Fifth European Workshop on OpenMP

Part II: OMPlab

Aachen, Germany

September 24-26, 2003

http://www.rz.rwth-aachen.de/ewomp03/OMPlab.html

Slides and papers of OMPtalks are available online now!

Slides of OMPlab are available online now!

Descriptions of Tools for OMPlab are collected here.

 
Introduction
      This second part of the workshop is a unique opportunity to not only discuss ideas with many experts in the field of OpenMP programming, but also to put hands on real hardware and work on real codes.

      Two days of talks and discussions during OMP
talks
      will be highly motivating to carry out practical experiments and to try out new ideas. This will be a unique chance for application programmers and compiler and tool developers to sit together, work on their own codes and apply their own compilers and tools in a large production environment.

      Note, that it is not obligatory to register for both parts of EWOMP'03.


    We are glad to provide a nice "playground" for such experiments:
    • an SMP cluster of 8 Sun Fire 6800 and 4 Sun Fire 15K SMP servers equipped with 24 respectively 72 UltraSPARC III Cu processors in Aachen
    • access to an IBM Regatta system with 32 Power4 CPUs in the nearby Research Center Jülich.
    • access to a 4-CPU Fujitsu Primepower system running the Fujitsu Fortran 95, C99, and C++ compilers with a complete OpenMP 2.0 implementation including nested parallelism, dynamic thread adjustment, and an OpenMP debugger and performance analyser.
    There already is a nice software suite available on these systems, which will be furthermore completed by the compilers and tools provided by participants of this workshop. So far on the Sun Fire systems there is available:
    • Sun Studio ONE compilers for Fortran95, C and C++ supporting OpenMP together with the very handy Sun performance analyzer.
    • The Sun HPC ClusterTools V5 Environment contains a fully thread-safe MPI2 environment, well suited for hybrid OpenMP+MPI applications. The Sun performance analyzer is also able to handle hybrid programs effectively.
    • KAI/Intels KAP Pro/Toolset consisting of the guidef90, guidec and guidec++ compilers, the guideview performance analyzer and the assuref90, assurec and assurec++ OpenMP program verification tools with the assureview GUI.
    • The TotalView Debugger from Etnus.
    • The Vampir and VampirTrace tools for the performance analysis of MPI codes.
  
Lab Material
  
Important Dates
April 16Call for papers and participation
June 29(*)Deadline for submission of OMPlab project proposals (max 5 pages).
July 21(*)Notification of authors about accepted projects for OMPlab.
August 4End of early registration period
September 15Deadline for installing tools and applications for OMPlab
September 15End of late registration period . Onsite registration will be possible. Payment in cash only.
September 22-23EWOMP'03 part I: OMPtalks
September 24-26EWOMP'03 part II: OMPlab
      (*) The composition of the OMP
lab
    will be an iterative process, which is supposed to converge into interesting cooperations.
  
Registration

The registration (SSL) is opened. It is handled by the Aachen tourist office.
(You can use the unsecure non-SSL version, in case the SSL-version doesn't work for you)
The registration is limited to 40 participants for the OMPlab part on a first-come, first-served basis.

early registration fee80 EUR
early registration fee for undergraduate and graduate students60 EUR
late registration fee100 EUR

 

      The registration fee includes access to presentations and hands-on sessions, printed handouts, breaks, and the "pizza & beer" style dinner on the evening of September 25th. (
This date has been shifted!
      )


    Students are requested to provide an affirmation of their advisor.
  
Submission / Participation
      We would like you to describe your OpenMP compiler, your tool or your parallel application, which you want to bring into the OMP
lab
      session and give a short oral introduction (~15 minutes). Please also express your desires and expectations for this lab.

      We plan to organize the participation in this lab session in project groups of 1-3 persons. Therefore please specify the number of persons who will be representing your project.

      • If you are an OpenMP application developer -
        please provide a rough overview of the background of your OpenMP application and the algorithm you chose. Describe your parallelisation approach and provide an easy-to-use makefile mechanism and suitable datasets such that a tool developer can easily run your application and make performance experiments in a limited amount of time.
        Which kind of experiments do you want to make?
        Do you want to get new ideas for the parallelization of your code?
        Do you have an algorithm, which is particularly difficult to parallelize with OpenMP?
        Parallelizing algorithms handling trees or lists is currently a hot topic.
        Also mixed OpenMP and MPI parallelization (particularly on clusters of SMP's) is a rewarding issue.
        Do you want to improve the performance and/or the scalability of your code?
        Do you want to profit from a specific feature of one (or some) of the contributed compilers/tools?
      • If you are a compiler/tool developer -
        please provide a quick reference guide and/or a user's guide to your tool such that an OpenMP application developer can easily apply your tool to his code.
        Which kind of experiments do you want to make? Do you want to validate the effectiveness of your tool? Do you want to get new ideas for further development of your tool? Do you just want to win new customers for your tool (which is considered a valid approach in this context)?

Please take into account that still the time for OMPlab is limited, don't put your expectations too high, try to be realistic.
Please submit a description of your contribution of up to 5 pages in length in Acrobat .pdf or Word .doc format.
All submissions should be sent to ewomp03@rz.rwth-aachen.de.
As soon as the program committee has taken any decisions, we will update this web information to make the composition of this lab an iterative process.
If your contribution has been accepted, you are welcome to provide hard copy or online information for the lab (user guides, etc.)
We will grant access to our machines for implementing, porting and preparation of your compilers/tools/application.

  
Contributed Projects
ParticipantsTitledescriptionkeywords

Brian Wiley, Nawal Copty, Ruud van der Pas
Sun Microsystems

Sun ONE Studio 8

html

RWTH-Primer

Slides

C, C++, F95 Compiler + Performance Tools

Matthijs van Waveren
Fujitsu Systems Europe
DSSE (Distributed Scalable Service Environment)
The service of DSSE Workbench for EWOMP03 would be to allow participants to seamlessly use the OpenMP 2.0 compilers (Fortran, C, and C++) on the PRIMEPOWER in Paris from Aachen.
pdf

Slides

C, C++, F95 Compiler, Debugger + Performance Tools

Nested Parallelization supported

Luiz DeRose,
IBM Research
The IBM Regatta System

Slides

C, C++, F95 Compiler, Debugger + Performance Tools

Luiz DeRose,
IBM Research
An Implementation of the POMP Performance Monitoring Interface

Slides

Performance Analysis Tool
Mitsuhisa Sato, Yoshinori OjimaExperiment with Cluster-enabled OpenMP: OpenMP for Software DSM System on PC Clusterspdf (4)

Slides

Access to an Distributed Shared Memory (DSM) System with OpenMP C and F77 compiler

Francisco de Sande et.al.The llc language and its implementationpdf (4)

Slides

OpenMP extension for multi-level parallelism and distributed memory architectures.

Sanjiv Shah,
Grant Haab,
Intel Corp.
Intel ThreadCheckerhtml

Verification Tool

Dieter an Mey

TotalView Parallel Debuggerpdf(5)

Slides

Debugger

Bernd Mohr (Felix Wolf)
Research Center Jülich
KOJAK - A Tool Set for Automatic Performance Analysis of Parallel Programspdf (4)

Slides

Performance Analysis Tool
Eduard Ayguade
European Center for Parallelism of Barcelona
Paraverhtml

Slides

Short Guide

Performance Analysis Tool
Federico Massaioli, (Massimo Bernaschi, Filippo Castiglione) CASPUR, RomeOpenMP parallelization of agent-based modelspdf (5)

Slides

Application

Alistair Rendell
ANU
Gaussian 03 Application, Fortran
Jörg Multhoff
Aachen University
Finite Element Programming in C++ using OpenMPpdf (3)

Slides

Application, C++

H. Martin Bücker, Arno Rasch, Aachen UniversityNested Parallelism when Differentiating OpenMP Programspdf (1)

Application

Thomas Beelitz, Christian H. Bischof, Bruno Lang, Univ. Wuppertal, Aachen Univ.Intervals and OpenMP: Towards an Efficient Parallel Results-Verifying Nonlinear Solverpdf (5)

Slides

Application, C++
Hans Thermann, Bernd Wickerath, Daniel Grates, Stephan Schmidt, Dieter an Mey, Chrstian Terboven
Aachen University

Panta - Navier-Stokes solver for the modeling of turbomachinery

Hybrid parallelization, nested OpenMP version in preparation

pdf (1), html

Slides

Application, Fortran
Ulrich Wepler, Dieter an Mey, Thomas Haarmann, Alexander Spiegel, Wolfgang KoschelThe Thermoflow60 Finite-Element Programpdf (5)

Slides (1), Slides (2)

Application, Fortran
Greg Bronevetsky et.al.
Cornell Univ. US
Blocking Application-level Checkpointing for OpenMP Programspdf (1)

Slides

Checkpointing

  
Agenda
EWOMP '03 Part II: OMPlab
Wed, Sept 24, 2003Thu, Sept 25, 2003Fri, Sept 26, 2003
8:30 registration8:30 hands-on and demos8:30 hands-on

9:00 tutorials / presentations: Programming Environments and Tools I

9:00 OMPlab Programming Environment (Slides)
Dieter an Mey

9:10 Sun OpenMP Programming Environment (Slides)
Brian Wiley

9:40 IBM OpenMP Programming Environment (Slides)
Luiz Derose

10:10 Fujitsu OpenMP Programming Environment (Slides)
Matthijs van Waveren

10:40 coffee break10:30 coffee break10:30 coffee break

11:10 short tutorials / presentations: Programming Environments and Tools II

11:10 Intel Thread Checker
Sanjiv Shah, Grant Haab

11:40 "Experiment with Cluster-enabled OpenMP: OpenMP for Software DSM System on PC Clusters" (Slides)
Mitsuhisa Sato, Yoshinori Ojima

11:55 KOJAK - A Tool Set for Automatic Performance Analysis of Parallel Programs (Slides)
Bernd Mohr, Felix Wolf

12:10 The llc language and its implementation (Slides)
Antonio J. Dorta, Jose Rodríguez,
Casiano Rodríguez, Francisco de Sande

12:25 Paraver (Slides)
Eduard Ayguade

12:40 An Implementation of the POMP Performance Monitoring Interface (Slides)
Luiz DeRose

12:55 TotalView Parallel Debugger (Slides)
Dieter an Mey

11:00
hands-on and demos

12:00 plenum: status reports, final discussion, wrap-up
13:10 lunch13:00 lunch
visit computer museum
13:00 lunch

14:30 short presentations: Applications

14:30 Finite Element Programming in C++ using OpenMP (Slides)
Jörg Multhoff

14:40 Nested Parallelism when Differentiating OpenMP Programs
H. Martin Bücker, Arno Rasch

14:50 Intervals and OpenMP: Towards an Efficient Parallel Results-Verifying Nonlinear Solver (Slides)
Thomas Beelitz, Christian H. Bischof, Bruno Lang

15:00 Hybrid Parallelization of the CFD Code PANTA (Slides)
Hans Thermann, Bernd Wickerath, Daniel Grates, Stephan Schmidt, Dieter an Mey, Chrstian Terboven

15:10 The Thermoflow60 Finite-Element Program (Slides 1, Slides 2)
Ulrich Wepler, Dieter an Mey, Thomas Haarmann, Alexander Spiegel, Wolfgang Koschel

15:20 OpenMP parallelization of agent-based models (Slides)
Federico Massaioli, Massimo Bernaschi

15:30 Gaussion '03
Alistair Rendell

15:40 Blocking Application-level Checkpointing for OpenMP Programs (Slides)
Greg Bronevetsky

15:15 hands-on and demos14:30 hands-on, "clean-up"
15:40 coffee break15:30 coffee break16:00 end of EWOMP'03
16:20 hands-on16:00 hands-on and demos
17:30 plenum
 
18:30 end of day19:00 end of day  
 20:00 social dinner, restaurant "Kazan" 

The presentations and the plenum sessions will take place in the lecture room (Hörsaal), coffee breaks in front of the lecture room, and the hands-on session in the PC rooms (CIP I + CIP III).

There will be WLAN access, and a few places with twisted pair access in the lecture room and in the PC rooms, such that you can use your own laptops for the lab, if you want.

During this workshop our schedule is quite flexible. There is room for parallel discussion groups and demonstrations etc.. Any proposal is welcome.

  
Sponsors

   Thank you
Our main sponsor is Sun Microsystems
Sun is the CCC's major provider of supercomputing machinery
Sun
OpenMPThe OpenMP Architecture Review Board (ARB)
Pallas, Germany
Pallas is the CCC's major provider of third party parallel programming tools
Pallas
  
Contact
      Dieter an Mey
        phone: +49-241-8024377 (mobile within the Center)

        mobile: +49-175-2609782 (Sept. 21 - 26 only)

        email:
ewomp03@rz.rwth-aachen.de
 

last modified: September 16, 2003
  • Keine Stichwörter