Cheetah: A Framework for Scalable Hierarchical Collective Operations
Collective communication operations, used by many scientific applications, tend to limit overall parallel application performance and scalability. Computer systems are becoming more heterogeneous with increasing node and core-per-node counts. Also, a growing number of data-access mechanisms, of vary...
Gespeichert in:
Hauptverfasser: | , , , , , , |
---|---|
Format: | Tagungsbericht |
Sprache: | eng |
Schlagworte: | |
Online-Zugang: | Volltext bestellen |
Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
container_end_page | 83 |
---|---|
container_issue | |
container_start_page | 73 |
container_title | |
container_volume | |
creator | Graham, Richard Venkata, Manjunath Gorentla Ladd, Joshua Shamis, Pavel Rabinovitz, Ishai Filipov, Vasily Shainer, Gilad |
description | Collective communication operations, used by many scientific applications, tend to limit overall parallel application performance and scalability. Computer systems are becoming more heterogeneous with increasing node and core-per-node counts. Also, a growing number of data-access mechanisms, of varying characteristics, are supported within a single computer system. We describe a new hierarchical collective communication framework that takes advantage of hardware-specific data-access mechanisms. It is flexible, with run-time hierarchy specification, and sharing of collective communication primitives between collective algorithms. Data buffers are shared between levels in the hierarchy reducing collective communication management overhead. We have implemented several versions of the Message Passing Interface (MPI) collective operations, MPI Barrier() and MPI Bcast(), and run experiments using up to 49, 152 processes on a Cray XT5, and a small InfiniBand based cluster. At 49, 152 processes our barrier implementation outperforms the optimized native implementation by 75%. 32 Byte and one Mega-Byte broadcasts outperform it by 62% and 11%, respectively, with better scalability characteristics. Improvements relative to the default Open MPI implementation are much larger. |
doi_str_mv | 10.1109/CCGrid.2011.42 |
format | Conference Proceeding |
fullrecord | <record><control><sourceid>acm_6IE</sourceid><recordid>TN_cdi_ieee_primary_5948598</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><ieee_id>5948598</ieee_id><sourcerecordid>acm_books_10_1109_CCGrid_2011_42</sourcerecordid><originalsourceid>FETCH-LOGICAL-a160t-3db6ebae3359337198952f78471d2d6325ece60703809b755b29d550a85889d93</originalsourceid><addsrcrecordid>eNqNj0FLAzEQhSMiKnWvXvwBnnadZDJJ5ihBW6HgRc8haWbpoqWy24v_3pb15MnT4_E-HnxK3WrotAZ-iHE5DrUzoHVnzZlq2Afwjskikzv_069UM01DAeO8I098rS7jVuSQtzfqos-fkzS_uVDvz09vcdWuX5cv8XHdZu3g0GItTkoWRGJErzkwmd4H63U11aEh2YgDDxiAiycqhisR5EAhcGVcqLv5dxCR9DUOuzx-J2IbiMNxvZ_XvNmlst9_TElDOommWTSdRJM1R7L9H5nKOEiPP6_5TrY</addsrcrecordid><sourcetype>Publisher</sourcetype><iscdi>true</iscdi><recordtype>conference_proceeding</recordtype></control><display><type>conference_proceeding</type><title>Cheetah: A Framework for Scalable Hierarchical Collective Operations</title><source>IEEE Electronic Library (IEL) Conference Proceedings</source><creator>Graham, Richard ; Venkata, Manjunath Gorentla ; Ladd, Joshua ; Shamis, Pavel ; Rabinovitz, Ishai ; Filipov, Vasily ; Shainer, Gilad</creator><creatorcontrib>Graham, Richard ; Venkata, Manjunath Gorentla ; Ladd, Joshua ; Shamis, Pavel ; Rabinovitz, Ishai ; Filipov, Vasily ; Shainer, Gilad</creatorcontrib><description>Collective communication operations, used by many scientific applications, tend to limit overall parallel application performance and scalability. Computer systems are becoming more heterogeneous with increasing node and core-per-node counts. Also, a growing number of data-access mechanisms, of varying characteristics, are supported within a single computer system. We describe a new hierarchical collective communication framework that takes advantage of hardware-specific data-access mechanisms. It is flexible, with run-time hierarchy specification, and sharing of collective communication primitives between collective algorithms. Data buffers are shared between levels in the hierarchy reducing collective communication management overhead. We have implemented several versions of the Message Passing Interface (MPI) collective operations, MPI Barrier() and MPI Bcast(), and run experiments using up to 49, 152 processes on a Cray XT5, and a small InfiniBand based cluster. At 49, 152 processes our barrier implementation outperforms the optimized native implementation by 75%. 32 Byte and one Mega-Byte broadcasts outperform it by 62% and 11%, respectively, with better scalability characteristics. Improvements relative to the default Open MPI implementation are much larger.</description><identifier>ISBN: 9780769543956</identifier><identifier>ISBN: 0769543952</identifier><identifier>ISBN: 1457701294</identifier><identifier>ISBN: 9781457701290</identifier><identifier>EISBN: 9780769543956</identifier><identifier>EISBN: 0769543952</identifier><identifier>DOI: 10.1109/CCGrid.2011.42</identifier><language>eng</language><publisher>Washington, DC, USA: IEEE Computer Society</publisher><subject>Algorithm design and analysis ; Classification algorithms ; Clustering algorithms ; Collectives ; Computer systems organization -- Architectures -- Distributed architectures ; Computer systems organization -- Dependable and fault-tolerant systems and networks ; Framework ; General and reference -- Cross-computing tools and techniques -- Performance ; Hierarchy ; Memory management ; Network topology ; Networks -- Network performance evaluation ; Sockets ; Software and its engineering -- Software organization and properties -- Software system structures -- Distributed systems organizing principles ; Topology</subject><ispartof>2011 11th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing, 2011, p.73-83</ispartof><woscitedreferencessubscribed>false</woscitedreferencessubscribed></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><linktohtml>$$Uhttps://ieeexplore.ieee.org/document/5948598$$EHTML$$P50$$Gieee$$H</linktohtml><link.rule.ids>309,310,780,784,789,790,2058,27925,54920</link.rule.ids><linktorsrc>$$Uhttps://ieeexplore.ieee.org/document/5948598$$EView_record_in_IEEE$$FView_record_in_$$GIEEE</linktorsrc></links><search><creatorcontrib>Graham, Richard</creatorcontrib><creatorcontrib>Venkata, Manjunath Gorentla</creatorcontrib><creatorcontrib>Ladd, Joshua</creatorcontrib><creatorcontrib>Shamis, Pavel</creatorcontrib><creatorcontrib>Rabinovitz, Ishai</creatorcontrib><creatorcontrib>Filipov, Vasily</creatorcontrib><creatorcontrib>Shainer, Gilad</creatorcontrib><title>Cheetah: A Framework for Scalable Hierarchical Collective Operations</title><title>2011 11th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing</title><addtitle>ccgrid</addtitle><description>Collective communication operations, used by many scientific applications, tend to limit overall parallel application performance and scalability. Computer systems are becoming more heterogeneous with increasing node and core-per-node counts. Also, a growing number of data-access mechanisms, of varying characteristics, are supported within a single computer system. We describe a new hierarchical collective communication framework that takes advantage of hardware-specific data-access mechanisms. It is flexible, with run-time hierarchy specification, and sharing of collective communication primitives between collective algorithms. Data buffers are shared between levels in the hierarchy reducing collective communication management overhead. We have implemented several versions of the Message Passing Interface (MPI) collective operations, MPI Barrier() and MPI Bcast(), and run experiments using up to 49, 152 processes on a Cray XT5, and a small InfiniBand based cluster. At 49, 152 processes our barrier implementation outperforms the optimized native implementation by 75%. 32 Byte and one Mega-Byte broadcasts outperform it by 62% and 11%, respectively, with better scalability characteristics. Improvements relative to the default Open MPI implementation are much larger.</description><subject>Algorithm design and analysis</subject><subject>Classification algorithms</subject><subject>Clustering algorithms</subject><subject>Collectives</subject><subject>Computer systems organization -- Architectures -- Distributed architectures</subject><subject>Computer systems organization -- Dependable and fault-tolerant systems and networks</subject><subject>Framework</subject><subject>General and reference -- Cross-computing tools and techniques -- Performance</subject><subject>Hierarchy</subject><subject>Memory management</subject><subject>Network topology</subject><subject>Networks -- Network performance evaluation</subject><subject>Sockets</subject><subject>Software and its engineering -- Software organization and properties -- Software system structures -- Distributed systems organizing principles</subject><subject>Topology</subject><isbn>9780769543956</isbn><isbn>0769543952</isbn><isbn>1457701294</isbn><isbn>9781457701290</isbn><isbn>9780769543956</isbn><isbn>0769543952</isbn><fulltext>true</fulltext><rsrctype>conference_proceeding</rsrctype><creationdate>2011</creationdate><recordtype>conference_proceeding</recordtype><sourceid>6IE</sourceid><sourceid>RIE</sourceid><recordid>eNqNj0FLAzEQhSMiKnWvXvwBnnadZDJJ5ihBW6HgRc8haWbpoqWy24v_3pb15MnT4_E-HnxK3WrotAZ-iHE5DrUzoHVnzZlq2Afwjskikzv_069UM01DAeO8I098rS7jVuSQtzfqos-fkzS_uVDvz09vcdWuX5cv8XHdZu3g0GItTkoWRGJErzkwmd4H63U11aEh2YgDDxiAiycqhisR5EAhcGVcqLv5dxCR9DUOuzx-J2IbiMNxvZ_XvNmlst9_TElDOommWTSdRJM1R7L9H5nKOEiPP6_5TrY</recordid><startdate>20110523</startdate><enddate>20110523</enddate><creator>Graham, Richard</creator><creator>Venkata, Manjunath Gorentla</creator><creator>Ladd, Joshua</creator><creator>Shamis, Pavel</creator><creator>Rabinovitz, Ishai</creator><creator>Filipov, Vasily</creator><creator>Shainer, Gilad</creator><general>IEEE Computer Society</general><general>IEEE</general><scope>6IE</scope><scope>6IL</scope><scope>CBEJK</scope><scope>RIE</scope><scope>RIL</scope></search><sort><creationdate>20110523</creationdate><title>Cheetah</title><author>Graham, Richard ; Venkata, Manjunath Gorentla ; Ladd, Joshua ; Shamis, Pavel ; Rabinovitz, Ishai ; Filipov, Vasily ; Shainer, Gilad</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-a160t-3db6ebae3359337198952f78471d2d6325ece60703809b755b29d550a85889d93</frbrgroupid><rsrctype>conference_proceedings</rsrctype><prefilter>conference_proceedings</prefilter><language>eng</language><creationdate>2011</creationdate><topic>Algorithm design and analysis</topic><topic>Classification algorithms</topic><topic>Clustering algorithms</topic><topic>Collectives</topic><topic>Computer systems organization -- Architectures -- Distributed architectures</topic><topic>Computer systems organization -- Dependable and fault-tolerant systems and networks</topic><topic>Framework</topic><topic>General and reference -- Cross-computing tools and techniques -- Performance</topic><topic>Hierarchy</topic><topic>Memory management</topic><topic>Network topology</topic><topic>Networks -- Network performance evaluation</topic><topic>Sockets</topic><topic>Software and its engineering -- Software organization and properties -- Software system structures -- Distributed systems organizing principles</topic><topic>Topology</topic><toplevel>online_resources</toplevel><creatorcontrib>Graham, Richard</creatorcontrib><creatorcontrib>Venkata, Manjunath Gorentla</creatorcontrib><creatorcontrib>Ladd, Joshua</creatorcontrib><creatorcontrib>Shamis, Pavel</creatorcontrib><creatorcontrib>Rabinovitz, Ishai</creatorcontrib><creatorcontrib>Filipov, Vasily</creatorcontrib><creatorcontrib>Shainer, Gilad</creatorcontrib><collection>IEEE Electronic Library (IEL) Conference Proceedings</collection><collection>IEEE Proceedings Order Plan All Online (POP All Online) 1998-present by volume</collection><collection>IEEE Xplore All Conference Proceedings</collection><collection>IEEE Electronic Library (IEL)</collection><collection>IEEE Proceedings Order Plans (POP All) 1998-Present</collection></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext_linktorsrc</fulltext></delivery><addata><au>Graham, Richard</au><au>Venkata, Manjunath Gorentla</au><au>Ladd, Joshua</au><au>Shamis, Pavel</au><au>Rabinovitz, Ishai</au><au>Filipov, Vasily</au><au>Shainer, Gilad</au><format>book</format><genre>proceeding</genre><ristype>CONF</ristype><atitle>Cheetah: A Framework for Scalable Hierarchical Collective Operations</atitle><btitle>2011 11th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing</btitle><stitle>ccgrid</stitle><date>2011-05-23</date><risdate>2011</risdate><spage>73</spage><epage>83</epage><pages>73-83</pages><isbn>9780769543956</isbn><isbn>0769543952</isbn><isbn>1457701294</isbn><isbn>9781457701290</isbn><eisbn>9780769543956</eisbn><eisbn>0769543952</eisbn><abstract>Collective communication operations, used by many scientific applications, tend to limit overall parallel application performance and scalability. Computer systems are becoming more heterogeneous with increasing node and core-per-node counts. Also, a growing number of data-access mechanisms, of varying characteristics, are supported within a single computer system. We describe a new hierarchical collective communication framework that takes advantage of hardware-specific data-access mechanisms. It is flexible, with run-time hierarchy specification, and sharing of collective communication primitives between collective algorithms. Data buffers are shared between levels in the hierarchy reducing collective communication management overhead. We have implemented several versions of the Message Passing Interface (MPI) collective operations, MPI Barrier() and MPI Bcast(), and run experiments using up to 49, 152 processes on a Cray XT5, and a small InfiniBand based cluster. At 49, 152 processes our barrier implementation outperforms the optimized native implementation by 75%. 32 Byte and one Mega-Byte broadcasts outperform it by 62% and 11%, respectively, with better scalability characteristics. Improvements relative to the default Open MPI implementation are much larger.</abstract><cop>Washington, DC, USA</cop><pub>IEEE Computer Society</pub><doi>10.1109/CCGrid.2011.42</doi><tpages>11</tpages></addata></record> |
fulltext | fulltext_linktorsrc |
identifier | ISBN: 9780769543956 |
ispartof | 2011 11th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing, 2011, p.73-83 |
issn | |
language | eng |
recordid | cdi_ieee_primary_5948598 |
source | IEEE Electronic Library (IEL) Conference Proceedings |
subjects | Algorithm design and analysis Classification algorithms Clustering algorithms Collectives Computer systems organization -- Architectures -- Distributed architectures Computer systems organization -- Dependable and fault-tolerant systems and networks Framework General and reference -- Cross-computing tools and techniques -- Performance Hierarchy Memory management Network topology Networks -- Network performance evaluation Sockets Software and its engineering -- Software organization and properties -- Software system structures -- Distributed systems organizing principles Topology |
title | Cheetah: A Framework for Scalable Hierarchical Collective Operations |
url | https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2024-12-24T01%3A46%3A13IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-acm_6IE&rft_val_fmt=info:ofi/fmt:kev:mtx:book&rft.genre=proceeding&rft.atitle=Cheetah:%20A%20Framework%20for%20Scalable%20Hierarchical%20Collective%20Operations&rft.btitle=2011%2011th%20IEEE/ACM%20International%20Symposium%20on%20Cluster,%20Cloud%20and%20Grid%20Computing&rft.au=Graham,%20Richard&rft.date=2011-05-23&rft.spage=73&rft.epage=83&rft.pages=73-83&rft.isbn=9780769543956&rft.isbn_list=0769543952&rft.isbn_list=1457701294&rft.isbn_list=9781457701290&rft_id=info:doi/10.1109/CCGrid.2011.42&rft_dat=%3Cacm_6IE%3Eacm_books_10_1109_CCGrid_2011_42%3C/acm_6IE%3E%3Curl%3E%3C/url%3E&rft.eisbn=9780769543956&rft.eisbn_list=0769543952&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_id=info:pmid/&rft_ieee_id=5948598&rfr_iscdi=true |