SASE VI and the statistical analyses of series in events in computer systems

Abstract

We describe recent results in the development of methodology of the statistical analysis of univariate series of events (point processes) and give some references to applications in the analysis and evaluation of computer system performance data. In addition, we describe the SASE VI program which has been developed to implement the methodology for the statistical analysis of series of events in the monograph on this subject by Cox and Lewis. Various subroutines perform, among other things, tests for monotone and cyclic trends, tests for renewal and Poisson processes and two different types of spectral analysis. The program can also be used to analyze any series of positive random variables such as counts of events in successive fired time intervals in a point process. It has been programmed in both FORTRAN and APL. Multivariate series of events (point processes) present a much more difficult task and the methodology for their analysis has only recently been developed in a perforce fairly tentative manner. Applications in the analysis of computer system data and neurophysiological data are given. One problem here is the need for new data analytic methods for the analysis of data when trying to build models, and the lack of simple models for non-normal, positive multivariate time series. Some starts in these directions are described. (Author)