Automatic Workload Characterization

Workload characterization is an important and basic step for performance modeling and capacity planning. Traditional approaches for workload characterization often require sophisticated and time-consuming user assistance. In this paper we describe a number of algorithms for automatically classifying workloads and for determining what resources a workload used even when the operating system does not provide the data. The algorithms use statistics to find the processes and other resources that worked together to complete each type of transaction. Generally we will define a transaction as the work supporting a business activity, such as an entry in a log file. For Web-based commerce, business activity data is logged and readily accessible. We also show how automatic workload characterization can be done without a log file.