Performance analysis of high-speed computers

The choice of a computer system depends on many factors of which one of the most important criterion is the performance. In scientific/engineering environments where high-speed (mini)supercomputers are widely used, performance evaluation process is very complex due to the multidimensional nature of a computer´s performance. Many of these computers are able to do scalar, vector and parallel processing. In this report different performance evaluation and analysis methods for high-speed computers have been developed. Attention has been paid to the compilers, program development, scalar, vector and parallel floating point computation and I/O. A method how to calculate single performance numbers has been presented and with this method the overall performances of computers can be estimated. A throughput test has been developed to measure the multiprogramming support of the operating system and hardware as well as the overheads in context switching, paging and swapping. These methods have been used to evaluate in practice the performances of some recently announced high-speed computers: Alliant FX/80, Ardent Titan, Convex C2, ETA-10P, FPS Model 500, and Stellar GS1000 and GS2000. Most of the results have been compared to those of a conventional scalar computer, VAX 8700.