Online error detection through observation of traffic self-similarity

The authors present a new and universally applicable approach to error detection in packet or cell communication networks. The error detection uses measured traffic load data. The advantage is that detection of an error in a low layer reduces the probability of an undetected error in higher layers, and makes time-costly error detection in higher layers unnecessary. Most systems use static traffic load thresholds for error detection. The authors present an approach which can achieve considerably higher sensitivity. Their basic idea is to exploit the property of self-similarity in network traffic. This analytical redundancy gives a reference behaviour of the network traffic load, which allows the detection of faulty behaviour in the real network traffic load. For the error detection, the validity of the given self-similar property is checked through a deviation indicator Q based on second-order properties of the time series´ distributions. This is a sufficient condition for normal (error-free) behaviour