The complexity of computer systems requires to consider the interaction of several workloads. The number of business and technical workloads potentially affecting a computer system is large, but only a limited number of them are usually required to properly model the system. In this paper, we discuss regression-based estimates of service times required for model parametrization and we focus on the selection of significant workloads. We present an experimental comparison, using real performance logs of a distributed enterprise application, illustrating the benefits of constrained and stepwise estimations over the traditional approach based on ordinary linear regression.
展开▼
