Distributed client/server architectures are the technological backbone of today's data centres. A usage-based allocation of infrastructure costs to business processes or users is often not possible as the necessary resource consumption measurements incur too much overhead. Reinhard Brandl proposes a method to derive estimates for the expected resource consumption of customer-oriented services during standard load tests. This facilitates the determination of usage-based cost allocation keys significantly. He implements the concept in a software tool kit and evaluates it successfully in a set of experiments with multi-tier database applications. In particular, he uses the determined consumption estimates as input parameters for Queuing Network Models which lead to highly accurate performance predictions. Finally, he analyzes how the method can be integrated into existing IT processes at the BMW Group.