A New Algorithm for Long Flows Statistics—MGCBF

Abstract

Abstract. Long flows identification and characteristics analysis play more and more important role in modern traffic analysis because long flows take main traffic payload of network. Based on the flows length distribution and long flows characteristics of the Internet, this paper presents a novel long flows' counting and information maintenance algorithm called Multi-granularity Counting Bloom Filter (MGCBF). Using a little fix memory, the MGCBF maintains the counters for all incoming flows with small error probability, and keeps information of long flows whose length are bigger than an optional threshold set by users. This paper builds up an architecture for long flows' information statistics based on this algorithm. And the space used, calculation complexity and error probability of this architecture are also discussed at following. The experiment applied this architecture on the CERNET TRACEs, which indicates that the MGCBF algorithm can reduce the resource usage in counting flows and flows information maintenance dramatically with losing little measurement's accuracy.