Computing in Object-Oriented Parallel Environments: Second International Symposium, ISCOPE 98 Santa Fe, NM, USA, December 8–11, 1998 Proceedings

253 Pages·1998·3.244 MB·English

by  Masahiro Tatsumi, Akio Yamamoto (auth.), Denis Caromel, Rodney R. Oldehoeft, Marydell Tholburn (eds.)

Lecture Notes in Computer Science 1505
Editedby G.Goos,J. Hartmanisand J.van Leeuwen
3
Berlin
Heidelberg
NewYork
Barcelona
HongKong
London
Milan
Paris
Singapore
Tokyo
Denis Caromel Rodney R. Oldehoeft
Marydell Tholburn (Eds.)
Computing
in Object-Oriented
Parallel Environments
Second International Symposium, ISCOPE 98
Santa Fe, NM, USA, December 8-11, 1998
Proceedings
1 3
SeriesEditors
GerhardGoos,KarlsruheUniversity,Germany
JurisHartmanis,CornellUniversity,NY,USA
JanvanLeeuwen,UtrechtUniversity,TheNetherlands
VolumeEditors
DenisCaromel
UniversityofNice–INRIASophiaAntipolis
2004RoutedesLucioles,BP93,F-06902SophiaAntipolisCedex,France
E-mail:[email protected]
RodneyR.Oldehoeft
ColoradoStateUniversity,ComputerScienceDepartment
FortCollins,CO80523,USA
E-mail:[email protected]
MarydellTholburn
LosAlamosNationalLaboratory,CIC-8,MSB272
LosAlamos,NM87545,USA
E-mail:[email protected]
Cataloging-in-Publicationdataappliedfor
DieDeutscheBibliothek-CIP-Einheitsaufnahme
Computinginobject-orientedparallelenvironments:second
internationalsymposium;proceedings/ISCOPE98,SantaFe,NM,
December8-11,1998.DenisCaromel... (ed.).-Berlin;Heidelberg
;NewYork;Barcelona;HongKong;London;Milan;Paris;
Singapore;Tokyo:Springer,1998
(Lecturenotesincomputerscience;Vol.1505)
ISBN3-540-65387-2
CRSubjectClassification(1991):D,G.1-2,F.3
ISSN0302-9743
ISBN3-540-65387-2Springer-VerlagBerlinHeidelbergNewYork
Thisworkissubjecttocopyright.Allrightsarereserved,whetherthewholeorpartofthematerialis
concerned,specificallytherightsoftranslation,reprinting,re-useofillustrations,recitation,broadcasting,
reproductiononmicrofilmsorinanyotherway,andstorageindatabanks.Duplicationofthispublication
orpartsthereofispermittedonlyundertheprovisionsoftheGermanCopyrightLawofSeptember9,1965,
initscurrentversion,andpermissionforusemustalwaysbeobtainedfromSpringer-Verlag.Violationsare
liableforprosecutionundertheGermanCopyrightLaw.
(cid:1)c Springer-VerlagBerlinHeidelberg1998
PrintedinGermany
Typesetting:Camera-readybyauthor
SPIN10638994 06/3142–543210 Printedonacid-freepaper
Preface
This volume contains the Proceedings of the International Symposium on Com-
puting in Object-Oriented Parallel Environments (ISCOPE ’98), held at Santa
Fe, New Mexico, USA on December 8{11, 1998. ISCOPE is in its second year,1
and continues to grow both in attendance and in the diversity of the subjects
covered. ISCOPE’97 and its predecessor conferences focused more narrowly on
scienti(cid:12)c computing in the high-performance arena. ISCOPE ’98 retains this
emphasis, but has broadened to include discrete-event simulation, mobile com-
puting, and web-based metacomputing.
The ISCOPE ’98 Program Committee received 39 submissions, and accep-
ted 10 (26%) as Regular Papers, based on their excellent content, maturity of
development, and likelihood for widespread interest. These 10 are divided into
three technical categories.
Applications: The (cid:12)rst paper describes an approach to simulating advanced
nuclear power reactor designs that incorporates multiple local solution me-
thodsandanaturalextensiontoparallelexecution.Thesecondpaperdiscus-
ses a Time Warp simulation kernel that is highly con(cid:12)gurable and portable.
The third gives an account of the development of software for simulating
high-intensity charged particle beams in linear particle accelerators, based
on the POOMA framework, that shows performance considerably better
than an HPF version, along with good parallel speedup.
Runtime and Libraries: The (cid:12)rst paper in this category evaluates Java as a
language and system for high-performance numerical computing, exposing
some issues to face in language features and compilation strategies. The se-
cond describes using the Illinois Concert system to parallelize an adaptive
mesh re(cid:12)nement code, showing that a combination of aggressive compiler
optimizations and advanced run-time support can yield good parallel per-
formance for dynamic applications. The third paper presents a uni(cid:12)ed fra-
mework for building a numerical linear algebra library for dense and sparse
matrices, achieving high performance and minimizing architectural depen-
dencies. In the fourth paper, a parallel run-time substrate is presented that
supportsaglobaladdressingscheme,objectmobility,andautomaticmessage
forwarding for implementing adaptive applications on distributed-memory
machines.
Numerics and Algorithms:The(cid:12)rstpaperdescribesasoftwarepackageforpar-
titioningdataonstructuredgrids,supportinginherentandnewpartitioning
algorithms, and describes its use in two applications. The second describes
a family of multiple minimum degree algorithms to generate permutations
of large, sparse, symmetric matrices to minimize time and space required
in matrix factorization. The (cid:12)nal regular paper discusses two optimizing
transformations for numerical frameworks, one that reduces inter-processor
communication and another that improves cache utilization.
1 The ISCOPE’97 Proceedings are available from Springer as LNCS, Vol. 1343.
VI Preface
Inaddition,theProgramCommitteeselected15submissionsasShortPapers.
These papers were deemed to represent important work of a more specialized
nature or to describe projects that are still in development. The Short Papers
are divided into four technical categories.
Metacomputing: The (cid:12)rst paper presents a Java-based infrastructure to com-
bine web-based metacomputing with cluster-based parallel computing. The
second describes an experimental metacomputing system that is dynami-
cally recon(cid:12)gurable in its use of systems and networks, and also in its own
capabilities.Thethirdpaperoutlinesadistributedplatformtoeasethecom-
binationofheterogeneousnetworks,concentratingonthedesignofitskernel
software.Thefourthpaperpresentslanguageconstructsforthesimultaneous
creation of entire static object networks which have useful properties.
Frameworks and Run-time: The (cid:12)rst paper describes a class library for FIFO
queues that can be incorporated with Time Warp simulation mechanisms
and retain the advantages of inlined data structures and e(cid:14)cient state sa-
ving.Thesecondpaperpresentsathreadpro(cid:12)lingsystemthatiscognizantof
the underlying concurrent run-time environment. The third paper evaluates
a high-level, portable, multithreaded run-time system for supporting con-
current object-oriented languages. The fourth describes a run-time library
for data-parallel applications that covers a spectrum of parallel granulari-
ties, problem regularities and user-de(cid:12)ned data structures. The last paper
in this section describes the design and use of a component architecture for
large-scale simulations of scienti(cid:12)c problems, based in turn on the POOMA
framework.
Numerics and Algorithms:The(cid:12)rstpaperdiscussestheparallelizationandim-
plementation of Monte Carlo simulations for physical problems. The second
presents a parallel implementation of the dynamic recursion method for tri-
diagonalizing sparse matrices e(cid:14)ciently. The third discusses the design of
software for solving sparse, symmetric systems of linear equations by di-
rect methods. The fourth paper describes a template library of two-phase
container classes and communication primitives for parallel dynamic mesh
applications.
Arrays: The (cid:12)rst of two papers in this category describes the Blitz++ libr-
ary, meant to provide a base environment of vectors, arrays and matrices
for scienti(cid:12)c computing with C++. The second discusses the design of ar-
rays and expression evaluation strategies in the new POOMA II framework
development.
Thiscollectionof25papersrepresentsthestateoftheartinapplyingobject-
oriented methods to parallel computing. ISCOPE ’98 is truly international in
scope, with its 72 contributing authors representing 24 research institutions in
9 countries. The ISCOPE ’98 organizers are con(cid:12)dent that the reader will share
their excitement about this dynamic and important area of computer science
and applications research.
Attheendofthisvolume,theAuthorContactssectiondetailsthea(cid:14)liations,
postal addresses, and email addresses of all the proceedings authors.
PrefaceVII
ISCOPE ’98 is partially supported by the Mathematical, Information, and
Computational Sciences Division, O(cid:14)ce of Energy Research, U.S. Department
of Energy.
Steering Committee
Dennis Gannon, Indiana University
Denis Caromel, University of Nice{INRIA Sophia Antipolis
Yutaka Ishikawa, Real World Computing Partnership
John Reynders, Los Alamos National Lab
Satoshi Matsuoka, Tokyo Institute of Technology
Jo¨rg Nolte, German National Research Center for Information Technology
Organizing Chairs
Dennis Gannon, Indiana University, General Chair
Denis Caromel, University of Nice{INRIA Sophia Antipolis, Program
Yutaka Ishikawa, Real World Computing Partnership, Posters
John Reynders, Los Alamos National Lab, Workshops/BOF
Rodney R. Oldehoeft, Colorado State University, Proceedings
Marydell Tholburn, Los Alamos National Lab, Local Arrangements/Publicity
Program Committee
Ole Agesen, Sun Microsystems Labs, USA
Denis Caromel, University of Nice{INRIA Sophia Antipolis, France
Antonio Corradi, University of Bologna, Italy
Geo(cid:11)rey Fox, Northeast Parallel Architecture Center, Syracuse Univ., USA
Dennis Gannon, Indiana University, USA
Jean-Marc Geib, University of Lille, France
Andrew Grimshaw, University of Virginia, USA
Urs H¨olzle, University of California-Santa Barbara, USA
Yutaka Ishikawa, Real World Computing Partnership, Japan
Jean-Marc Jezequel, IRISA/CNRS, France
Pierre Kuonen, EPFL, Switzerland
Satoshi Matsuoka, Tokyo Institute of Technology, Japan
Jo¨rg Nolte, Institute GMD-FIRST, Germany
Rodney R. Oldehoeft, Colorado State University, USA
John Reynders, Los Alamos National Lab, USA
Wolfgang Schroeder-Preikschat, Magdeburg, GMD, Germany
Anthony Skjellum, Mississippi State University, USA
David F. Snelling, Fujitsu European Center for Information Technology, UK
VIII Preface
Kenjiro Taura, University of Tokyo, Japan
MaryDell Tholburn, Los Alamos National Lab, USA
Andrew L. Wendelborn, University of Adelaide, Australia
Russel Winder, King’s College London, UK
Additional External Referees
Cli(cid:11) Addison Sven van den Berghe Danilo Beuche
Lars Buettner P. Calegari Peter Chow
Paul Coddington George Crawford David Detlefs
Rossen Dimitrov Stephane Ecolivet Anto^nio A. Fro¨hlich
Robert George Matt Gleeson Abdelaziz Guerrouat
F. Guidec Fred(cid:19)eric Guyomarc’h Ute Haack
Peter Harrison Greg Henley Wai Ming Ho
Naoki Kobayashi Evelina Lamma Pascale Launay
Andrea Omicini Jean-Louis Pazat R. Radhakrishnan
Tomasz Radzik Y.S. Ramakrishna C. Stefanelli
F. Zambonelli
Table of Contents
Regular Papers
Applications
Object-Oriented Approach for an Iterative Calculation Method and Its
Parallelization with Domain Decomposition Method ::::::::::::::::::: 1
Masahiro Tatsumi, Akio Yamamoto
An Object-Oriented Time Warp Simulation Kernel :::::::::::::::::::: 13
Radharamanan Radhakrishnan, Dale E. Martin, Malolan Chetlur,
Dhananjai Madhava Rao, Philip A. Wilsey
Particle Beam Dynamics Simulations Using the POOMA Framework ::::: 25
William Humphrey, Robert Ryne, Timothy Cleland, Julian Cummings,
Salman Habib, Graham Mark, Ji Qiang
Runtime and Libraries
An Evaluation of Java for Numerical Computing::::::::::::::::::::::: 35
Brian Blount, Siddhartha Chatterjee
High-Level Parallel Programming of an Adaptive Mesh Application Using
the Illinois Concert System:::::::::::::::::::::::::::::::::::::::::: 47
Bishwaroop Ganguly, Andrew Chien
TheMatrixTemplateLibrary:AGenericProgrammingApproachtoHigh
Performance Numerical Linear Algebra ::::::::::::::::::::::::::::::: 59
Jeremy G. Siek, Andrew Lumsdaine
The Mobile Object Layer: A Run-Time Substrate for Mobile Adaptive
Computations ::::::::::::::::::::::::::::::::::::::::::::::::::::: 71
Nikos Chrisochoides, Kevin Barker, D(cid:19)emian Nave, Chris Hawblitzel
Numerics and Algorithms I
Software Tools for Partitioning Block-Structured Applications ::::::::::: 83
Jarmo Rantakokko
An Object-Oriented Collection of Minimum Degree Algorithms ::::::::: 95
Gary Kumfert, Alex Pothen
Optimizing Transformations of Stencil Operations for Parallel Object-
Oriented Scienti(cid:12)c Frameworks on Cache-Based Architectures ::::::::::: 107
Federico Bassetti, Kei Davis, Dan Quinlan