Showing papers on "Benchmark (computing) published in 1983"

Performance of a modular interactive data analysis system (MIDAS)

Abstract: A processor cluster, part of a multiprocessor system named MIDAS (modular interactive data analysis system), has been constructed and tested. The architecture permits considerable flexibility in organizing the processing elements for different applications. The current tests involved 8 general CPUs from commercial computers, 2 special purpose pipelined processors and a specially designed communications system. Results on a variety of programs indicated that the cluster performs from 8 to 16 times faster than a standard computer with an identical CPU. The range represents the effect of differing CPU and I/O requirements-ranging from CPU intensive to I/O intensive. A benchmark test indicated that the cluster performed at approximately 85percent the speed of the CDC 7600. Plans for further cluster enhancements and multicluster operation are discussed. 5 references.

Processing power on the IBM personal computer

Abstract: In the last 20 years there has been nearly a six-order-of-magnitude increase in computing power, and it is predicted that another five-to-six-orders-of-magnitude increase will occur by the year 2000 [1]. Although supercomputers such as the Cray-1 and Cyber 205 can perform certain vector operations with burst rates of up to 100 million floating-point operations per second (Mflops) [2], it is remarkable that a popular and affordable personal computer is only three-orders-of-magnitude slower. Indeed, with suitable software, the IBM Personal Computer (PC) represents a significant advance in usable memory size and floating-point computing power when compared with previous popular microcomputers.It is purpose of this paper to discuss several floating-point compute-intensive benchmark programs which have been run on the IBM PC. These programs utilize the Intel 8087 Numeric Data Processor (NDP), and when coded in FORTH, can approach the NDP maximum compute rate of 0.05 Mflops.

Experiment in benchmarking database systems

Abstract: This paper describes and analyzes an interactive benchmark of the INGRES database system, which we have recently performed at the Los Alamos National Laboratory. The benchmark includes single-user runs and an extensive set of multi-user runs composed by synthesizing workloads of increasing complexity. Basic estimates were obtained for the response type of a range of interactive queries and for several other measures of the database management system performance.

Benchmarks for a computer system for nasa's shuttle procedures simulator

Abstract: The plans for the Shuttle Procedures Simulator called for more computational power than was available in order to meet future requirements. The three computer systems selected as candidates for meeting these requirements were Univac 1 lOO/8O, Univac 1 1 OO/4O and a conf iguration of ten PerkinElmer 8/32's. The performance of these systems was measured in terms of their ability t o execute a real-time flight simulation. Two benchmarks were used for this purpose. One was a single processor benchmark called KOPBM and the second was a multiprocessor benchmark derived from a simulation software load actually used for training astronauts. KOPBM indicated that a Univac 1100/80 processor was 1.70 times as fast as a Univac 1100/40 processor and that a Perkin-Elmer 8/32 was 0.48 times as fast as a Univac 1100/40 processor. The multiprocessor benchmark, however, showed that a Univac 1100/80 system was 1.18 times as fast as a Univac 1100/40 system with the same number of processors. Thus, the results of the uniprocessor benchmark and the multiprocessor benchmark differed s ignificantly. Subsequent to analysis of the results, the difference in the results of the two benchmarks was attributed to the difference in the memory architecture of the Univac systems and to the computational c haracteristics of the real-time flight simulation.