Newsgroups: comp.parallel
From: meuer@rz.uni-mannheim.de (Hans Werner Meuer)
Subject: MPI and HPF in Practice /Tutorial
Organization: Rechenzentrum Uni Mannheim
Date: 29 May 1995 15:25:01 GMT
Message-ID: <3qcp0d$229@usenet.srv.cis.pitt.edu>

(This information is also accessible together with a registration  
form via -----> URL:      
http://parallel.rz.uni-mannheim.de/sc/sc95.html )


Invitation:

The Board of the "Verein zur Fvrderung der wissenschaftlichen  
Weiterbildung an der Universitdt Mannheim e. V." invites you to the  
Tutorial MPI and HPF in Practice

Chairman :  Hans-Martin Wacker, DLR Kvln

Lecturers:  Jack Dongarra, University of Tennessee
            Chuck Koelbel, Rice University

Date :      June 26 - 27, 1995

Location :  Wartburg Hotel Mannheim, F4, 4


Fee:
                Academic        DM  870.00
                Regular         DM 1250.00
               
                Fees includes tutorial notes, coffee, lunches, social  
                event
                Members of the Fvrderverein receive a 10% discount



Registration :  Please register by June 2, 1995 (with enclosed     
                registration form or via WWW). For cancellation after                
                June 9, 1995 - should  a substitute not be provided -             
                a 50% withdrawal fee will becharged, after June 16,
                1995 the full registration fee is due.
                You will be billed after registration.

Information :   The Supercomputer Seminar with main focus on  
                Applications,Architectures, and Trends will take   
                place from June 22 to 24,1995
               
                Two additional Tutorials will be held from June 26 to  
                27, 1995 (both in German):
                
                WWW f|r kommerzielle Anwender und potentielle
                Informationsanbieter im Internet
               
                SMP-Systeme - Architekturen und Anwendungen

                Please ask for information!


In case of questions please contact:
Angela Jvck-Regut, Tel: ++49 621 292-5233
email: joeck@rz.uni-mannheim.de or
Werner Huff-H|bner, Tel: ++49 621 292-1432
email: verein@rummelplatz.uni-mannheim.de

Rechenzentrum Universitdt Mannheim
68131 Mannheim / GERMANY
FAX: ++49 621 292-5012,
email: meuer@rz.uni-mannheim.de

Dr.-Ing. Herbert Gassert
Chairman of the Board
Augustaanlage 33
D-68165 Mannheim

------------------------------

About the Tutorial
------------------


Background:

The break through for parallel computers expected by computer  
specialists could not be reached in 1994. The number of installations  
of parallel computers for scientific technical use did increase  
significantly, but almost all newly installed parallel computers were  
obtained by universities and research institutions to be used mainly  
for research on parallel processing. Very few computers are employed  
to produce important application programs. Especially the industry  
whose purchasing policy is vital seems to be very reluctant when it  
comes to use parallel computers mainly because until today there is  
no sufficient application software available for parallel
computers. Even software available on the market for parallel  
computers often is of inferior quality, the calculations in many  
cases differ compared to the original version of the same software  
for computers of classical architectures. The reason is that until  
recently no standardization for parallel computers applied and  
software had to be especially designed for a certain computer  
architecture. To demonstrate performance was mostly the goal for the  
developed codes. Thus developed software was not portable and of  
limited quality. Because of this the software platform generally used  
for parallel computers is quite small limiting the market for  
parallel computers very much. So in 1995 too some parallel computer  
manufacturers will have to face financial difficulties.

Necessary for the break through of parallel computers is a sufficient  
software platform. With the availability of the standards Message  
Passing Interface (MPI) and High Performance FORTRAN (HPF) now for  
the first time efficient interfaces are available to the user  
permitting a development of performant and portable software fitting  
the quality requirements of today.


Goal of the Tutorial:

The main emphasis will be on the standard tools for parallel  
processing available today and especially on MPI and HPF. The goal  
will be to give the attendants of the tutorial enough knowledge to be  
put in the position to develop software on their own which can be  
used for parallel computers of different architectures. Mainly the  
standpoints of performance, portability and software quality will be  
reinforced.
During the introduction of the tutorial the parallel computers  
available today with their architectures will be shown and compared.  
Especially different benchmarks will be looked at and explained as  
well as their underlying software methods judged.
The question which architecture or architectures will have the  
greatest chance to push through on the market will be discussed at  
the end of each day - especially in viewpoint to the impact on the  
software development.

Two of the leading experts and developers in the field of MPI and  
HPF, Jack Dongarra and Chuck Koelbel, both having developed and  
worked actively on these standards, will give the interested  
specialist in Europe and particulary in Germany the extraordinary  
chance to reach the growingly important State of the Art during the  
10th anniversary of the "Mannheim Supercomputing Days".

The number of participants is limited, making it possible to ask  
questions - the speakers are open to questions and ideas during the  
entire tutorial.


Target Group:

   Engineers, natural scientists, mathematicians, computer  
   specialists with experience in the field of software development  
   for conventional computers like workstations, mainframes and  
   vector computers wanting to develop professional software for
   parallel computers which in regard to performance,  
   portability and software quality fit the State of the Art.

   Technically oriented employees of computer manufacturers who want  
   to acquire sound knowledge in the area of parallel computing in  
   order to assess the present and future scientific-technical market  
   and to better evaluate the resulting needs of customers.

   Employees of software houses whose job it is to develop software  
   for parallel computers in the future.

   All persons interested to gain knowledge during a two day compact  
   courseabout the status, the potential, and the developing  
   computing from the viewpoint of an experienced user. You will meet
   specialists in the area of parallel computers like Jack Dongarra  
   and Chuck Koelbel, who had an essential influence in the      
   standardization parallel computing.



--------------------------

Program Monday, June 26, 1995



8.15            Welcome and Introduction
                Hans-Martin Wacker



I. Introduction to Parallel Computing

Jack Dongarra

8.30            Architecture of Parallel Computers
                - Shared Memory Architecture
                - Local Memory Architecture
                - Virtual Shared Memory Architecture
                - Examples of Existing Parallel Computers
                - Introduction to Parallel Programming
                - Shared Memory Parallelization
                - Local Memory Parallelization
                - Message Passing

10.00           Coffee and Communication Break


II. Portable High Performance Software

Jack Dongarra

10.30           Message Passing Interface - MPI
                - Definition and Status of Standardization
                - Usage and Programming
                - Status of the Implementation for Existing Parallel  
                  Computers

12.00           Lunch

13.30           Performance of Parallel Programs
                - Linpack Benchmark
                - NAS Parallel Benchmark
                - ParkBench
                - Portability of High Performance Software
                - Quality Aspects of Parallel Software

15.00           Coffee and Communication Break

15.30           Libraries
                - LAPACK
                - ScalAPACK
                - BLACS

                National High Performance Software Exchange

III. Discussion

17.00           Final Discussion with Jack Dongarra

19.00           Social Event



------------------------------

Program Tuesday, June 27, 1995

IV. Introduction to Automatic Parallelization

Chuck Koelbel

8.30            Automatic Parallelization Techniques
                - Coarse Grain and Fine Grain Parallelism
                - Status of Automatic Parallelization Techniques
                - Efficiency of Automatic Parallelization
                - FORTRAN 90

10.00           Coffee and Communication Break

V. High Performance FORTRAN (HPF)

Chuck Koelbel

10.30           HPF - Features and Compilers
                - Introduction
                - Data Parallel Features
                - Data Mapping Features
                - HPF Compilers

12.00           Lunch

13.30           Using HPF on SMP- and MPP-Systems
                - Linear Algebra: Gaussian Elimination
                - Iterative Methods: Jacobi and Conjugate Gradient
                - Sparse Methods: Relaxation on Irregular Meshes

15.00           Coffee and Communication Break


15.30           Performance Scalability and Future of HPF
                - Performance and Scalability of HPF-Programs
                - Portability of HPF-Programs
                - Software Engineering and Quality Aspects
                  of HPF compared to Message Passing Software


VI. Discussion

17.00           Final Discussion with Chuck Koelbel


18.00           End of Tutorial

-----------------------------------
Lecturers
---------

Jack Dongarra is Distinguished Professor of the University of  
Tennessee, Department of Computer Science and Distinguished Scientist  
of the Oak Ridge National Laboratory, Mathematical Science Section.  
Dongarra has a worldwide reputation as the originator of the famous  
LINPACK Benchmark. This benchmark measures the performance of various  
computers ranging from PCs to Supercomputers. At present he directs a  
group, called the Innovative Computer Laboratory (ICL), which is part  
of the Computer Science Department at the University of Tennessee and  
is engaged in research in various areas of high-performance  
computing: BLACS, BLAS, HeNCE, LAPACK, MPI, NHSE, Parkbench, PVM3,  
ScalAPACK, Templates, Report,XNETLIB and TOP500.
Jack Dongarra received a Ph.D. in Applied Mathematics from the  
University of New Mexico in 1980. He has published numerous articles,  
papers, reports and technical memoranda. Some recent books include  
"MPI: A Message-Passing Interface Standard", some recent papers  
include: "An Introduction to the MPI Standard",University of  
Tennessee Technical report CS-95-274, January 1995.



Charles (Chuck) Koelbel is one of the leading American scientists in  
the field of high performance computing and parallel computing. He  
received his Ph.D. in Computer Science from Purdue University in  
1990. Since that time he has been a Research Scientist at the Center  
for Research on Parallel Computation at Rice University. His research  
interests center on mapping programs onto distributed memory  
multiprocessors, in particular the compilation of irregular  
algorithms for these machines.
Since coming to Rice he has worked closely with Ken Kennedy and  
others to apply his methods to the Fortran D compiler project. In  
addition to his research responsibilities, Dr. Koelbel served in 1992  
and 1993 as executive director of the High Performance Fortran (HPF)  
Forum, a group of academic, industrial, and government organizations  
defining extensions to the Fortran language for use on
scalable parallel machines. Koelbel continues to be active with that  
group. He published "The High Performance Fortran Handbook", MIT  
Press, MA, 1993 and ten Technical Reports on HPF.


Hans-Martin Wacker, Director of Data Processing and Information  
Technology at DLR (the German NASA), is Chairman of the Tutorial. Dr.  
Wacker has worked in different positions for Universities and  
Research Centers more than 30 years.


 
---------------------------------------------------------------------
Hans-Werner Meuer			e-mail:  meuer@rz.uni-mannheim.de
Computing Center			phone :  ++49 621 292 5258
University of Mannheim		fax   :  ++49 621 292 5012
L15,16          	     
D-68131 Mannheim             
URL:  http://www.uni-mannheim.de/members/rum/meuer.html


---------------------------------------------------------------------
 

