Can parallel programming be made easy for scientists

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

The general opinion is that parallel programming is much harder than sequential programming. It is true if the progammer would like to reach over 90 Our P-GRADE environment was designed to meet these natural requirements of scientists. It is a completely graphical environment that supports the whole life-cycle of parallel program development. The programming language, called GRAPNEL, is a graphical extension of C, C++ or FORTRAN where graphics is used to express activities related to parallelism (like process creation, communication, etc.) and at the same time graphics hides the low level details of message passing library calls like PVM and MPI calls. Program constructs independent of parallelism can be inherited from sequential C, C++ or FOR- TRAN code. Moreover complete sequential C, C++ or FORTRAN libraries can be used in the GRAPNEL program and in this way parallelizing sequential code becomes extremely easy. Usage of predefined process topology templates enables the user to quickly generate very large parallel programs, too. A user-friendly dragg-and-drop style graphical editor (GRED) helps the pro- grammer to generate any necessary graphical constructs of GRPNEL. The DI- WIDE distributed debugger provides systematic and automatic discovery of deadlock situations that are the most common problems of message passing parallel programs. DIWIDE also supports replay technique and hence the cyclic debugging techniques like breakpoint, step-by-step execution can be applied even in a non-deterministic parallel programming system. Performance analysis is sup- ported by the GRM monitor and the PROVE execution visualization tool. The instrumentation is completely automatic, filters can be easily added or removed for the GRM monitor. The execution visualization can be done both off-line and on-line providing various synchronized trace-event views as well as statistics windows on processor utilization and communications. The connection between the source code and the trace-events can be easily identifed by the source code click-back and click-forward facilities. GRM and PROVE are able to support the observation of real-size, long-running parallel programs, too. In many cases per- formance bottlenecks are due to wrong mapping of processes to processors. An easy-to-use mapping tool supports the user to quickly rearrange the processes on the processors of the parallel system. The talk will highlight those features of P-GRADE that makes parallel pro- gramming really easy for non-hacker programmers, including scientists.

Original languageEnglish
Title of host publicationLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
PublisherSpringer Verlag
Pages7
Number of pages1
Volume2073
ISBN (Print)3540422323, 9783540422327
Publication statusPublished - 2001
EventInternational Conference on Computational Science, ICCS 2001 - San Francisco, United States
Duration: May 28 2001May 30 2001

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume2073
ISSN (Print)03029743
ISSN (Electronic)16113349

Other

OtherInternational Conference on Computational Science, ICCS 2001
CountryUnited States
CitySan Francisco
Period5/28/015/30/01

Fingerprint

Parallel programming
Parallel Programming
Parallel Programs
Message passing
D.3.2 [Programming Languages]: Language Classifications - Fortran
Visualization
C++
Communication
Message Passing
Computer programming
Computer programming languages
Parallelism
Life cycle
Monitor
Trace
Topology
Statistics
Deadlock
Tool Support
Debugging

ASJC Scopus subject areas

  • Computer Science(all)
  • Theoretical Computer Science

Cite this

Kacsuk, P. (2001). Can parallel programming be made easy for scientists. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 2073, pp. 7). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 2073). Springer Verlag.

Can parallel programming be made easy for scientists. / Kacsuk, P.

Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). Vol. 2073 Springer Verlag, 2001. p. 7 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 2073).

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Kacsuk, P 2001, Can parallel programming be made easy for scientists. in Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). vol. 2073, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 2073, Springer Verlag, pp. 7, International Conference on Computational Science, ICCS 2001, San Francisco, United States, 5/28/01.
Kacsuk P. Can parallel programming be made easy for scientists. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). Vol. 2073. Springer Verlag. 2001. p. 7. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).
Kacsuk, P. / Can parallel programming be made easy for scientists. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). Vol. 2073 Springer Verlag, 2001. pp. 7 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).
@inproceedings{ca943d3c22ad48b6bcd8bb5e2eba4725,
title = "Can parallel programming be made easy for scientists",
abstract = "The general opinion is that parallel programming is much harder than sequential programming. It is true if the progammer would like to reach over 90 Our P-GRADE environment was designed to meet these natural requirements of scientists. It is a completely graphical environment that supports the whole life-cycle of parallel program development. The programming language, called GRAPNEL, is a graphical extension of C, C++ or FORTRAN where graphics is used to express activities related to parallelism (like process creation, communication, etc.) and at the same time graphics hides the low level details of message passing library calls like PVM and MPI calls. Program constructs independent of parallelism can be inherited from sequential C, C++ or FOR- TRAN code. Moreover complete sequential C, C++ or FORTRAN libraries can be used in the GRAPNEL program and in this way parallelizing sequential code becomes extremely easy. Usage of predefined process topology templates enables the user to quickly generate very large parallel programs, too. A user-friendly dragg-and-drop style graphical editor (GRED) helps the pro- grammer to generate any necessary graphical constructs of GRPNEL. The DI- WIDE distributed debugger provides systematic and automatic discovery of deadlock situations that are the most common problems of message passing parallel programs. DIWIDE also supports replay technique and hence the cyclic debugging techniques like breakpoint, step-by-step execution can be applied even in a non-deterministic parallel programming system. Performance analysis is sup- ported by the GRM monitor and the PROVE execution visualization tool. The instrumentation is completely automatic, filters can be easily added or removed for the GRM monitor. The execution visualization can be done both off-line and on-line providing various synchronized trace-event views as well as statistics windows on processor utilization and communications. The connection between the source code and the trace-events can be easily identifed by the source code click-back and click-forward facilities. GRM and PROVE are able to support the observation of real-size, long-running parallel programs, too. In many cases per- formance bottlenecks are due to wrong mapping of processes to processors. An easy-to-use mapping tool supports the user to quickly rearrange the processes on the processors of the parallel system. The talk will highlight those features of P-GRADE that makes parallel pro- gramming really easy for non-hacker programmers, including scientists.",
author = "P. Kacsuk",
year = "2001",
language = "English",
isbn = "3540422323",
volume = "2073",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
publisher = "Springer Verlag",
pages = "7",
booktitle = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

}

TY - GEN

T1 - Can parallel programming be made easy for scientists

AU - Kacsuk, P.

PY - 2001

Y1 - 2001

N2 - The general opinion is that parallel programming is much harder than sequential programming. It is true if the progammer would like to reach over 90 Our P-GRADE environment was designed to meet these natural requirements of scientists. It is a completely graphical environment that supports the whole life-cycle of parallel program development. The programming language, called GRAPNEL, is a graphical extension of C, C++ or FORTRAN where graphics is used to express activities related to parallelism (like process creation, communication, etc.) and at the same time graphics hides the low level details of message passing library calls like PVM and MPI calls. Program constructs independent of parallelism can be inherited from sequential C, C++ or FOR- TRAN code. Moreover complete sequential C, C++ or FORTRAN libraries can be used in the GRAPNEL program and in this way parallelizing sequential code becomes extremely easy. Usage of predefined process topology templates enables the user to quickly generate very large parallel programs, too. A user-friendly dragg-and-drop style graphical editor (GRED) helps the pro- grammer to generate any necessary graphical constructs of GRPNEL. The DI- WIDE distributed debugger provides systematic and automatic discovery of deadlock situations that are the most common problems of message passing parallel programs. DIWIDE also supports replay technique and hence the cyclic debugging techniques like breakpoint, step-by-step execution can be applied even in a non-deterministic parallel programming system. Performance analysis is sup- ported by the GRM monitor and the PROVE execution visualization tool. The instrumentation is completely automatic, filters can be easily added or removed for the GRM monitor. The execution visualization can be done both off-line and on-line providing various synchronized trace-event views as well as statistics windows on processor utilization and communications. The connection between the source code and the trace-events can be easily identifed by the source code click-back and click-forward facilities. GRM and PROVE are able to support the observation of real-size, long-running parallel programs, too. In many cases per- formance bottlenecks are due to wrong mapping of processes to processors. An easy-to-use mapping tool supports the user to quickly rearrange the processes on the processors of the parallel system. The talk will highlight those features of P-GRADE that makes parallel pro- gramming really easy for non-hacker programmers, including scientists.

AB - The general opinion is that parallel programming is much harder than sequential programming. It is true if the progammer would like to reach over 90 Our P-GRADE environment was designed to meet these natural requirements of scientists. It is a completely graphical environment that supports the whole life-cycle of parallel program development. The programming language, called GRAPNEL, is a graphical extension of C, C++ or FORTRAN where graphics is used to express activities related to parallelism (like process creation, communication, etc.) and at the same time graphics hides the low level details of message passing library calls like PVM and MPI calls. Program constructs independent of parallelism can be inherited from sequential C, C++ or FOR- TRAN code. Moreover complete sequential C, C++ or FORTRAN libraries can be used in the GRAPNEL program and in this way parallelizing sequential code becomes extremely easy. Usage of predefined process topology templates enables the user to quickly generate very large parallel programs, too. A user-friendly dragg-and-drop style graphical editor (GRED) helps the pro- grammer to generate any necessary graphical constructs of GRPNEL. The DI- WIDE distributed debugger provides systematic and automatic discovery of deadlock situations that are the most common problems of message passing parallel programs. DIWIDE also supports replay technique and hence the cyclic debugging techniques like breakpoint, step-by-step execution can be applied even in a non-deterministic parallel programming system. Performance analysis is sup- ported by the GRM monitor and the PROVE execution visualization tool. The instrumentation is completely automatic, filters can be easily added or removed for the GRM monitor. The execution visualization can be done both off-line and on-line providing various synchronized trace-event views as well as statistics windows on processor utilization and communications. The connection between the source code and the trace-events can be easily identifed by the source code click-back and click-forward facilities. GRM and PROVE are able to support the observation of real-size, long-running parallel programs, too. In many cases per- formance bottlenecks are due to wrong mapping of processes to processors. An easy-to-use mapping tool supports the user to quickly rearrange the processes on the processors of the parallel system. The talk will highlight those features of P-GRADE that makes parallel pro- gramming really easy for non-hacker programmers, including scientists.

UR - http://www.scopus.com/inward/record.url?scp=84949653231&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84949653231&partnerID=8YFLogxK

M3 - Conference contribution

SN - 3540422323

SN - 9783540422327

VL - 2073

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 7

BT - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

PB - Springer Verlag

ER -