STATUS REPORT Sc? (1@’Rq» Lu? Ul __ V? 211 — Q CERN Participation in the GP MIMD ESPRIT Project

Total Page:16

File Type:pdf, Size:1020Kb

STATUS REPORT Sc? (1@’Rq» Lu? Ul __ V? 211 — Q CERN Participation in the GP MIMD ESPRIT Project CERN LIBRARIES, GENEVA CERN/DRDU 9 1-12 Danc/M5 llltlllllllllulll II\)Illll|lI)I|l)H)|lllllI|ll)) afch , ;, — P A $(100000690 <_ fri · \ xg ·-J ’ STATUS REPORT Sc? (1@’rQ» lu? Ul __ V? 211 — Q CERN participation in the GP MIMD ESPRIT project P.C. Burkimsher, R.W. Dobinson ' , A. King, B. Martin, and I.M. Willers, CERN. J—L. Pages, University of Lausanne A. Schneider, University of Geneva. In collaboration with INMOS (UK), lead partner. Meiko (UK), Parsys (UK), Parsytec (D) and Telmat (F), main partners. and Grupo APD (E), INESC (P), Siemens (D), SICS (S), Southampton University (UK), IRISA (F) OCR Output spokesman z JMU Introduction A proposal was submitted to ESPRIT in January 1990 by INMOS and four European manufacturers of Transputer based computers. The aim of the proposal is to develop high performance parallel processing computers using a new generation of Transputers. ESPRIT is the EEC supported European Strategic Programme for Research and Development in Information Technology. The technical, manpower and budget details of the project are contained in the technical annex [l] of the official contract that has been signed by the lead and main project partners with the EEC. CERN has taken part in the preparatory work for the GP MIMD ESPRIT proposal and must now decide on the continuation of this work within the framework of the approved project. Transputer Technology Today's microprocessors make available on a single chip a significant fraction of the computing performance of a traditional mainframe. A recent trend has been to construct very powerful computing systems of interconnected microprocessors: up to hundreds or even thousands of processor nodes, each with local memory, joined by a fast switching network. Such machines are called MIMD (Multiple Instruction, Multiple Data) computers. Programs written for such systems can achieve high performance by treating different parts of a problem concurrently on multiple processors which communicate via the exchange of messages. Making multiple processors work together on a common problem has hitherto often involved major efforts in both·hardware and software. The Transputer, Fig. l, is a novel type of microprocessor, a basic building block explicitly designed to allow connection to other Transputers. Large assemblies of Transputers and switches are more straightforward to assemble and program than systems using traditional microprocessors, see Fig. 2. The basic building blocks used in the GP-MIMD project are a new INMOS microprocessor, the Hl Transputer, and the ClO4 network routing chip which are being developed with the help of funding from the EEC. INMOS claims that each Hl will have a sustained performance in excess of 60 MIPS and 20 MFLOPS, which compares favourably with present state—of—the—art American RISC processors available from Intel, MIPS and SUN. Each of the four on-chip links of the H1 have a data transfer speed of l0 Mbytes/second. The new Hl Transputer with its higher computational performance and enhanced networking capabilities will form the basis for the construction of general purpose computers as well as advanced dedicated systems. GP MIMD Project Outline A three year project, commencing on January lst 1991, was given initial approval by ESPRIT in May 1990. The proposed role of CERN in the project is as follows: l. CERN will mount high energy physics applications using new generation Transputer hardware and software. Both on—line and off-line applications will be targetted. We will pursue methods and tools for programming parallel systems: 2. CERN will develop high performance input/output interfaces between Transputer based machines, data acquisition equipment and high speed data recording devices. This will include the development of a VME interface and a fiber optics connection to the GP MIMD machine. OCR Output A grant would be made by ESPRIT to CERN for the hiring of new personnel (8 man years) and equipment. The total value of this grant is approximately 1.4M sf. EEC funding is conditional on CERN providing complementary resources of 690k sf and 15.75 man years, to be spread over the three year duration of the project. The overall aims of the GP MIMD project, as approved by the EEC, are as follows: l. To develop in Europe an open standard architecture for message passing MIMD machines which provide computational and communications capabilities for up to 1000 processing nodes: 2. To provide light weight software for real time computing and a distributed UNIX-like operating system for general purpose computing; 3. To construct prototype computers and demonstrate the machine's capabilities across a range of applications: 4. To develop new VLSI components, in line with the recent EEC Open Microprocessor Initiative. INMOS will design, by the end of the project, a processor with a target performance of 400 MIPS and 100 MFLOPS on a single chip, with Gigabit/second links. Relevance to High Energy Physics During the last two years, CERN has gained experience in exploiting the present generation of T80O Transputers and has established good contacts with INMOS, Meiko and Southampton University which have lead to our involvement in this project. The TSOO Transputers have proved themselves to be promising components for building on-line systems for high energy physics experiments, in UA6 and ZEUS [2]. An interface between Fastbus and Transputers has been constructed at CERN in collaboration with INFN Rome and is under test. Several European institutes are using parallel computer systems to achieve high computing performance, at relatively low cost, for a range of off—1ine applications. This important new trend in parallel computing will, in our view, be of direct relevance to future experiments and accelerators. It will be of importance in real time acquisition and control systems, for example, in event building and higher level trigger systems. It could also provide significant computing power for simulation and event processing, capacity which could be made available more economically than using present traditional mainframes. Future high energy physics computing needs are predicted to continue the past strong upward trend. Software is recognised to be a particularly challenging and important aspect. Work has already begun to explore methods and design tools for designing parallel real time systems and for running some of CERN's large Fortran programs on a GP-MIMD machine [3]. Valuable experience would be gained by mounting applications on the evolving industry standard UNIX parallel processing platforms. To successfully understand and exploit parallel computing systems, it is important to learn to apply parallel programming techniques to solve high energy physics problems. As a result of our continuing collaboration with the H1 chip manufacturer, INMOS, and the current major European parallel computer manufacturers, Meiko (UK), Parsys (UK), Parsytec (D) and Telmat (F), considerable expertise and experience would be gained in the area of parallel processing, both in hardware and software. OCR Output References [1] GP MIMD P5404, Technical Annex for ESPRIT General Purpose MIMD Machines. [2] Transputers in Particle Physics Experiments, R.w. Dobinson, J-L. Pages and J.C. Vermeulen, to be published in Particle world. [3] Parallelizing HEP FORTRAN Programs for the GP—MIMD Distributed Memory Machine, A. Schneider and A. King, to be presented at Computing in High Energy Physics 1991, Tsukuba City, Japan, March 11-15, 1991. OCR Output Floating point processor Main processor On chip memory Links to other Transputers External memory interface Figure 1 Transputer block diagram OCR Output H1 H1 H1 H1 C104 Switch H1 H1 H1 H1 Figure 2. Network of H1 Transpute-rs linked via a C104 switch.
Recommended publications
  • The Helios Operating System
    The Helios Operating System PERIHELION SOFTWARE LTD May 1991 COPYRIGHT This document Copyright c 1991, Perihelion Software Limited. All rights reserved. This document may not, in whole or in part be copied, photocopied, reproduced, translated, or reduced to any electronic medium or machine readable form without prior consent in writing from Perihelion Software Limited, The Maltings, Charlton Road, Shepton Mallet, Somerset BA4 5QE. UK. Printed in the UK. Acknowledgements The Helios Parallel Operating System was written by members of the He- lios group at Perihelion Software Limited (Paul Beskeen, Nick Clifton, Alan Cosslett, Craig Faasen, Nick Garnett, Tim King, Jon Powell, Alex Schuilen- burg, Martyn Tovey and Bart Veer), and was edited by Ian Davies. The Unix compatibility library described in chapter 5, Compatibility,im- plements functions which are largely compatible with the Posix standard in- terfaces. The library does not include the entire range of functions provided by the Posix standard, because some standard functions require memory man- agement or, for various reasons, cannot be implemented on a multi-processor system. The reader is therefore referred to IEEE Std 1003.1-1988, IEEE Stan- dard Portable Operating System Interface for Computer Environments, which is available from the IEEE Service Center, 445 Hoes Lane, P.O. Box 1331, Pis- cataway, NJ 08855-1331, USA. It can also be obtained by telephoning USA (201) 9811393. The Helios software is available for multi-processor systems hosted by a wide range of computer types. Information on how to obtain copies of the Helios software is available from Distributed Software Limited, The Maltings, Charlton Road, Shepton Mallet, Somerset BA4 5QE, UK (Telephone: 0749 344345).
    [Show full text]
  • ISRA VISION at a Glance
    ISRA VISION Initiating Coverage 18 February 2019 Better than the human eye – Growth story is expected continue In a highly fragmented market environment ISRA VISION ranks Target price (EUR) 36 Share price (EUR) 28 among the leading players for machine vision technologies with a strong focus on software solutions. Driven by market trends as well as R&D efforts, we expect ISRA’s growth story to continue and Forecast changes forecast a revenue growth CAGR of 10.4% until FY 2022/23e. EBIT % 2019e 2020e 2021e is expected to increase by 10.6% during this period. In addition to organic growth, M&A driven growth is very likely. We initiate the Revenues NM NM NM EBITDA NM NM NM coverage of ISRA VISION with Buy and a TP of EUR 36.00. EBIT adj NM NM NM EPS reported NM NM NM EPS adj NM NM NM ISRA VISION among the technology leaders Source: Pareto ISRA is one of the leading providers for machine vision solutions with specialisation in 3D machine vision. The company’s Ticker ISRG.DE, ISR GR competences are mainly concentrated around the internally Sector Industrials developed software application. Future demand is driven by more Shares fully diluted (m) 21.9 Market cap (EURm) 605 complexity and miniaturization, quality requirements and entering Net debt (EURm) -18 into new markets. Customer benefits are cost savings, process Minority interests (EURm) 2 optimization, higher flexibility and a ROI of clearly <1 year. Enterprise value 19e (EURm) 590 Free float (%) 65 Growth story should continue - 5yrs growth CAGR of 10.4% The well diversified customer base in various end-markets helps ISRA to reduce cyclicality.
    [Show full text]
  • Uva-DARE (Digital Academic Repository)
    UvA-DARE (Digital Academic Repository) Scalable distributed data structures for database management Karlsson, S.J. Publication date 2000 Document Version Final published version Link to publication Citation for published version (APA): Karlsson, S. J. (2000). Scalable distributed data structures for database management. General rights It is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s), other than for strictly personal, individual use, unless the work is under an open content license (like Creative Commons). Disclaimer/Complaints regulations If you believe that digital publication of certain material infringes any of your rights or (privacy) interests, please let the Library know, stating your reasons. In case of a legitimate complaint, the Library will make the material inaccessible and/or remove it from the website. Please Ask the Library: https://uba.uva.nl/en/contact, or a letter to: Library of the University of Amsterdam, Secretariat, Singel 425, 1012 WP Amsterdam, The Netherlands. You will be contacted as soon as possible. UvA-DARE is a service provided by the library of the University of Amsterdam (https://dare.uva.nl) Download date:11 Oct 2021 atabasee Management Scalablee Distributed Data Structures s for r Databasee Management Academischh Proefschrift terr verkrijging van de graad van doctor aann de Universiteit van Amsterdam opp gezag van de Rector Magnificus prof.. dr. J. J. M. Franse tenn overstaan van een door het collegee voor promoties ingestelde commissie, inn het openbaar te verdedigen inn de Aula der Universiteit opp donderdag 14 december 2000, te 11.00 uur doorr Jonas S Karlsson geborenn te Enköping, Sweden Promotor:: Prof.
    [Show full text]
  • ALTERNATING-DIRECTION LINE-RELAXATION METHODS on MULTICOMPUTERS* J)RN HOFHAUS and ERIC E VAN DE VELDE$ Abstract
    SIAM J. ScI. COMPUT. () 1996 Society for Industrial and Applied Mathematics Vol. 17, No. 2, pp. 454-478, March 1996 010 ALTERNATING-DIRECTION LINE-RELAXATION METHODS ON MULTICOMPUTERS* J)RN HOFHAUS AND ERIC E VAN DE VELDE$ Abstract. We study the multicomputer performance of a three-dimensional Navier-Stokes solver based on alternating-direction line-relaxation methods. We compare several multicomputer implementations, each of which combines a particular line-relaxation method and a particular distributed block-tridiagonal solver. In our experiments, the problem size was determined by resolution requirements of the application. As a result, the granularity of the computations of our study is finer than is customary in the performance analysis of concurrent block-tridiagonal solvers. Our best results were obtained with a modified half-Gauss-Seidel line-relaxation method implemented by means of a new iterative block-tridiagonal solver that is developed here. Most computations were performed on the Intel Touchstone Delta, but we also used the Intel Paragon XP]S, the Parsytec SC-256, and the Fujitsu S-600 for comparison. Key words. Navier-Stokes equations, concurrency, parallelism, block-tridiagonal systems, tridiagonal systems, ADI, alternating directions AMS subject classifications. 65M20, 65N40, 65Y05, 76C05, 76M20 1. Introduction. When using alternating-direction line-relaxation methods for systems of partial-differential equations discretized on a rectangular grid, one must solve many block- tridiagonal systems of linear equations in every relaxation step. This type of computation surfaces in many applications. In our work, we faced it when parallelizing a highly vectorized solver for the three-dimensional, unsteady, and incompressible Navier-Stokes equations [4] for use on multicomputers.
    [Show full text]
  • Implementation of an Environment for Monte Carlo Simulation of Fully 3-D Positron Tomography on a High-Performance Parallel Platform
    Parallel Computing 24 (1998) 1523±1536 Implementation of an environment for Monte Carlo simulation of fully 3-D positron tomography on a high-performance parallel platform Habib Zaidi *, Claire Labbe, Christian Morel Division of Nuclear Medicine, Geneva University Hospital CH-1211 Geneva 4, Switzerland Received 15 January 1998; received in revised form 15 April 1998 Abstract 12/02/1998 ext-2001-007 This paper describes the implementation of the Eidolon Monte Carlo program designed to simulate fully three-dimensional (3-D) cylindrical positron tomographs on a MIMD parallel architecture. The original code was written in Objective-C and developed under the NeXT- STEP development environment. Dierent steps involved in porting the software on a parallel architecture based on PowerPC 604 processors running under AIX 4.1 are presented. Basic aspects and strategies of running Monte Carlo calculations on parallel computers are de- scribed. A linear decrease of the computing time was achieved with the number of computing nodes. The improved time performances resulting from parallelisation of the Monte Carlo calculations makes it an attractive tool for modelling photon transport in 3-D positron tomography. The parallelisation paradigm used in this work is independent from the chosen parallel architecture. Ó 1998 Elsevier Science B.V. All rights reserved. Keywords: Monte Carlo simulation; Positron emission tomography; Random numbers; Image recon- struction; Parallel computer 1. Introduction Positron emission tomography (PET) is a well-established imaging modality which provides physicians with information about the body's chemistry not available through any other procedure. Three-dimensional (3-D) PET provides qualitative and * Corresponding author. E-mail: [email protected] 0167-8191/98/$ ± see front matter Ó 1998 Elsevier Science B.V.
    [Show full text]
  • Logp: Towards a Realistic Model of Parallel Computation
    LogP: Towards a Realistic Model of Parallel Computation David Culler, Richard Karp,y David Patterson, Abhijit Sahay, Klaus Erik Schauser, Eunice Santos, Ramesh Subramonian, and Thorsten von Eicken Computer Science Division, University of California, Berkeley Abstract A vast body of theoretical research has focused either on overly simplistic models of parallel computation, notably the PRAM, or overly specific models that have few representatives in the real world. Both kinds of models encourage exploitation of formal loopholes, rather than rewarding development of techniques that yield performance across a range of current and future parallel machines. This paper offers a new parallel machine model, called LogP, that reflects the critical technology trends underlying parallel computers. It is intended to serve as a basis for developing fast, portable parallel algorithms and to offer guidelines to machine designers. Such a model must strike a balance between detail and simplicity in order to reveal important bottlenecks without making analysis of interesting problems intractable. The model is based on four parameters that specify abstractly the computing bandwidth, the communi- cation bandwidth, the communication delay, and the efficiency of coupling communication and computation. Portable parallel algorithms typically adapt to the machine configuration, in terms of these parameters. The utility of the model is demonstrated through examples that are implemented on the CM-5. Keywords: massively parallel processors, parallel models, complexity analysis, parallel algo- rithms, PRAM 1 Introduction Our goal is to develop a model of parallel computation that will serve as a basis for the design and analysis of fast, portable parallel algorithms, i.e., algorithms that can be implemented effectively on a wide variety of current and future parallel machines.
    [Show full text]
  • Uva-DARE (Digital Academic Repository)
    UvA-DARE (Digital Academic Repository) A Communication Kernel for Parallel Programming Support on a Massively Parallel Processor System Overeinder, B.J.; Vesseur, J.J.J.; van der Linden, F.; Sloot, P.M.A. Publication date 1995 Published in Proceedings of the Workshop on Parallel Programming and Computation (ZEUS'95) and the 4th Nordic Transputer Conference (NTUG'95) Link to publication Citation for published version (APA): Overeinder, B. J., Vesseur, J. J. J., van der Linden, F., & Sloot, P. M. A. (1995). A Communication Kernel for Parallel Programming Support on a Massively Parallel Processor System. In L. Finmo, & P. Fritzon (Eds.), Proceedings of the Workshop on Parallel Programming and Computation (ZEUS'95) and the 4th Nordic Transputer Conference (NTUG'95) (pp. 259-266). IOS Press. General rights It is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s), other than for strictly personal, individual use, unless the work is under an open content license (like Creative Commons). Disclaimer/Complaints regulations If you believe that digital publication of certain material infringes any of your rights or (privacy) interests, please let the Library know, stating your reasons. In case of a legitimate complaint, the Library will make the material inaccessible and/or remove it from the website. Please Ask the Library: https://uba.uva.nl/en/contact, or a letter to: Library of the University of Amsterdam, Secretariat, Singel 425, 1012 WP Amsterdam, The Netherlands. You will be contacted as soon as possible. UvA-DARE is a service provided by the library of the University of Amsterdam (https://dare.uva.nl) Download date:01 Oct 2021 A Communication Kernel for Parallel Programming Support on a Massively Parallel Processor System B.
    [Show full text]
  • TNMOC-Newsletter-Q1-2015.Pdf
    InSync Members’ News TNMoC will be hosting a Members’ Open Day on Saturday 28th March 2015, 10:00am for 10:30am. This is your opportunity to share your ideas regarding the future of the membership programme. Important changes are afoot both in the museum and in the Members’ organisation. We want you to be amongst the first to hear about them. In particular, we are keen to consult more with Members and to enable you to become more involved in the Museum. We’d like to share our plans with you during the open day and hear your views, so we hope that you will be able to attend. The programme for the day will include an exclusive, in-depth look at the Cambridge University EDSAC reconstruction. If you’ve been to the museum recently you’ll have seen that the project is progressing very well. Members of the EDSAC team will be there to tell you all about it. Volunteers and the Museum’s management team will be on hand to chat with Members about the museum and its ever-changing range of exhibits. If there is sufficient interest we will arrange one or more specialist guided tours for Members. Full details will be sent to you nearer the time. If you have any questions you are invited to contact John Linford at [email protected]. In the meantime, please put the Open Day in your diary. We look forward to seeing you! Connecting with the Museum of Computing Keep up to date with the latest information from the museum by visiting our various social media and website presences.
    [Show full text]
  • A High-Performance, Portable Implementation of the MPI Message
    A HighPerformance Portable Implementation of the MPI Message Passing Interface Standard William Gropp Ewing Lusk Mathematics and Computer Science Division Argonne National Lab oratory Nathan Doss Anthony Skjellum Department of Computer Science NSF Engineering Research Center for CFS y Mississippi State University Abstract MPI Message Passing Interface is a sp ecication for a standard library for message passing that was dened by the MPI Forum a broadly based group of parallel com puter vendors library writers and applications sp ecialists Multiple implementations of MPI havebeendevelop ed In this pap er we describ e MPICH unique among existing implementations in its design goal of combining p ortability with high p erformance We do cument its p ortability and p erformance and describ e the architecture bywhichthese features are simultaneously achieved We also discuss the set of to ols that accompany the free distribution of MPICH which constitute the b eginnings of a p ortable parallel programming environment A pro ject of this scop e inevitably imparts lessons ab out parallel computing the sp ecication b eing followed the currenthardware and software environment for parallel computing and pro ject management we describ e those we have learned Finallywe discuss future developments for MPICH including those nec essary to accommo date extensions to the MPI Standard now b eing contemplated by the MPI Forum Intro duction The messagepassing mo del of parallel computation has emerged as an expressive e cient and wellundersto o d paradigm
    [Show full text]
  • The Helios File Server Installation Information
    The Helios File Server Installation Information PERIHELION SOFTWARE LTD November 1990 Copyright This document Copyright °c 1991, Perihelion Software Limited. All rights re- served. This document may not, in whole or in part be copied, photocopied, reproduced, translated, or reduced to any electronic medium or machine read- able form without prior consent in writing from Perihelion Software Limited, The Maltings, Charlton Road, Shepton Mallet, Somerset BA4 5QE, UK. This manual is Edition 2.0, November 1990. Acorn and ARM are trademarks of Acorn Computers Ltd. Amiga is a registered trademark of Commodore-Amiga, Inc. Apple is a registered trademark of Apple Computers, Inc. Atari is a trademark of the Atari Corporation. Commodore is a registered trademark of Commodore Electronics, Ltd. Ethernet is a trademark of Xerox Corporation. Helios is a trademark of Perihelion Software Limited. IBM is a registered trademark of International Business Machines, Inc. Inmos, occam, T414, T425 and T800 are trademarks of the Inmos group of companies. Intel and iPSC are registered trademarks of Intel Corporation. Macintosh is a trademark of Apple Computers, Inc. Meiko and Cesius are trademarks of Meiko Limited. Motorola is a trademark of Motorola, Inc. MS-DOS is a registered trademark of The Microsoft Corporation. Parsytec, Paracom and SuperCluster are trademarks of Parsytec GmbH. POSIX refers to the standard de¯ned by IEEE Standard 1003.1-1988; Posix refers to the library calls based upon this standard. Transtech is a trademark of Transtech Devices Ltd. Sun, SunOs and SunView are trademarks of Sun Microsystems. Telmat and T.Node are trademarks of Telmat Informatique. Unix is a registered trademark of AT&T.
    [Show full text]
  • Versatility of Bulk Synchronous Parallel Computing: from the Heterogeneous Cluster to the System on Chip
    Dissertation Versatility of Bulk Synchronous Parallel Computing: From the Heterogeneous Cluster to the System on Chip Olaf Bonorden Fakultät für Elektrotechnik, Informatik und Mathematik Institut für Informatik und Heinz Nixdorf Institut February 2008 ii Zusammenfassung Der Bedarf an Rechenleistung in den Wissenschaften und der Industrie steigt ständig an. Bei der Entwicklung schnellerer Prozessoren gelangt man allerdings an physi- kalische Grenzen, so dass zur weiteren Leistungssteigerung sehr viele Prozessoren zusammen als Parallelcomputer benutzt werden müssen. Zur Nutzung dieser Super- computer ist es wichtig, allgemeine Modelle und Werkzeuge zu haben, die es erlau- ben, unabhängig von der speziellen Hardware effiziente Algorithmen zu entwickeln und zu implementieren. In dieser Dissertation werden Modelle für parallele Systeme vorgestellt, ein Über- blick über Algorithmen für diese Modelle gegeben und effiziente Implementierungen für verschiedene Architekturen entwickelt. Der Schwerpunkt liegt dabei auf der Fami- lie der Bulk Synchronous Parallel Modelle, da diese die Entwicklung portabler, aber trotzdem effizienter paralleler Programme erlauben. Für die Implementierungen werden zwei Architekturen betrachtet: ein On-Chip- Parallelcomputer und Workstation-Cluster. Im Bereich des On-Chip-Systems zeigt die Arbeit, wie das benutzte Modell die Entwicklung applikationsunabhängiger, effizien- ter, paralleler Systeme unterstützen kann. Die Workstation-Cluster, auf denen nur freie Rechenkapazitäten genutzt werden dürfen, stehen auf der anderen Seite des Spektrums paralleler Systeme. Sie unter- scheiden sich vom On-Chip-System nicht nur durch viel größere Latenzen, geringe- re Kommunikationsbandbreite und größeren Arbeitsspeicher, sie können auch hete- rogen sein, d. h., verschiedene Computertypen enthalten. Hierdurch und durch die variable, sich ständig ändernde, nutzbare Rechenkapazität der einzelnen Knoten er- geben sich besondere Herausforderungen, z. B.
    [Show full text]
  • A Parallel Implementation of Genetic Programming That Achieves Super-Linear Performance
    A PARALLEL IMPLEMENTATION OF GENETIC PROGRAMMING THAT ACHIEVES SUPER-LINEAR PERFORMANCE David Andre John R. Koza Visiting Scholar, Computer Science Computer Science Department Department Stanford University Stanford University EMAIL: [email protected] EMAIL: [email protected] WWW: http://www-cs- WWW: http://www- faculty.stanford.edu/~koza/ leland.stanford.edu/~phred/ Abstract: This paper describes the successful parallel implementation of genetic programming on a network of processing nodes using the transputer architecture. With this approach, researchers of genetic algorithms and genetic programming can acquire computing power that is intermediate between the power of currently available workstations and that of supercomputers at intermediate cost. This approach is illustrated by a comparison of the computational effort required to solve a benchmark problem. Because of the decoupled character of genetic programming, our approach achieved a nearly linear speed up from parallelization. In addition, for the best choice of parameters tested, the use of subpopulations delivered a super linear speed-up in terms of the ability of the algorithm to solve the problem. Several examples are also presented where the parallel genetic programming system evolved solutions that are competitive with human performance on the same problem. 1 Introduction Amenability to parallelization is an appealing feature of genetic algorithms, evolutionary programming, evolution strategies, classifier systems, and genetic programming [1][2][3][4]. The probability of success in applying the genetic algorithm to a particular problem often depends on the adequacy of the size of the population in relation to the difficulty of the problem. Although many moderately sized problems can be solved by the genetic algorithm using currently available single-processor workstations, more substantial problems usually require larger populations.
    [Show full text]