GPP: Simplifying the implementations of parallel geophysical algorithms on large-scale clusters

PreviousNext No Access2016 Workshop: Workshop High Performance Computing, Beijing, China, 14-16 November 2016GPP: Simplifying the implementations of parallel geophysical algorithms on large-scale clustersAuthors: Changhai Zhao*Jiamin WenGuoan LuoJiguo DuMinqiang ShangZengbo WangChao LiChanghai Zhao*Geophysical Technique Research Center of BGP, CNPCSearch for more papers by this author, Jiamin WenGeophysical Technique Research Center of BGP, CNPCSearch for more papers by this author, Guoan LuoGeophysical Technique Research Center of BGP, CNPCSearch for more papers by this author, Jiguo DuGeophysical Technique Research Center of BGP, CNPCSearch for more papers by this author, Minqiang ShangGeophysical Technique Research Center of BGP, CNPCSearch for more papers by this author, Zengbo WangGeophysical Technique Research Center of BGP, CNPCSearch for more papers by this author, and Chao LiBeihang UniversitySearch for more papers by this authorhttps://doi.org/10.1190/hpc2016-015 SectionsAboutPDF/ePub ToolsAdd to favoritesDownload CitationsTrack CitationsPermissions ShareFacebookTwitterLinked InRedditEmail Abstract GPP is a parallel programming model for building highly scalable parallel applications to process large scale seismic data on large clusters that typically contain hundreds or thousands of computing nodes. GPP is composed of two key components: 1) a set of elegant and clean asynchronous communication primitives; 2) an event-driven fault tolerance mechanism. The communication primitives contain three basic one-sided asynchronous communication interfaces: Put, Get, and PutGet, based on which, the complex computations of parallel seismic applications can be expressed in a elegant and clean way. GPP is a better choice than MPI to implement complex parallel algorithms. GPP hires an event-driven fault tolerance mechanism to build a nonstop parallel program. GPP runtime system converts each detected process failure into an event containing detailed failure information of the execution context, and then schedules the event up to application layer by executing user-directed event handlers to drive the program recover from fault. Our implementation of GPP runs on a large cluster of commodity machines and is highly scalable. All the key modules in GeoEast have been re-implemented using GPP over the past six years. Keywords: algorithms, parallel, programmingPermalink: https://doi.org/10.1190/hpc2016-015FiguresReferencesRelatedDetails 2016 Workshop: Workshop High Performance Computing, Beijing, China, 14-16 November 2016ISSN (online):2159-6832Copyright: 2016 Pages: 63 publication data© 2016 Published in electronic format with permission by the Society of Exploration GeophysicistsPublisher:Society of Exploration Geophysicists HistoryPublished Online: 16 Nov 2016 CITATION INFORMATION Changhai Zhao*, Jiamin Wen, Guoan Luo, Jiguo Du, Minqiang Shang, Zengbo Wang, and Chao Li, (2016), "GPP: Simplifying the implementations of parallel geophysical algorithms on large-scale clusters," SEG Global Meeting Abstracts : 32-33. https://doi.org/10.1190/hpc2016-015 Plain-Language Summary KeywordsalgorithmsparallelprogrammingPDF DownloadLoading ...