Applications of Multivariate Statistical Techniques for Computer Performance Evaluation

摘要

In many situations the computer performance evaluation (CPE) analyst has collected an abundance of computer system performance data from the target system's accounting files and software monitors. Traditionally, regression analysis provided the primary means of examining CPE data sets, with the emphasis being on modeling specific workload and performance parameters. Multivariate analysis techniques provide the analyst with additional analysis tools for the examination of relationships, dimension, and structure of large amounts of data. This study examines possible CPE applications for four multivariate analysis techniques: Canonical Correlation, Factor Analysis, Discriminant Analysis, and Cluster Analysis. Also included in the study was the use of ordinary least squares regression modeling and ridge regression modeling, to exemplify the traditional problems encountered with use of regression analysis. Depending on the performance evaluation requirements, one or more of the multivariate techniques or ridge regression could be used to perform a preliminary or supplementary CPE data analysis. (Author)