Measurement of flow burstiness by fractal technique

Abstract

Burstiness is an important network feature. Many studies apply crude methods using "burst" or flow size to diagnose bursty flows, but these approaches cannot reveal traffic details. Using the self-similarity of network flow, the study transforms the time-domain traffic into space-domain data, and adopts the variations of fractal dimensions to determine the variations of burstiness. The resulting variations of burstiness indicate the irregular situation of network. Experimental results show that the variation of the fractal dimension is proportional to the variation of burstiness. Thus, a variation that exceeds a pre-defined threshold indicates irregular traffic. Triggering a simple alarm upon detecting irregular traffic can ameliorate the following traffic. Therefore, the proposed method is an effective gauge for flow monitoring because it watches the fractal dimensions derived from network traffic. The proposed method can help network administrators monitor traffic in detail and better manage their networks.