Loading…
Fat-Trees Routing and Node Ordering Providing Contention Free Traffic for MPI Global Collectives
As the size of High Performance Computing clusters grows, the increasing probability of interconnect hot spots degrades the latency and effective bandwidth the network provides. This paper presents a solution to this scalability problem for real life constant bisectional-bandwidth fat-tree topologie...
Saved in:
Main Author: | |
---|---|
Format: | Conference Proceeding |
Language: | eng ; jpn |
Subjects: | |
Online Access: | Request full text |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
cited_by | |
---|---|
cites | |
container_end_page | 770 |
container_issue | |
container_start_page | 761 |
container_title | |
container_volume | |
creator | Zahavi, E. |
description | As the size of High Performance Computing clusters grows, the increasing probability of interconnect hot spots degrades the latency and effective bandwidth the network provides. This paper presents a solution to this scalability problem for real life constant bisectional-bandwidth fat-tree topologies. It is shown that maximal bandwidth and cut-through latency can be achieved for MPI global collective traffic. To form such congestion-free configuration, MPI programs should utilize collective communication, MPI-node-order should be topology aware, and the packets routing should match the MPI communication patterns. First, we show that MPI collectives can be classified into unidirectional and bidirectional shifts. Using this property, we propose a scheme for congestion-free routing of the global collectives in fully and partially populated fat trees running a single job. Simulation results of the proposed routing, MPI-node-order and communication patterns show a 40% throughput improvement over previously published results for all-to-all collectives. |
doi_str_mv | 10.1109/IPDPS.2011.219 |
format | conference_proceeding |
fullrecord | <record><control><sourceid>ieee_CHZPO</sourceid><recordid>TN_cdi_ieee_primary_6008918</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><ieee_id>6008918</ieee_id><sourcerecordid>6008918</sourcerecordid><originalsourceid>FETCH-LOGICAL-i156t-cc902e04a91f88570d31b51425c3cc2e0b463ad71b4c4d052801403555af54453</originalsourceid><addsrcrecordid>eNotT8tOwzAQNAIkUOmVCxf_QMqu4_XjiAotlQqNoJyL4zjIKCTICZX4e1JgLzO785CWsUuEGSLY61VxWzzPBCDOBNojNrXagFaWJGmtjn93VCiMlILwhJ0j5ZAJ0HTGpn3_DuMoZaWic_a6cEO2TSH0_Kn7GmL7xl1b8ceuCnyTqpAOlyJ1-1gd2Lxrh9AOsWv5YgzxbXJ1HT2vu8QfihVfNl3pmtHWNMEPcR_6C3Zau6YP03-csJfF3XZ-n603y9X8Zp1FJDVk3lsQAaSzWBtDGqocS8LxAZ97PyqlVLmrNJbSywpIGEAJORG5mqSkfMKu_npjCGH3meKHS987BWAsmvwH_eRWGA</addsrcrecordid><sourcetype>Publisher</sourcetype><iscdi>true</iscdi><recordtype>conference_proceeding</recordtype></control><display><type>conference_proceeding</type><title>Fat-Trees Routing and Node Ordering Providing Contention Free Traffic for MPI Global Collectives</title><source>IEEE Xplore All Conference Series</source><creator>Zahavi, E.</creator><creatorcontrib>Zahavi, E.</creatorcontrib><description>As the size of High Performance Computing clusters grows, the increasing probability of interconnect hot spots degrades the latency and effective bandwidth the network provides. This paper presents a solution to this scalability problem for real life constant bisectional-bandwidth fat-tree topologies. It is shown that maximal bandwidth and cut-through latency can be achieved for MPI global collective traffic. To form such congestion-free configuration, MPI programs should utilize collective communication, MPI-node-order should be topology aware, and the packets routing should match the MPI communication patterns. First, we show that MPI collectives can be classified into unidirectional and bidirectional shifts. Using this property, we propose a scheme for congestion-free routing of the global collectives in fully and partially populated fat trees running a single job. Simulation results of the proposed routing, MPI-node-order and communication patterns show a 40% throughput improvement over previously published results for all-to-all collectives.</description><identifier>ISSN: 1530-2075</identifier><identifier>ISBN: 9781612844251</identifier><identifier>ISBN: 1612844251</identifier><identifier>EISBN: 9780769545776</identifier><identifier>EISBN: 0769545777</identifier><identifier>DOI: 10.1109/IPDPS.2011.219</identifier><language>eng ; jpn</language><publisher>IEEE</publisher><subject>Bandwidth ; Clustering algorithms ; Degradation ; Indexes ; Routing ; Topology ; Traffic control</subject><ispartof>2011 IEEE International Symposium on Parallel and Distributed Processing Workshops and Phd Forum, 2011, p.761-770</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/6008918$$EHTML$$P50$$Gieee$$H</linktohtml><link.rule.ids>309,310,780,784,789,790,2058,27925,54555,54920,54932</link.rule.ids><linktorsrc>$$Uhttps://ieeexplore.ieee.org/document/6008918$$EView_record_in_IEEE$$FView_record_in_$$GIEEE</linktorsrc></links><search><creatorcontrib>Zahavi, E.</creatorcontrib><title>Fat-Trees Routing and Node Ordering Providing Contention Free Traffic for MPI Global Collectives</title><title>2011 IEEE International Symposium on Parallel and Distributed Processing Workshops and Phd Forum</title><addtitle>ipdpsw</addtitle><description>As the size of High Performance Computing clusters grows, the increasing probability of interconnect hot spots degrades the latency and effective bandwidth the network provides. This paper presents a solution to this scalability problem for real life constant bisectional-bandwidth fat-tree topologies. It is shown that maximal bandwidth and cut-through latency can be achieved for MPI global collective traffic. To form such congestion-free configuration, MPI programs should utilize collective communication, MPI-node-order should be topology aware, and the packets routing should match the MPI communication patterns. First, we show that MPI collectives can be classified into unidirectional and bidirectional shifts. Using this property, we propose a scheme for congestion-free routing of the global collectives in fully and partially populated fat trees running a single job. Simulation results of the proposed routing, MPI-node-order and communication patterns show a 40% throughput improvement over previously published results for all-to-all collectives.</description><subject>Bandwidth</subject><subject>Clustering algorithms</subject><subject>Degradation</subject><subject>Indexes</subject><subject>Routing</subject><subject>Topology</subject><subject>Traffic control</subject><issn>1530-2075</issn><isbn>9781612844251</isbn><isbn>1612844251</isbn><isbn>9780769545776</isbn><isbn>0769545777</isbn><fulltext>true</fulltext><rsrctype>conference_proceeding</rsrctype><creationdate>2011</creationdate><recordtype>conference_proceeding</recordtype><sourceid>6IE</sourceid><recordid>eNotT8tOwzAQNAIkUOmVCxf_QMqu4_XjiAotlQqNoJyL4zjIKCTICZX4e1JgLzO785CWsUuEGSLY61VxWzzPBCDOBNojNrXagFaWJGmtjn93VCiMlILwhJ0j5ZAJ0HTGpn3_DuMoZaWic_a6cEO2TSH0_Kn7GmL7xl1b8ceuCnyTqpAOlyJ1-1gd2Lxrh9AOsWv5YgzxbXJ1HT2vu8QfihVfNl3pmtHWNMEPcR_6C3Zau6YP03-csJfF3XZ-n603y9X8Zp1FJDVk3lsQAaSzWBtDGqocS8LxAZ97PyqlVLmrNJbSywpIGEAJORG5mqSkfMKu_npjCGH3meKHS987BWAsmvwH_eRWGA</recordid><startdate>201105</startdate><enddate>201105</enddate><creator>Zahavi, E.</creator><general>IEEE</general><scope>6IE</scope><scope>6IL</scope><scope>CBEJK</scope><scope>RIE</scope><scope>RIL</scope></search><sort><creationdate>201105</creationdate><title>Fat-Trees Routing and Node Ordering Providing Contention Free Traffic for MPI Global Collectives</title><author>Zahavi, E.</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-i156t-cc902e04a91f88570d31b51425c3cc2e0b463ad71b4c4d052801403555af54453</frbrgroupid><rsrctype>conference_proceedings</rsrctype><prefilter>conference_proceedings</prefilter><language>eng ; jpn</language><creationdate>2011</creationdate><topic>Bandwidth</topic><topic>Clustering algorithms</topic><topic>Degradation</topic><topic>Indexes</topic><topic>Routing</topic><topic>Topology</topic><topic>Traffic control</topic><toplevel>online_resources</toplevel><creatorcontrib>Zahavi, E.</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 Online</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>Zahavi, E.</au><format>book</format><genre>proceeding</genre><ristype>CONF</ristype><atitle>Fat-Trees Routing and Node Ordering Providing Contention Free Traffic for MPI Global Collectives</atitle><btitle>2011 IEEE International Symposium on Parallel and Distributed Processing Workshops and Phd Forum</btitle><stitle>ipdpsw</stitle><date>2011-05</date><risdate>2011</risdate><spage>761</spage><epage>770</epage><pages>761-770</pages><issn>1530-2075</issn><isbn>9781612844251</isbn><isbn>1612844251</isbn><eisbn>9780769545776</eisbn><eisbn>0769545777</eisbn><abstract>As the size of High Performance Computing clusters grows, the increasing probability of interconnect hot spots degrades the latency and effective bandwidth the network provides. This paper presents a solution to this scalability problem for real life constant bisectional-bandwidth fat-tree topologies. It is shown that maximal bandwidth and cut-through latency can be achieved for MPI global collective traffic. To form such congestion-free configuration, MPI programs should utilize collective communication, MPI-node-order should be topology aware, and the packets routing should match the MPI communication patterns. First, we show that MPI collectives can be classified into unidirectional and bidirectional shifts. Using this property, we propose a scheme for congestion-free routing of the global collectives in fully and partially populated fat trees running a single job. Simulation results of the proposed routing, MPI-node-order and communication patterns show a 40% throughput improvement over previously published results for all-to-all collectives.</abstract><pub>IEEE</pub><doi>10.1109/IPDPS.2011.219</doi><tpages>10</tpages></addata></record> |
fulltext | fulltext_linktorsrc |
identifier | ISSN: 1530-2075 |
ispartof | 2011 IEEE International Symposium on Parallel and Distributed Processing Workshops and Phd Forum, 2011, p.761-770 |
issn | 1530-2075 |
language | eng ; jpn |
recordid | cdi_ieee_primary_6008918 |
source | IEEE Xplore All Conference Series |
subjects | Bandwidth Clustering algorithms Degradation Indexes Routing Topology Traffic control |
title | Fat-Trees Routing and Node Ordering Providing Contention Free Traffic for MPI Global Collectives |
url | http://sfxeu10.hosted.exlibrisgroup.com/loughborough?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2024-12-25T18%3A07%3A57IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-ieee_CHZPO&rft_val_fmt=info:ofi/fmt:kev:mtx:book&rft.genre=proceeding&rft.atitle=Fat-Trees%20Routing%20and%20Node%20Ordering%20Providing%20Contention%20Free%20Traffic%20for%20MPI%20Global%20Collectives&rft.btitle=2011%20IEEE%20International%20Symposium%20on%20Parallel%20and%20Distributed%20Processing%20Workshops%20and%20Phd%20Forum&rft.au=Zahavi,%20E.&rft.date=2011-05&rft.spage=761&rft.epage=770&rft.pages=761-770&rft.issn=1530-2075&rft.isbn=9781612844251&rft.isbn_list=1612844251&rft_id=info:doi/10.1109/IPDPS.2011.219&rft.eisbn=9780769545776&rft.eisbn_list=0769545777&rft_dat=%3Cieee_CHZPO%3E6008918%3C/ieee_CHZPO%3E%3Cgrp_id%3Ecdi_FETCH-LOGICAL-i156t-cc902e04a91f88570d31b51425c3cc2e0b463ad71b4c4d052801403555af54453%3C/grp_id%3E%3Coa%3E%3C/oa%3E%3Curl%3E%3C/url%3E&rft_id=info:oai/&rft_id=info:pmid/&rft_ieee_id=6008918&rfr_iscdi=true |