© 1981 by British Computer Society
The design of calibration experiments for synthetic jobs
Industrial Engineering Department, Texas A & M University, College Station, Texas, USA
The workload processed by a computer system can have a dramatic impact on the system's measured performance. A computer performance evaluation study conducted using empirical techniques requires an executable test workload. A test workload composed of synthetic jobs appears the most promising. Such a synthetic mix requires development of predictor equations which allow the setting of the parameters to produce the desired resource demand patterns.
A statistical methodology composed of experimental design techniques and regression analysis is proposed to aid in the development of prediction equations relating the resource descriptor variables and the synthetic job parameters. Two synthetic jobs are posed and the experimental design procedure is demonstrated with workload data from the Amdahl 470/V6 at Texas A & M University. Some factors related to validation of test workloads are presented.
Received July 1979. revised February 1980.
Industrial Engineering Department, Texas A & M University, College Station, Texas 77843, USA