Workload characterization and performance evaluation in a research environment

This paper describes the process of bench-marking the diverse research environment that constitutes the workload of VAX/VMS at the University Analysis and Control Center at North Carolina State University. The benchmarking process began with a study of the system load and performance characteristics over the six-month period from January to June of 1981. Statistics were compiled on the number of active users, CPU usage by individual accounts, and peak load periods. Individual users were interviewed to determine the nature and major computing characteristics of the research they were conducting on VAX. Information from all sources was compiled to produce a benchmark that closely paralleled actual system activity. An analytic model was introduced and used in conjunction with the benchmark data and hardware characteristics to derive performance measures for the system. Comparisons with measured system performance were conducted to demonstrate the accuracy of the model. The model was then employed to predict performance as the system workload was increased, to suggest improvements for the system, and to examine the effects of those improvements.