A large set of codes is used to construct various throughput models of the overall performance of any supercomputer system. The model is described with re sults from runs on CRAY-1S and Cyber 205 Class VI su percomputers. A unique reference throughput is estab lished on a CRAY-1S with precise vector performance and CPU utilization. The sensitivity of the modeling to variations in the code mix is discussed. The performance of multiprocessor, parallel supercomputers is estimated.
References
1.
Dongarra, J.1988. Technical Memo 23. Argonne, Ill.: Argonne National Laboratory, April 19.
2.
Gentzsch, W.1988. Comparison of supercomputers and minisupercomputers for computational fluid dynamics. J. Supercomput. 2(4):63-71.
3.
Temperton, C.1987. Assembler language fast Fourier transforms. (available in the MAGEV library at JVNC and from K. Fong and B. McNamara at the NMFECC).
