🚀 go-pugleaf

Attached please find the titles, speakers and abstracts of the talks
that will take place during the High Performance Computing Day, part of
the SHARE Europe Anniversary meeting, to be held in The Hague, The
Netherlands, on Wednesday October 27, 1993.

There is a special one-day fee for those only interested in the HPC Day.

If you are interested in attending, please contact SHARE Europe HQ
(48, route des Acacias, CH 1227, Carouge, Switzerland, Phone
+41 22 3003775, Fax +41 22 3001119).

........................................................................

Session 3.1H  Time: 8:45-9:45

Title: An overview of HPC-systems

Speaker: Ad Emmen (SARA, Amsterdam)

Abstract:

  There are many different types of  architectures available when it
  comes to high-performance computing, ranging from traditional
  parallel vectorcomputers to message-passing systems with thousands
  of processors. Furthermore, people are clustering together powerfull
  workstations into one parallel system to get hihjer performance.

  The lecture will give an overview of the main architectures employed,
  try the classify the available (commercial) systems and comment on the
  worldwide distribution.

CV:

  Ad Emmen is Manager User support at SARA, the Dutch national
  supercomputer center. He is involved in high-performance computing
  since 1983 and is editor of several journals in the field, including
  "Supercomputer" and "Supercomputer - European Watch".

........................................................................

Session 3.2H  Time: 10:15-11:10

Title: SP1: The first benchmarks

Speaker: Joanne Martin (IBM, Kingston)

Abstract:

  The 9076 SP1 is IBM's first offering in the Scalable POWERparallel
  Series.  Based on RISC System/6000 processors, the SP1 is comprised
  of 8 to 64 processors and has a peak system performance of 1 to 8
  GFLOPS.

  In this talk, we will discuss the effective performance of the SP1
  based on a series of benchmarks, which includes synthetic routines to
  measure specific system components, computational kernels that
  demonstrates performance on commonly used routines, well known
  benchmarks for which performance data is expected in the
  supercomputing community (e.g. TPP, SPEC/Perfect), and full
  applications and production use.

CV:

  Joanne L. Martin, Ph.D. is a Senior Technical Staff Member and Manager,
  of Applications Technology and High-Performance Studies.

  Joanne joined IBM as a Research Staff Member at the T.J. Watson
  Research Center in November, 1984. She had received her Ph.D. in
  mathematics from The Johns Hopkins University in 1981 and conducted
  research in performance evaluation for supercomputers at the Los Alamos
  National Laboratory prior to joining IBM.

  In May, 1991, Joanne was appointed manager of her present department,
  and is responsible for performance analysis and modeling for IBM's
  Scalable Parallel systems, within the newly created POWER Parallel
  business unit. In January, 1993 Joanne was appointed a member of IBM's
  Senior Technical Staff.

  Joanne has maintained active participation in the external scientific
  community, as founding editor-in-chief of the International Journal of
  Supercomputer Applications, serving as the General Chair of
  Supercomputing 90, being sought as an advisor to the Dept. of Energy
  and the National Science Foundation and she was named to Who's Who in
  Science and Engineering for 1992-1993.

........................................................................

Session 3.3H  Time: 11:15-12:10

Title: Discussion of customer requirements and scalable parallel future.

Speaker: Joanne Martin (IBM, Kingston)

Abstract:

  During this session Dr Joanne Martin address the requirements provided
  by Share Europe during Share SM 93.

  She will also discuss the possible evolution of the Scalable Parallel
  Systems.

........................................................................

Session 3.4H  Time: 14:00-14:45

Title: Overview of PVM

Speaker: John Zollweg (Cornell Theory Center)

Abstract:

  Parallel Virtual Machine (PVM) has become a very popular message-
  passing environment on many systems of workstations with TCP/IP
  communication. The standard PVM package will be briefly described.
  Some enhancements to this system have been made by IBM that allow one
  to run PVM over a point-to-point switch with low message latency.
  Results will be presented for some scientific applications that require
  frequent communication, showing the effects of the enhanced PVM.

CV:

  John Zollweg is Project Manager for the Strategic Applications Program
  in Cornell Theory Center. He has had extensive experience with parallel
  processing, including PVM, both as a researcher who used the Theory
  Center's platforms for many years (in chemical engineering), and now as
  a Theory Center staff member coordinating early use by researchers of
  the strategic platforms. John is now involved in a Joint Verification
  Study with IBM.  He has already ported his own research code to the
  SP1, and is co-chairing the SP1 Early User Workshop in June.

........................................................................

Session 3.5H  Time: 14:50-15:35

Title:  High Performance Fortran

Speaker: Clemens-August Thole (GMD-I1T)

Abstract:

  The architecture of high performance computers gets more complex. Large
  numbers of parallel nodes and non-uniform memory access times are
  challenges for compilers, which make efficient code generation for
  sequential Fortran 90 programs very difficult. In  order to generate
  better code compilers need additional information like the
  specification of

    - opportunities for parallel execution
    - distribution of data objects onto the memories of the nodes

  The High Performance Fortran Forum (HPFF) was founded as a coalition of
  about 40 industrial and academic groups (including most vendors of
  parallel computers), who had several meetings since March 1992 in order
  to define language extensions to Fortran 90 (High Performance Fortran).
  A final document was accepted in March 1993.

  HPF supports especially the mapping of data parallel program on
  parallel computers. Compiler directives specify the distribution of
  data objects. Some language extensions to Fortran 90 enhance the
  possibilities to write Fortran 90 programs in such a way that they can
  be executed in parallel.

  The presentation will give an overview about HPF and its features and
  will inform about the status of the initiative.

CV:

  Clemens-August Thole is member of the HPFF (High Performance Fortran
  Forum), that developed High Performance Fortran. He organized a group
  in Europe complementary to the HPFF in the US, and was instrumental in
  making HPF an international effort.

........................................................................

Session 3.6H  Time: 16:00-16:45

Title: MCNP4, a Parallel Monte Carlo Implementation on a Workstation Network

Speaker: Frank Schmitz (Kernforschungzentrum Karlsruhe)

Abstract:

  The Monte Carlo code MCNP4 has been implemented on a workstation
  network to allow parallel computing of Monte Carlo transport processes.
  This has been achieved by making use of the communication tool PVM
  (Parallel Virtual Machine) and introducing some changes in the MCNP4
  code. The PVM daemons and user libraries have been installed on
  different workstations to allow working on the same platform.
  Essential features of PVM and the structure of the parallelized MCNP4
  version are discussed in this paper. Experiences are described and
  problems are explained and solved with the extended version of MCNP.
  The efficiency of the parallelized MCNP4 is assessed for two realistic
  sample problems from the field of fusion neutronics. Compared with the
  fastest workstation in the network, a speed-up factor near five has
  been obtained by using a network of ten workstations, different in
  architecture and performance.

CV:

  Frank Schmitz has been working since seven years in the environment of
  supercomputing and is now looking for tools to use the workstations in
  the Nuclear Research Center. Vectorizing is done for many problems, now
  parallelizing will be the next job.

........................................................................

Session 3.7H  Time: 16:50-17:35

Title: SP1 compared to a workstation cluster: a user experience

Speaker: John Zollweg (Cornell Theory Center)

Abstract: Not available

CV:

  John Zollweg is Project Manager for the Strategic Applications Program
  in Cornell Theory Center. He has had extensive experience with parallel
  processing, including PVM, both as a researcher who used the Theory
  Center's platforms for many years (in chemical engineering), and now as
  a Theory Center staff member coordinating early use by researchers of
  the strategic platforms. John is now involved in a Joint Verification
  Study with IBM.  He has already ported his own research code to the
  SP1, and is co-chairing the SP1 Early User Workshop in June.

........................................................................

Other talks related to HPC during SHARE Europe AM93:

   Session 1.3C  Date: October 25  Time: 11:15-12:00
   Title: Highly Parallel DB Computing
   Speaker:

   Session 2.1D  Date: October 26  Time: 8:45-9:45
   Title: Is a parallel System/390 in Your Future?
   Speaker: Scott Loveland (IBM Poughkeepsie, USA)

   Session 2.2D  Date: October 26  Time: 10:15-11:10
   Title: Is a parallel System/390 in Your Future? (Session Continued)
   Speaker: Scott Loveland (IBM Poughkeepsie, USA)

   Session 2.3D  Date: October 26  Time: 11:15-12:10
   Title: Parallel Batch
   Speaker: Tom Monza (IBM Poughkeepsie, USA)

   Session 2.EI  Date: October 26  Time: 11:15-12:00
   Title: Inside the Human Brain with Parallel Processing
   Speaker: Dr. James A. Browm (IBM Santa Teresa, USA)