Scheduling a 100,000 Core Supercomputer for Maximum Utilization and Capability

In late 2009, the National Institute for Computational Sciences placed in production the world's fastest academic supercomputer (third overall), a Cray XT5 named Kraken, with almost 100,000 compute cores and a peak speed in excess of one Petaflop. Delivering over 50% of the total cycles availab...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Andrews, Phil, Kovatch, Patricia, Hazlewood, Victor, Baer, Troy
Format: Tagungsbericht
Sprache:eng
Schlagworte:
Online-Zugang:Volltext bestellen
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
container_end_page 427
container_issue
container_start_page 421
container_title
container_volume
creator Andrews, Phil
Kovatch, Patricia
Hazlewood, Victor
Baer, Troy
description In late 2009, the National Institute for Computational Sciences placed in production the world's fastest academic supercomputer (third overall), a Cray XT5 named Kraken, with almost 100,000 compute cores and a peak speed in excess of one Petaflop. Delivering over 50% of the total cycles available to the National Science Foundation users via the TeraGrid, Kraken has two missions that have historically proven difficult to simultaneously reconcile: providing the maximum number of total cycles to the community, while enabling full machine runs for "hero" users. Historically, this has been attempted by allowing schedulers to choose the correct time for the beginning of large jobs, with a concomitant reduction in utilization. At NICS, we used the results of a previous theoretical investigation to adopt a different approach, where the "clearing out" of the system is forced on a weekly basis, followed by consecutive full machine runs. As our previous simulation results suggested, this lead to a significant improvement in utilization, to over 90%. The difference in utilization between the traditional and adopted scheduling policies was the equivalent of a 300+ Teraflop supercomputer, or several million dollars of compute time per year.
doi_str_mv 10.1109/ICPPW.2010.63
format Conference Proceeding
fullrecord <record><control><sourceid>ieee_6IE</sourceid><recordid>TN_cdi_ieee_primary_5599101</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><ieee_id>5599101</ieee_id><sourcerecordid>5599101</sourcerecordid><originalsourceid>FETCH-LOGICAL-i90t-8c8dae2035d2ace0cf4db5658ece7605dbe336657bb3d1dfbb97c1c08172fc2e3</originalsourceid><addsrcrecordid>eNotjEtLw0AUhccX2NYuXbmZH2DqvfPMLCVULVQttOKyzCs60jQhTcD66w3o2Ry-78Ah5BphhgjmblGsVu8zBgMrfkLGoJWRAqXWp2TEOGeZVAbOyBgFE0IbzOU5GQEayPgAl2R6OHzBECEZajEiL2v_GUO_S_sPaikC3A4jLeo20nXfxNbXVdN3saVl3dJn-52qvqJvXdqlH9ulek_tPtDCNtYNqjtekYvS7g5x-t8TsnmYb4qnbPn6uCjul1ky0GW5z4ONDLgMzPoIvhTBSSXz6KNWIIOLnCsltXM8YCidM9qjhxw1Kz2LfEJu_m5TjHHbtKmy7XErpTEIyH8BcSJRZg</addsrcrecordid><sourcetype>Publisher</sourcetype><iscdi>true</iscdi><recordtype>conference_proceeding</recordtype></control><display><type>conference_proceeding</type><title>Scheduling a 100,000 Core Supercomputer for Maximum Utilization and Capability</title><source>IEEE Electronic Library (IEL) Conference Proceedings</source><creator>Andrews, Phil ; Kovatch, Patricia ; Hazlewood, Victor ; Baer, Troy</creator><creatorcontrib>Andrews, Phil ; Kovatch, Patricia ; Hazlewood, Victor ; Baer, Troy</creatorcontrib><description>In late 2009, the National Institute for Computational Sciences placed in production the world's fastest academic supercomputer (third overall), a Cray XT5 named Kraken, with almost 100,000 compute cores and a peak speed in excess of one Petaflop. Delivering over 50% of the total cycles available to the National Science Foundation users via the TeraGrid, Kraken has two missions that have historically proven difficult to simultaneously reconcile: providing the maximum number of total cycles to the community, while enabling full machine runs for "hero" users. Historically, this has been attempted by allowing schedulers to choose the correct time for the beginning of large jobs, with a concomitant reduction in utilization. At NICS, we used the results of a previous theoretical investigation to adopt a different approach, where the "clearing out" of the system is forced on a weekly basis, followed by consecutive full machine runs. As our previous simulation results suggested, this lead to a significant improvement in utilization, to over 90%. The difference in utilization between the traditional and adopted scheduling policies was the equivalent of a 300+ Teraflop supercomputer, or several million dollars of compute time per year.</description><identifier>ISSN: 0190-3918</identifier><identifier>ISBN: 1424479185</identifier><identifier>ISBN: 9781424479184</identifier><identifier>EISSN: 2332-5690</identifier><identifier>EISBN: 0769541577</identifier><identifier>EISBN: 9780769541570</identifier><identifier>DOI: 10.1109/ICPPW.2010.63</identifier><language>eng</language><publisher>IEEE</publisher><subject>Aggregates ; High-performance computing ; Processor scheduling ; Production ; Resource management ; Runtime ; scheduling ; Supercomputers ; systems software</subject><ispartof>2010 39th International Conference on Parallel Processing Workshops, 2010, p.421-427</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/5599101$$EHTML$$P50$$Gieee$$H</linktohtml><link.rule.ids>309,310,780,784,789,790,2058,27925,54920</link.rule.ids><linktorsrc>$$Uhttps://ieeexplore.ieee.org/document/5599101$$EView_record_in_IEEE$$FView_record_in_$$GIEEE</linktorsrc></links><search><creatorcontrib>Andrews, Phil</creatorcontrib><creatorcontrib>Kovatch, Patricia</creatorcontrib><creatorcontrib>Hazlewood, Victor</creatorcontrib><creatorcontrib>Baer, Troy</creatorcontrib><title>Scheduling a 100,000 Core Supercomputer for Maximum Utilization and Capability</title><title>2010 39th International Conference on Parallel Processing Workshops</title><addtitle>icppw</addtitle><description>In late 2009, the National Institute for Computational Sciences placed in production the world's fastest academic supercomputer (third overall), a Cray XT5 named Kraken, with almost 100,000 compute cores and a peak speed in excess of one Petaflop. Delivering over 50% of the total cycles available to the National Science Foundation users via the TeraGrid, Kraken has two missions that have historically proven difficult to simultaneously reconcile: providing the maximum number of total cycles to the community, while enabling full machine runs for "hero" users. Historically, this has been attempted by allowing schedulers to choose the correct time for the beginning of large jobs, with a concomitant reduction in utilization. At NICS, we used the results of a previous theoretical investigation to adopt a different approach, where the "clearing out" of the system is forced on a weekly basis, followed by consecutive full machine runs. As our previous simulation results suggested, this lead to a significant improvement in utilization, to over 90%. The difference in utilization between the traditional and adopted scheduling policies was the equivalent of a 300+ Teraflop supercomputer, or several million dollars of compute time per year.</description><subject>Aggregates</subject><subject>High-performance computing</subject><subject>Processor scheduling</subject><subject>Production</subject><subject>Resource management</subject><subject>Runtime</subject><subject>scheduling</subject><subject>Supercomputers</subject><subject>systems software</subject><issn>0190-3918</issn><issn>2332-5690</issn><isbn>1424479185</isbn><isbn>9781424479184</isbn><isbn>0769541577</isbn><isbn>9780769541570</isbn><fulltext>true</fulltext><rsrctype>conference_proceeding</rsrctype><creationdate>2010</creationdate><recordtype>conference_proceeding</recordtype><sourceid>6IE</sourceid><sourceid>RIE</sourceid><recordid>eNotjEtLw0AUhccX2NYuXbmZH2DqvfPMLCVULVQttOKyzCs60jQhTcD66w3o2Ry-78Ah5BphhgjmblGsVu8zBgMrfkLGoJWRAqXWp2TEOGeZVAbOyBgFE0IbzOU5GQEayPgAl2R6OHzBECEZajEiL2v_GUO_S_sPaikC3A4jLeo20nXfxNbXVdN3saVl3dJn-52qvqJvXdqlH9ulek_tPtDCNtYNqjtekYvS7g5x-t8TsnmYb4qnbPn6uCjul1ky0GW5z4ONDLgMzPoIvhTBSSXz6KNWIIOLnCsltXM8YCidM9qjhxw1Kz2LfEJu_m5TjHHbtKmy7XErpTEIyH8BcSJRZg</recordid><startdate>201009</startdate><enddate>201009</enddate><creator>Andrews, Phil</creator><creator>Kovatch, Patricia</creator><creator>Hazlewood, Victor</creator><creator>Baer, Troy</creator><general>IEEE</general><scope>6IE</scope><scope>6IL</scope><scope>CBEJK</scope><scope>RIE</scope><scope>RIL</scope></search><sort><creationdate>201009</creationdate><title>Scheduling a 100,000 Core Supercomputer for Maximum Utilization and Capability</title><author>Andrews, Phil ; Kovatch, Patricia ; Hazlewood, Victor ; Baer, Troy</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-i90t-8c8dae2035d2ace0cf4db5658ece7605dbe336657bb3d1dfbb97c1c08172fc2e3</frbrgroupid><rsrctype>conference_proceedings</rsrctype><prefilter>conference_proceedings</prefilter><language>eng</language><creationdate>2010</creationdate><topic>Aggregates</topic><topic>High-performance computing</topic><topic>Processor scheduling</topic><topic>Production</topic><topic>Resource management</topic><topic>Runtime</topic><topic>scheduling</topic><topic>Supercomputers</topic><topic>systems software</topic><toplevel>online_resources</toplevel><creatorcontrib>Andrews, Phil</creatorcontrib><creatorcontrib>Kovatch, Patricia</creatorcontrib><creatorcontrib>Hazlewood, Victor</creatorcontrib><creatorcontrib>Baer, Troy</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>Andrews, Phil</au><au>Kovatch, Patricia</au><au>Hazlewood, Victor</au><au>Baer, Troy</au><format>book</format><genre>proceeding</genre><ristype>CONF</ristype><atitle>Scheduling a 100,000 Core Supercomputer for Maximum Utilization and Capability</atitle><btitle>2010 39th International Conference on Parallel Processing Workshops</btitle><stitle>icppw</stitle><date>2010-09</date><risdate>2010</risdate><spage>421</spage><epage>427</epage><pages>421-427</pages><issn>0190-3918</issn><eissn>2332-5690</eissn><isbn>1424479185</isbn><isbn>9781424479184</isbn><eisbn>0769541577</eisbn><eisbn>9780769541570</eisbn><abstract>In late 2009, the National Institute for Computational Sciences placed in production the world's fastest academic supercomputer (third overall), a Cray XT5 named Kraken, with almost 100,000 compute cores and a peak speed in excess of one Petaflop. Delivering over 50% of the total cycles available to the National Science Foundation users via the TeraGrid, Kraken has two missions that have historically proven difficult to simultaneously reconcile: providing the maximum number of total cycles to the community, while enabling full machine runs for "hero" users. Historically, this has been attempted by allowing schedulers to choose the correct time for the beginning of large jobs, with a concomitant reduction in utilization. At NICS, we used the results of a previous theoretical investigation to adopt a different approach, where the "clearing out" of the system is forced on a weekly basis, followed by consecutive full machine runs. As our previous simulation results suggested, this lead to a significant improvement in utilization, to over 90%. The difference in utilization between the traditional and adopted scheduling policies was the equivalent of a 300+ Teraflop supercomputer, or several million dollars of compute time per year.</abstract><pub>IEEE</pub><doi>10.1109/ICPPW.2010.63</doi><tpages>7</tpages></addata></record>
fulltext fulltext_linktorsrc
identifier ISSN: 0190-3918
ispartof 2010 39th International Conference on Parallel Processing Workshops, 2010, p.421-427
issn 0190-3918
2332-5690
language eng
recordid cdi_ieee_primary_5599101
source IEEE Electronic Library (IEL) Conference Proceedings
subjects Aggregates
High-performance computing
Processor scheduling
Production
Resource management
Runtime
scheduling
Supercomputers
systems software
title Scheduling a 100,000 Core Supercomputer for Maximum Utilization and Capability
url https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2024-12-23T06%3A59%3A16IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-ieee_6IE&rft_val_fmt=info:ofi/fmt:kev:mtx:book&rft.genre=proceeding&rft.atitle=Scheduling%20a%20100,000%20Core%20Supercomputer%20for%20Maximum%20Utilization%20and%20Capability&rft.btitle=2010%2039th%20International%20Conference%20on%20Parallel%20Processing%20Workshops&rft.au=Andrews,%20Phil&rft.date=2010-09&rft.spage=421&rft.epage=427&rft.pages=421-427&rft.issn=0190-3918&rft.eissn=2332-5690&rft.isbn=1424479185&rft.isbn_list=9781424479184&rft_id=info:doi/10.1109/ICPPW.2010.63&rft_dat=%3Cieee_6IE%3E5599101%3C/ieee_6IE%3E%3Curl%3E%3C/url%3E&rft.eisbn=0769541577&rft.eisbn_list=9780769541570&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_id=info:pmid/&rft_ieee_id=5599101&rfr_iscdi=true