On the construction of a representative synthetic workload
Communications of the ACM
| Hi-index | 0.00 |
This paper reports on the design and analysis of a statistical experiment conducted on a “live” job stream to determine the effect of segment size used for storage allocation on the system performance. Performance measures selected are turnaround time, cost and CPU utilization. The experiment consists of one factor, the segment size, at five levels. Uncontrolled factors such as EXCP's (number of I/0 starts) and core usage are included as covariates in the analysis of variance. This experiment is part of a continuing activity of Measurement, Evaluation and Simulation. It is designed to provide data for improving performance incrementally. The results of the experiment provided an optimal segment size for the given classing/scheduling algorithm and core-layout. Design objectives and details of the analysis are also presented.