High Performance Computing Workshop:
Sun Application Tuning Seminar

Date: Tuesday May 17 - Friday May 20, 2005
Location: IMM - DTU, building 321, room 033
Time: Tuesday - Thursday 09:00 - 17:00
Friday: 09:00 - 12:30
Speakers: Ruud van der Pas (Sun Microsystems) ,
Jonas Edberg (Sun Microsystems)

Introduction
Have you ever wanted to know more how to optimize a technical, scientific application for a modern RISC microprocessor? Are you interested in learning about parallelization? Now is the time to learn!
We will focus on serial program tuning, as well as tuning shared and distributed memory parallel applications using automatic parallelization, OpenMP resp. MPI. No background knowledge in these topics is assumed. We will cover the basics as well.
The general philosophy of the workshop is to build up understanding of key concepts that are relevant to obtain good application performance. Once this is achieved, it is much easier to use the development environment in the best possible way.
The Sun compilers, the Sun performance analyzer and the Sun HPC ClusterTools environment (MPI) will be covered in detail. It will be shown how these tools can be used to get optimal performance out of UltraSPARC IV and AMD Opteron based Sun workstations and servers.

Participants

Attendees should be comfortable with C or Fortran programming and interested in learning more about the technical details of application tuning. Although there is no special coverage of C++ and the examples are in Fortran and/or C, C++ programmers will certainly benefit from this course as well.

Prepared lab exercises will be made available to participants. These exercises have been selected to demonstrate features discussed in the presentations. Participants who prefer to work on their own application(s) instead will be welcome to do so. A suitable preparation of Makefiles and small plus medium sized data sets are of course desirable in this case.

The workshop language will be English.

The speakers
Ruud van der Pas ...

has a degree in numerical mathematics and physics. Since 1985, he has been involved with High Performance and Technical Computing. After having worked for 4 years as a consultant in numerical analysis and high-end computing at the University of Utrecht, and after positions at Convex and SGI, Ruud works for Sun Microsystems since 1998. He is a senior staff engineer in Sun's Scalable Systems Group and is actively involved in application tuning. He provides performance consultancy to customers and works with several engineering groups within Sun to enhance current and future products. Ruud will be the main speaker of parts I and II of the workshop

Jonas Edberg ...

has studied applied mechanics and received a Ph.D. degree from Luleå University of Technology in 1996 on "Numerical Modeling of Hot Rolling" using and developing finite element analysis (FEM). He now works with High Performance Technical Computing support within Sun Service PTS (Product Technical Support) in EMEA which provides world-wide customer supports for Sun HPC ClusterTools, Compiler/Development Tools and Sun ONE Grid Engine Software. Before joining Sun in 1998, he was a project leader and senior research scientist at Mefos which is a Industrial research institute for the metallurgical and metalworking industry. He worked as a collaborator with Lawrence Livermore laboratories on the development of the finite element codes Nike, Dyna and Topaz. Jonas will be the main speaker of part III of the workshop.

The Agenda
The full agenda is available as a PDF file.
The following topics will be covered in detail:
  •  Part I: Tuning Serial Applications
    •  The memory hierarchy in RISC cache based systems
    •  Serial optimization techniques
    •  The following Sun Studio components
      •  The Fortran and C compilers
      •  The Performance Library
      •  The Performance Analyzer
    •  Microprocessor and server architectures
    •  Serial optimization techniques
    •  Modulo Scheduling
    •  Case studies serial optimization
  •  Part II: Shared Memory Parallelization
    •  Introduction into parallelization
    •  Shared memory parallelization
      •  Data dependency analysis 
      •  Automatic Parallelization
      •  Explicit Shared Memory Parallelization with OpenMP
      •  Tuning OpenMP applications
      •  OpenMP Case Studies
  •  Part III: Distributed Memory Parallelization using MPI
    •  Introduction into explicit message passing programming using the Message Passing Interface (MPI)
    •  Sun HPC ClusterTools 5 overview
    •  Build & execution environment for MPI applications (Cluster Runtime Environment CRE)
    •  Debugging MPI codes on Sun using Prism
    •  Sun MPI implementation
    •  Runtime consideration and tuning
    •  Optimizing MPI communication
    •  Sun Scalable Scientific Subroutine Library (Sun S3L)
    •  MPI performance analysis

The Costs

The seminar is organized by the HPC Center at DTU in cooperation with and Sun Microsystems. There is no seminar fee. All other costs (e.g. travel, hotel, and consumptions) are at your own expenses.

Registration
Registration is necessary, and works on a first come, first serve basis. Please send an e-mail to Bernd Dammann, IMM DTU to register. It is also possible to register for part I, II and III individually.
Deadline: May 11, 2005.

Please note that you will receive a confirmation of your registration before or latest on May 11, 2005.

Course Material and Links

Contact

Bernd Dammann
Tel.: +45 45 25 33 71
E-mail: bd (at) imm.dtu.dk