In this paper, an on-line change detection algorithm for resource allocation in service-oriented systems is presented. The change detection is made basing on a dissimilarity measure between two estimated probability distributions. In our approach we take advantage of the fact that streams of requests in service-oriented systems can be modeled by non-homogenous Poisson processes. Thus, for Bhattacharyya distance measure and Kullback-Leibler divergence analytical expressions can be given. At the end of the paper a simulation study is presented. The aim of the simulation is to demonstrate an effect of applying adaptive approach in resource allocation problem.