Metrics for Characterizing Complexity of Network Traffic

We propose the study of different entropy metrics as measures of complexity of network traffic. Based on examples with synthetic data as well as different measurement traces we show that these metrics provide a new way of characterizing and studying traffic behaviour. Especially multiscale entropy analysis appears to be a powerful tool, capable of uncovering rich structures on different time-scales from different traffic measurements. We also discuss potential uses of these tools in design and validation of traffic models. Examples of potential application areas for these metrics to be studied in the future include traffic characterization, classification and anomaly detection.