In large-scale networks, such as the Internet, spatial-temporal correlations emerge from interactions among adaptive transport connections and from variations in user demands. Analyzing spatial-temporal characteristics of traffic in large-scale networks requires both a suitable analysis method and a means to reduce the amount of data that must be collected. In this chapter, we describe a novel technique that provides a useful way to observe network-wide congestion patterns shifting over time. To illustrate this technique and its potential promise, we report results from some simulation experiments, where we successfully identify network hotspots induced deliberately in a large-scale network and where we expose large-scale distributed denial-of-service (DDoS) attacks. Since observing the whole Internet in detail is impractical, we suggest a means to efficiently observe selective portions in detail, or to apply spatial aggregation to observe larger-scale networks with less detail. In either case, our analysis method lowers computing time requirements, while revealing shifting traffic patterns over both space and time.
