Abstract |
---|
This paper presents a workload characterization for workstations, servers and WWW server. Twelve data sets built from standard UNIX tools and from access.log files are analyzed with three different time scales. We demonstrate that the workload of these resources is statistically self-similar in the periods of irregular activity and for the three time scales. We present the underlying mathematical and statistical properties of self-similarity and some tests used to characterize the dataset self-similarity.The paper concludes with a discussion about the results, the understanding of this phenomenon and the ways to deal with it. |
Contact |
Franck Cappello LRI, batiment 490,Universite Paris-sud,91 405 Orsay cedex,France fci@lri.fr |