Generating workload for ERP applications through end-user organization categorization using high level business operation data

Abstract

For software companies performance testing is an essential part of new application development. In this paper we present a performance engineering method that extracts the workload of an existing legacy ERP application with more than 1 million users and generates workload for a radically new version of the application. The workload is used to classify groups of end user organizations, i.e., enterprises whose customers are end users of the application, with unsupervised machine learning techniques. The method shows that (1) workload for new application testing and architecture validation can be generated from legacy application behavior, (2) end user organizations have significantly different usage patterns, and (3) for ERP applications, high-level operations, such as a salary calculations, provide a useful method for analyzing and generating workload, as opposed to for instance low level page views. The method is evaluated within a Dutch software company, where it is found to be accurate and effective for performance engineering.