Now in its 10th year, the twice-yearly TOP500 list of supercomputers serves as a "Who's Who" in the field of High Performance Computing (HPC). The TOP500 list was started in 1993 as a project to compile and publish twice a year a list of the most powerful supercomputers in the world. But it is more than just a ranking system and serves as major source of information to analyze trends in HPC. The list of manufacturers active in this market segment has changed continuously and quite dramatically during the 10 year history of this project. And while the architectures of the systems in the list have also seen a constant change, it turns out that the overall increase in the performance levels recorded is rather smooth and predictable. HPC performance levels grow exponentially. The most important single factor for this growth is - of course the increase of processor performance described by Moore's Law. However, the TOP500 list clearly illustrates that HPC performance has actually outpaced Moore's Law, due to the increasing processor numbers in HPC systems. On the other hand changes in computer architecture make it more and more of a challenge to achieve high performance efficiencies in the Linpack benchmark used to rank the 500 systems. With knowledge and effort, the Linpack benchmark can still be implemented in very efficient ways as recently demonstrated by a new implementation developed at the U.S. Department of Energy's National Energy Research Scientific Computing Center for their 6,656 processor IBM SP system.

Since the very beginning of the International Supercomputer Conference (ISC) series, one important focus has been the analysis of the supercomputer market. For 20 years, statistics about this marketplace have been published at ISC. Initially these were based on simple market surveys and since 1993, they are based on the TOP500 project, which has become the accepted standard for such data. We take the occasion of the 20th anniversary of ISC to combine and extend several previously published articles based on these data. We analyze our methodologies for collecting our statistics and illustrate the major developments, trends and changes in the High Performance Computing (HPC) marketplace and the supercomputer market since the introduction of the Cray 1 system.

The problem of ranking the utility of supercomputer systems arises frequently in situations such as procurements and other types of evaluations of architectures. It is also central for any general ranking of supercomputers such as the Top500. Rankings of computer systems have traditionally solely focused on performance aspects. In recent years restrictions due to power and space requirements of large supercomputers have become very noticeable, which has increased the importance of including these factors in generalized rankings. In this paper we present an overview of the current practice for utility metrics and analyze their shortcomings. We then present and discuss in detail a new concept for a parameterized utility metric for supercomputers, which is based on effective performance, available memory size, actual power consumption, and (if desired) the floor space required for supercomputers. This metric is designed and proposed for augmenting the current Top500 ranking.

With the increasing gap between processor, memory, and interconnect speed, the performances of scientific applications on high performance computing systems have become dominated by the ability to move global data. However, many benchmarks in the field of high performance computing focus on measuring the achieved CPU speed in MFlop/s. In this paper, we introduced a novel benchmark, Apex-Map, which focuses on global data movement and measures how fast global data can be fed into computational units. Apex-Map is a parameterized synthetic performance probe and integrates concepts for tempo-ral and spatial locality into its design. By measuring the Apex-Map performance for a whole range of temporal and spatial localities performance surfaces can be generated which can be used to study the characteristics of the computational platforms and which are useful for performance comparison. Results on a vector platform and two superscalar platforms clearly reflect the design differences be-tween these two types of systems.

The memory wall and global data movement have become the dominant performance bottleneck for many scientific applications. New characterizations of data access streams and related benchmarks to measure their performances are therefore needed to compare HPC systems, software, and programming paradigms effectively. In this paper, we introduce a novel global data access benchmark, Apex-Map. It is a parameterized synthetic performance probe and integrates concepts for temporal and spatial locality into its design. We measured Apex-Map performance for a whole range of temporal and spatial localities on sev-eral advanced processors and parallel computing platforms and use the generated performance surfaces for performance comparisons and to study the characteristics of these different architectures. We demonstrate that the results of Apex-Map clearly reflect many specific characteristics of the used systems. We also show the utility of Apex-Map for analyzing the performance effects of three leading parallel programming models and demonstrate their relative merits.

APEX-Map is a synthetic performance probe for characterizing access behavior to global data structures. It is designed around concepts for temporal locality and spatial locality and can be used to analyze the performance characteristics of a computing platform across a whole range of localities. It can also be used to compare performance across different architectures or different programming paradigms. In this paper, we present the results of APEX-Map on the Cray X1 and use them to analyze this specific computing platform. We have implemented APEX-Map using MPI, SHMEM, and UPC and compare these three programming paradigms on the Cray X1. We are also going to discuss some performance problems we have found regarding the current MPI library implementation.

Since 1993 we compile and publish twice a year a list of the most powerful supercomputers in the world. In this article we analyse some major trends in High-Performance Computing (HPC) based on the quantitative data gathered over the years in this TOP500 project (see www.top500.org for a complete access to all data). We start with an overview of the project, including the motivation and rationale behind it. We present the results of the TOP500 project as of November 2002 focusing on the changes over time with respect to the performance levels, the overall dynamics in this market, the manufacturers active in this market-segment, the architectures, and technologies used for these systems, and the major centers using such systems.