A high performance implementation of MPI-IO for a Lustre file system environment

It is often the case that MPI‐IO performs poorly in a Lustre file system environment, although the reasons for such performance have heretofore not been well understood. We hypothesize that such performance is a direct result of the fundamental assumptions upon which most parallel I/O optimizations...

Celý popis

Uložené v:
Podrobná bibliografia
Vydané v:Concurrency and computation Ročník 22; číslo 11; s. 1433 - 1449
Hlavní autori: Dickens, Phillip M., Logan, Jeremy
Médium: Journal Article
Jazyk:English
Vydavateľské údaje: Chichester, UK John Wiley & Sons, Ltd 10.08.2010
Predmet:
ISSN:1532-0626, 1532-0634, 1532-0634
On-line prístup:Získať plný text
Tagy: Pridať tag
Žiadne tagy, Buďte prvý, kto otaguje tento záznam!
Abstract It is often the case that MPI‐IO performs poorly in a Lustre file system environment, although the reasons for such performance have heretofore not been well understood. We hypothesize that such performance is a direct result of the fundamental assumptions upon which most parallel I/O optimizations are based. In particular, it is almost universally believed that parallel I/O performance is optimized when aggregator processes perform large, contiguous I/O operations in parallel. Our research, however, shows that this approach can actually provide the worst performance in a Lustre environment, and that the best performance may be obtained by performing a large number of small, non‐contiguous I/O operations. In this paper, we provide empirical results demonstrating these non‐intuitive results and explore the reasons for such unexpected performance. We present our solution to the problem, which is embodied in a user‐level library termed Y‐Lib, which redistributes the data in a way that conforms much more closely with the Lustre storage architecture than does the data redistribution pattern employed by MPI‐IO. We provide a large body of experimental results, taken across two large‐scale Lustre installations, demonstrating that Y‐Lib outperforms MPI‐IO by up to 36% on one system and 1000% on the other. We discuss the factors that impact the performance improvement obtained by Y‐Lib, which include the number of aggregator processes and Object Storage Devices, as well as the power of the system's communications infrastructure. We also show that the optimal data redistribution pattern for Y‐Lib is dependent upon these same factors. Copyright © 2009 John Wiley & Sons, Ltd.
AbstractList It is often the case that MPI‐IO performs poorly in a Lustre file system environment, although the reasons for such performance have heretofore not been well understood. We hypothesize that such performance is a direct result of the fundamental assumptions upon which most parallel I/O optimizations are based. In particular, it is almost universally believed that parallel I/O performance is optimized when aggregator processes perform large, contiguous I/O operations in parallel. Our research, however, shows that this approach can actually provide the worst performance in a Lustre environment, and that the best performance may be obtained by performing a large number of small, non‐contiguous I/O operations. In this paper, we provide empirical results demonstrating these non‐intuitive results and explore the reasons for such unexpected performance. We present our solution to the problem, which is embodied in a user‐level library termed Y‐Lib, which redistributes the data in a way that conforms much more closely with the Lustre storage architecture than does the data redistribution pattern employed by MPI‐IO. We provide a large body of experimental results, taken across two large‐scale Lustre installations, demonstrating that Y‐Lib outperforms MPI‐IO by up to 36% on one system and 1000% on the other. We discuss the factors that impact the performance improvement obtained by Y‐Lib, which include the number of aggregator processes and Object Storage Devices, as well as the power of the system's communications infrastructure. We also show that the optimal data redistribution pattern for Y‐Lib is dependent upon these same factors. Copyright © 2009 John Wiley & Sons, Ltd.
It is often the case that MPI-IO performs poorly in a Lustre file system environment, although the reasons for such performance have heretofore not been well understood. We hypothesize that such performance is a direct result of the fundamental assumptions upon which most parallel I/O optimizations are based. In particular, it is almost universally believed that parallel I/O performance is optimized when aggregator processes perform large, contiguous I/O operations in parallel. Our research, however, shows that this approach can actually provide the worst performance in a Lustre environment, and that the best performance may be obtained by performing a large number of small, non-contiguous I/O operations. In this paper, we provide empirical results demonstrating these non-intuitive results and explore the reasons for such unexpected performance. We present our solution to the problem, which is embodied in a user-level library termed Y-Lib, which redistributes the data in a way that conforms much more closely with the Lustre storage architecture than does the data redistribution pattern employed by MPI-IO. We provide a large body of experimental results, taken across two large-scale Lustre installations, demonstrating that Y-Lib outperforms MPI-IO by up to 36% on one system and 1000% on the other. We discuss the factors that impact the performance improvement obtained by Y-Lib, which include the number of aggregator processes and Object Storage Devices, as well as the power of the system's communications infrastructure. We also show that the optimal data redistribution pattern for Y-Lib is dependent upon these same factors.
Author Dickens, Phillip M.
Logan, Jeremy
Author_xml – sequence: 1
  givenname: Phillip M.
  surname: Dickens
  fullname: Dickens, Phillip M.
  email: dickens@umcs.maine.edu
  organization: Department of Computer Science, University of Maine, Orono, ME, U.S.A
– sequence: 2
  givenname: Jeremy
  surname: Logan
  fullname: Logan, Jeremy
  organization: Department of Computer Science, University of Maine, Orono, ME, U.S.A
BookMark eNp1kEFPwjAYhhuDiYAm_oQevQzbde3YkQxEEhQOKsemlG9S3dbZDpV_78gMxkRP33d43jd5nx7qlLYEhC4pGVBCwmtdwYBGCT1BXcpZGBDBos7xD8UZ6nn_QgilhNEuWo7w1jxvcQUus65QpQZsiiqHAspa1caW2Gb4bjkLZgvcEFjh-c7XDnBmcsB-72soMJTvxtnykDlHp5nKPVx83z56vJk8pLfBfDGdpaN5oBkTNIBwE64ZI8k6TIZKK7URjEdcRaIZoWPQEA5VxiGm0SaLEkUSYBEIHSrKiOAJ66Ortrdy9m0HvpaF8RryXJVgd15SEVM25JyLBh20qHbWeweZ1KbdVjtlckmJPKiTjTp5UPfTfQxUzhTK7f9Cgxb9aHTs_-Vkupz85k0j7vPIK_cqRcxiLlf3UzmOV2PCnhKZsi_afo0w
CitedBy_id crossref_primary_10_1109_ACCESS_2020_2985928
crossref_primary_10_1016_j_cpc_2024_109246
crossref_primary_10_1016_j_parco_2021_102786
crossref_primary_10_1002_mp_13033
crossref_primary_10_1007_s11227_011_0669_2
crossref_primary_10_1002_cpe_1613
Cites_doi 10.1109/CCGRID.2007.51
10.1016/S0167-8191(01)00129-6
10.1109/CLUSTR.2003.1253332
10.1109/IPDPS.2007.370485
10.1145/301816.301826
10.1109/FMPC.1999.750599
10.1109/IPDPS.2007.370239
ContentType Journal Article
Copyright Copyright © 2009 John Wiley & Sons, Ltd.
Copyright_xml – notice: Copyright © 2009 John Wiley & Sons, Ltd.
DBID BSCLL
AAYXX
CITATION
7SC
8FD
H8D
JQ2
L7M
L~C
L~D
DOI 10.1002/cpe.1491
DatabaseName Istex
CrossRef
Computer and Information Systems Abstracts
Technology Research Database
Aerospace Database
ProQuest Computer Science Collection
Advanced Technologies Database with Aerospace
Computer and Information Systems Abstracts – Academic
Computer and Information Systems Abstracts Professional
DatabaseTitle CrossRef
Aerospace Database
Technology Research Database
Computer and Information Systems Abstracts – Academic
ProQuest Computer Science Collection
Computer and Information Systems Abstracts
Advanced Technologies Database with Aerospace
Computer and Information Systems Abstracts Professional
DatabaseTitleList
Aerospace Database
CrossRef
DeliveryMethod fulltext_linktorsrc
Discipline Computer Science
EISSN 1532-0634
EndPage 1449
ExternalDocumentID 10_1002_cpe_1491
CPE1491
ark_67375_WNG_D7WD03V9_C
Genre article
GrantInformation_xml – fundername: National Science Foundation
  funderid: 0702748
GroupedDBID .3N
.DC
.GA
.Y3
05W
0R~
10A
1L6
1OC
31~
33P
3SF
3WU
4.4
50Y
50Z
51W
51X
52M
52N
52O
52P
52S
52T
52U
52W
52X
5GY
5VS
66C
702
7PT
8-0
8-1
8-3
8-4
8-5
8UM
930
A03
AAESR
AAEVG
AAHQN
AAMNL
AANHP
AANLZ
AAONW
AASGY
AAXRX
AAYCA
AAZKR
ABCQN
ABCUV
ABEML
ABIJN
ACAHQ
ACBWZ
ACCZN
ACPOU
ACRPL
ACSCC
ACXBN
ACXQS
ACYXJ
ADBBV
ADEOM
ADIZJ
ADKYN
ADMGS
ADMLS
ADNMO
ADOZA
ADXAS
ADZMN
AEIGN
AEIMD
AEUYR
AEYWJ
AFBPY
AFFPM
AFGKR
AFWVQ
AFZJQ
AGQPQ
AGYGG
AHBTC
AITYG
AIURR
AJXKR
ALMA_UNASSIGNED_HOLDINGS
ALVPJ
AMBMR
AMYDB
ASPBG
ATUGU
AUFTA
AVWKF
AZBYB
AZFZN
BAFTC
BDRZF
BFHJK
BHBCM
BMNLL
BROTX
BRXPI
BSCLL
BY8
CS3
D-E
D-F
DCZOG
DPXWK
DR2
DRFUL
DRSTM
EBS
EJD
F00
F01
F04
F5P
FEDTE
G-S
G.N
GNP
GODZA
HF~
HGLYW
HHY
HVGLF
HZ~
IX1
JPC
KQQ
LATKE
LAW
LC2
LC3
LEEKS
LH4
LITHE
LOXES
LP6
LP7
LUTES
LW6
LYRES
MEWTI
MK4
MRFUL
MRSTM
MSFUL
MSSTM
MXFUL
MXSTM
N04
N05
N9A
O66
O9-
OIG
P2W
P2X
P4D
PQQKQ
Q.N
Q11
QB0
QRW
R.K
ROL
RX1
SUPJJ
TN5
UB1
V2E
W8V
W99
WBKPD
WIH
WIK
WOHZO
WQJ
WXSBR
WYISQ
WZISG
XG1
XV2
~IA
~WT
AAHHS
ACCFJ
ADZOD
AEEZP
AEQDE
AEUQT
AFPWT
AIWBW
AJBDE
ALUQN
RWI
WRC
AAYXX
CITATION
O8X
7SC
8FD
H8D
JQ2
L7M
L~C
L~D
ID FETCH-LOGICAL-c3361-e2d2b3309b298acaad63545a46002c7ece28af5e714df49a09e34e6c2a1306593
IEDL.DBID DRFUL
ISICitedReferencesCount 10
ISICitedReferencesURI http://www.webofscience.com/api/gateway?GWVersion=2&SrcApp=Summon&SrcAuth=ProQuest&DestLinkType=CitingArticles&DestApp=WOS_CPL&KeyUT=000280431500006&url=https%3A%2F%2Fcvtisr.summon.serialssolutions.com%2F%23%21%2Fsearch%3Fho%3Df%26include.ft.matches%3Dt%26l%3Dnull%26q%3D
ISSN 1532-0626
1532-0634
IngestDate Sun Nov 09 09:26:25 EST 2025
Sat Nov 29 01:41:10 EST 2025
Tue Nov 18 21:16:21 EST 2025
Wed Jan 22 16:19:40 EST 2025
Tue Nov 11 03:32:29 EST 2025
IsPeerReviewed true
IsScholarly true
Issue 11
Language English
License http://onlinelibrary.wiley.com/termsAndConditions#vor
LinkModel DirectLink
MergedId FETCHMERGED-LOGICAL-c3361-e2d2b3309b298acaad63545a46002c7ece28af5e714df49a09e34e6c2a1306593
Notes National Science Foundation - No. 0702748
ark:/67375/WNG-D7WD03V9-C
ArticleID:CPE1491
istex:51734B87519EF8407F97C21154D37ED12BBCD575
ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 23
PQID 1671385556
PQPubID 23500
PageCount 17
ParticipantIDs proquest_miscellaneous_1671385556
crossref_citationtrail_10_1002_cpe_1491
crossref_primary_10_1002_cpe_1491
wiley_primary_10_1002_cpe_1491_CPE1491
istex_primary_ark_67375_WNG_D7WD03V9_C
PublicationCentury 2000
PublicationDate 2010-08-10
10 August 2010
20100810
PublicationDateYYYYMMDD 2010-08-10
PublicationDate_xml – month: 08
  year: 2010
  text: 2010-08-10
  day: 10
PublicationDecade 2010
PublicationPlace Chichester, UK
PublicationPlace_xml – name: Chichester, UK
PublicationTitle Concurrency and computation
PublicationTitleAlternate Concurrency Computat.: Pract. Exper
PublicationYear 2010
Publisher John Wiley & Sons, Ltd
Publisher_xml – name: John Wiley & Sons, Ltd
References Thakur R, Gropp W, Lusk E. Optimizing noncontiguous accesses in MPI-IO. Parallel Computing 2002; 28(1):83-105.
2009
2002; 28
2007
2004
e_1_2_9_10_2
e_1_2_9_21_2
e_1_2_9_20_2
e_1_2_9_12_2
e_1_2_9_11_2
e_1_2_9_22_2
e_1_2_9_7_2
e_1_2_9_6_2
e_1_2_9_5_2
e_1_2_9_4_2
e_1_2_9_3_2
e_1_2_9_2_2
e_1_2_9_9_2
e_1_2_9_8_2
e_1_2_9_14_2
e_1_2_9_13_2
e_1_2_9_16_2
e_1_2_9_15_2
e_1_2_9_18_2
e_1_2_9_17_2
e_1_2_9_19_2
References_xml – reference: Thakur R, Gropp W, Lusk E. Optimizing noncontiguous accesses in MPI-IO. Parallel Computing 2002; 28(1):83-105.
– year: 2009
– start-page: 23
  end-page: 32
– volume: 28
  start-page: 83
  issue: 1
  year: 2002
  end-page: 105
  article-title: Optimizing noncontiguous accesses in MPI‐IO
  publication-title: Parallel Computing
– start-page: 104
  end-page: 111
– start-page: 182
  end-page: 189
– year: 2007
– year: 2004
– ident: e_1_2_9_17_2
– ident: e_1_2_9_21_2
– ident: e_1_2_9_8_2
  doi: 10.1109/CCGRID.2007.51
– ident: e_1_2_9_14_2
  doi: 10.1016/S0167-8191(01)00129-6
– ident: e_1_2_9_5_2
– ident: e_1_2_9_11_2
– ident: e_1_2_9_2_2
– ident: e_1_2_9_13_2
  doi: 10.1109/CLUSTR.2003.1253332
– ident: e_1_2_9_6_2
– ident: e_1_2_9_19_2
  doi: 10.1109/IPDPS.2007.370485
– ident: e_1_2_9_3_2
– ident: e_1_2_9_22_2
– ident: e_1_2_9_20_2
– ident: e_1_2_9_12_2
– ident: e_1_2_9_7_2
– ident: e_1_2_9_4_2
– ident: e_1_2_9_15_2
– ident: e_1_2_9_10_2
  doi: 10.1145/301816.301826
– ident: e_1_2_9_9_2
  doi: 10.1109/FMPC.1999.750599
– ident: e_1_2_9_16_2
– ident: e_1_2_9_18_2
  doi: 10.1109/IPDPS.2007.370239
SSID ssj0011031
Score 1.9941487
Snippet It is often the case that MPI‐IO performs poorly in a Lustre file system environment, although the reasons for such performance have heretofore not been well...
It is often the case that MPI-IO performs poorly in a Lustre file system environment, although the reasons for such performance have heretofore not been well...
SourceID proquest
crossref
wiley
istex
SourceType Aggregation Database
Enrichment Source
Index Database
Publisher
StartPage 1433
SubjectTerms Concurrency
Devices
grid computing
Infrastructure
Libraries
Lustre
Mathematical models
object-based file systems
Optimization
parallel I/O
Yttrium
Title A high performance implementation of MPI-IO for a Lustre file system environment
URI https://api.istex.fr/ark:/67375/WNG-D7WD03V9-C/fulltext.pdf
https://onlinelibrary.wiley.com/doi/abs/10.1002%2Fcpe.1491
https://www.proquest.com/docview/1671385556
Volume 22
WOSCitedRecordID wos000280431500006&url=https%3A%2F%2Fcvtisr.summon.serialssolutions.com%2F%23%21%2Fsearch%3Fho%3Df%26include.ft.matches%3Dt%26l%3Dnull%26q%3D
hasFullText 1
inHoldings 1
isFullTextHit
isPrint
journalDatabaseRights – providerCode: PRVWIB
  databaseName: Wiley Online Library Full Collection 2020
  customDbUrl:
  eissn: 1532-0634
  dateEnd: 99991231
  omitProxy: false
  ssIdentifier: ssj0011031
  issn: 1532-0626
  databaseCode: DRFUL
  dateStart: 20010101
  isFulltext: true
  titleUrlDefault: https://onlinelibrary.wiley.com
  providerName: Wiley-Blackwell
link http://cvtisr.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwpV1LSwMxEA5iPXixPrG-iCB6WtxNNt3NUVqrQq1FfN1CNjuBoralreLRn-Bv9JeY2UdVUBA87WWSXWYymy-ZmW8I2TMp4wIg8AD81AultZ6OfesJHYtUM21tVl9x0446nfjuTnaLrEqshcn5IaYXbugZ2f8aHVwn48NP0lAzBOfmWLhewZoqd_CqNC9b1-1pDAEbGORsqczzHW4vqWd9dliO_bYZVVCvL9-Q5le8mm04rep_PnWRLBQwkx7l62KJzEB_mVTLFg608OgVcnlEkbCYDj_rB2jvscwpR6PRgaXn3bP317ezC-pkqKZtvB8BipRONGeCpl_q5VbJdev4qnHqFW0WPMN53VmJpSzh3JcJk7E2WqcOhIRChxiyMxEYYLG2AqIgTG0otS-Bh1A3TAfYdl7yNTLbH_RhnVCTCGGETACEDXnq4AZwN9xYhxJ1ECU1clDqW5mCgxxbYTyonD2ZKacqhaqqkd2p5DDn3fhBZj8z2VRAj-4xTy0S6rZzoprRbdPnN1I13GSlTZXzHgyJ6D4MnsYqqLtDeiyEqLvJMhP--jbV6B7jc-OvgptkPk81QP7cLTI7GT3BNpkzz5PeeLRTrNYP2CXu4w
linkProvider Wiley-Blackwell
linkToHtml http://cvtisr.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwpV1La9wwEB7CbqC9NEkfNG0eKpT2ZGJLlm2RU9jNJkud7RLyuglZHkFou7tsktJjf0J-Y35JNH5sEmih0JMvI9nMaKTPmplvAD7akguJGAWIYRnEyrnAZKELpMlkabhxrqqvOMvT0Si7uFDjJdhta2FqfojFhRt5RrVfk4PThfTOA2uonaH3c6pc78aJSLMOdPvHg9N8EUSgDgY1XSoPQg_cW-7ZkO-0Y5-cRl1S7K8nUPMxYK1OnMHKf33rKrxogCbbq1fGGizh5CWstE0cWOPTr-B4jxFlMZs9VBCwyx9tVjmZjU0dOxoP737fDr8yL8MMy-mGBBmROrGaC5o9qph7DaeD_ZPeYdA0WgisEIm3Ey95IUSoCq4yY40pPQyJpYkpaGdTtMgz4ySmUVy6WJlQoYgxsdxE1HheiTfQmUwn-BaYLaS0UhWI0sWi9IADhR9unceJJkqLdfjcKlzbhoWcmmF81zV_MtdeVZpUtQ4fFpKzmnnjDzKfKpstBMz8G2WqpVKfjw50Pz3vh-JM6Z6frDWq9v5DQREzwenNlY4S_5ueSSkTP1llw7--TffG-_R896-C2_Ds8OQo1_lw9OU9PK8TD4hNdwM61_Mb3IRl-_P68mq-1SzdeyLJ8tM
linkToPdf http://cvtisr.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwpV1LTxsxEB6hpKp6AUpbQQvUlVB7WrHrR3YtTighNGoIESqPm-X1jiXUkkQBqh77E_ob-SV49hFAaqVKPe1l7F3NeOxvPTPfAOy4gguFmESIcRFJ7X1ks9hHymaqsNx6X9ZXnA3T0Si7uNDjJdhramEqfojFhRt5Rrlfk4PjrPC7D6yhbobBz6lyvS2VVrIF7d5J_3S4CCJQB4OKLpVHcQDuDfdszHebsU9OozYp9ucTqPkYsJYnTn_lv751FZZroMn2q5XxEpZwsgYrTRMHVvv0KzjZZ0RZzGYPFQTs8qrJKiezsalnR-PB3a_fg2MWZJhlQ7ohQUakTqzigmaPKuZew2n_4Gv3c1Q3WoicEJ1gJ17wXIhY51xn1llbBBgilZUUtHMpOuSZ9QrTRBZeahtrFBI7jtuEGs9r8QZak-kE14G5XCmndI6ovBRFABwownDnA060SZpvwKdG4cbVLOTUDOO7qfiTuQmqMqSqDfiwkJxVzBt_kPlY2mwhYOffKFMtVeZ8dGh66XkvFmfadMNkjVFN8B8KitgJTm-vTdIJv-mZUqoTJitt-Ne3me74gJ5v_1XwPTwf9_pmOBh9eQcvqrwDItPdhNbN_Ba34Jn7cXN5Pd-uV-49wl3yTg
openUrl ctx_ver=Z39.88-2004&ctx_enc=info%3Aofi%2Fenc%3AUTF-8&rfr_id=info%3Asid%2Fsummon.serialssolutions.com&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.atitle=A+high+performance+implementation+of+MPI%E2%80%90IO+for+a+Lustre+file+system+environment&rft.jtitle=Concurrency+and+computation&rft.au=Dickens%2C+Phillip+M.&rft.au=Logan%2C+Jeremy&rft.date=2010-08-10&rft.issn=1532-0626&rft.eissn=1532-0634&rft.volume=22&rft.issue=11&rft.spage=1433&rft.epage=1449&rft_id=info:doi/10.1002%2Fcpe.1491&rft.externalDBID=n%2Fa&rft.externalDocID=10_1002_cpe_1491
thumbnail_l http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/lc.gif&issn=1532-0626&client=summon
thumbnail_m http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/mc.gif&issn=1532-0626&client=summon
thumbnail_s http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/sc.gif&issn=1532-0626&client=summon