Computer Performance Evaluation of Individual Workloads on the ASD CDC Cyber System.

This paper describes computer performance evaluation and measurement techniques applied to a Control Data Corporation Cyber 74 system, located at Wright-Patterson AFB, Ohio. The performance of a computer system is influenced by its hardware configuration, system software, man-machine interaction, and application program workload. The focus of this investigation is to determine how application workload affects performance. Application workload is measured by the magnitude of job demands placed on the system hardware resources for example, CPU and I-O time, and central memory used. System accounting data provides resource usage values as well as related delay times for each job. Workload is then characterized by statistics for each job class and by total job statistics. The grouping of jobs into classes is done using clustering, a multivariate data analysis technique. Batch turnaround time is then analyzed for its dependence on workload. The methodology used and the results obtained provide performance conclusions and a background for further research. Author