Change detection schemes are used in numerous applications, ranging from industrial quality control over component failure in telecommunications networks to edge detection in images. Next to general schemes solving the theoretical problem, several application-specific solutions were developed. In this paper, we look at the problem from a network engineerpsilas point of view, more specifically related to building network traffic models. We first describe an automated version of a commonly used technique. Next, we introduce a novel change detection scheme that is able to classify a network trace based on the stationarity of its rate, variance or any higher order moment. This scheme employs a discrete traffic representation. Its use can be generalized to detecting and localizing changes in any given time series.
展开▼