Skip to Main Content
Summary form only given. Many key tasks in computer network monitoring, such as load balancing and the detection of anomalies (e.g., DoS attacks, link failures, and routing loops), fundamentally require a whole-network perspective - that is, methods of 'spatial' traffic analysis. A key challenge to advances in this area is posed by the fact that the 'space' in question, deriving from a network, is not some Euclidean sub-space but rather a graph. We focus on the problem of developing analyses sensitive to 'spatial' scale and introduce a framework for graph-based wavelets. In conjunction, we also consider the important issue of characterizing the underlying 'spatial' auto-correlation structure.