Joanne L. Martin

The Perfect Club Benchmarks: Effective Performance Evaluation of Supercomputers

This report presents a methodology for measuring the performance of supercomputers. It includes 13 Fortran programs that total over 50,000 lines of source code. They represent applications in several areas of engi neering and scientific computing, and in many cases the codes are currently being used by computational re search and development groups. We also present the PERFECT Fortran standard, a set of guidelines that allow portability to several types of machines. Furthermore, we present some performance measures and a method ology for recording and sharing results among diverse users on different machines. The results presented in this paper should not be used to compare machines, except in a preliminary sense. Rather, they are presented to show how the methodology has been applied, and to encourage others to join us in this effort. The results should be regarded as the first step toward our objec tive, which is to develop a publicly accessible data base of performance information of this type.

A Retrospective

gaining access to the most advanced computing technology for the first time as a result of that program. No other refereed journals were devoted specifically to supercomputing, and there were as yet no major conferences sponsored by professional organizations to provide forums for interaction among the diverse community of supercomputer users. It was clear that computational science would soon

SP2 system architecture

Scalable parallel systems are increasingly being used today to address existing and emerging application areas that require performance levels significantly beyond what symmetric multiprocessors are capable of providing. These areas include traditional technical computing applications, commercial computing applications such as decision support and transaction processing, and emerging areas such as “grand challenge” applications, digital libraries, and video production and distribution. The IBM SP2™ is a general-purpose scalable parallel system designed to address a wide range of these applications. This paper gives an overview of the architecture and structure of SP2, discusses the rationale for the significant system design decisions that were made, indicates the extent to which key objectives were met, and identifies future system challenges and advanced technology development areas.

Web Services: Promises and Compromises

Much of web services’ initial promise will be realized via integration within the enterprise, either with legacy applications or new business processes that span organizational silos. Enterprises need organizational structures that support this new paradigm.

Supercomputer Performance Evaluation: Status and Directions

Despite the increasing application of supercomputers to important problems in a wide spectrum of scientific disciplines, the process of measuring, evaluating, and predicting their performance is imprecise at best. To advance the science of supercomputer performance evaluation, measurements must be made in the context of defined models of architectures and applications. We present an overview of current practice in supercomputer performance evaluation, describe methods for characterizing applications and architectures, and point toward some approaches for accomplishing their pairing.

Special Section International Parallel Processing Projects: A Software Perspective

Industrial and academic researchers provide a glimpse into efforts to develop a faster, more powerful computer through methods beyond purely technological advances.

Performance Analysis of Vectorized Nuclear Codes On a Facom Vp-100 At the Japan Atomic Energy Research Institute

We have evaluated the performance of vectorized nuclear codes used at the Japan Atomic Energy Institute. Forty nu clear codes were restructured for vectori zation, with an average vectorization of 83%. The vector to scalar ratio for modi fied code performance is about 5. Vectori zation improved performance in most cases, but certain codes, including LOCA and Monte Carlo codes, required consid erable effort to vectorize, without high performance gains. We discuss the bar riers to achieving high performance, and the need to consider applications charac teristics when evaluating system performance.

An Invitation To Participate

ernment policies have placed supercomputers within reach of university researchers. This will enable the solution of computation-intensive basic research problems that previously were insurmountable, thus changing radically the patterns of many of the world’s leading scientific efforts. In this inaugural issue, Erich Bloch, Director of the National Science Foundation, provides an overview of the plans for NSF-sponsored supercomputing centers in the U.S. and of the progress to date. In subsequent issues, Supercomputer Applications will highlight the activities of particular supercomputing centers in the international scientific

The Missing Pieces

duced without loss of clarity or interpretation. For the diggers, a bit of synchronization can double productivity. For the pianists, there’s no such magic. Neither effort is resource limited ; rather, the complexities and definitions of the tasks impose different constraints on achievable speed. Major architectural advances have put those of us involved in high-speed computing in the position of seeking the magic that will allow any computational problem to be calculated at peak rates, even if peak rates require the consistent use of vectors or parallel

Book Reviews: High-Speed Computing: Scientific Applications and Algorithm Design: Robert Wilhelmson, ed. 1987. Urbana and Chicago: University of Illinois Press. Approx. 250 pp., illustrated.

Also in the first section, although somewhat out of place, is Stephen Wolfram’s article, &dquo;Cellular Automaton Supercomputing.&dquo; The techniques described by Wolfram, and their conformability to massively parallel processors like the Connection Machine, may be harbingers of the next generation of algorithms for supercomputers. Particularly noteworthy contributions in subsequent sections include John Kogut’s concise and lucid description of quantum chromodynamics (QCD), an article by Sangiovanni-Vincentelli and Webber entitled &dquo;Com-