lade...
Vorschaubild
Name:
E1-13.pdf
Größe:
8.37 MB
Format:
Adobe Portable Document Format