Searching for Periodic Flows in an Internet Backbone

Abstract

If questioned on a traffic pattern occurring every day at precisely the same time, one would intuitively attribute it to some timed application such as e.g. amirror update. Nonetheless has research so far neglected to investigate the key properties of above, apparently obvious, phenomenon. This report presents a methodology for detecting traffic exchanges at periodical intervals between hosts. We applied the methodology on NetFlow statistics of an Internet backbone. We discovered about 15% of the TCP traffic on a border router to be periodic; moreover, comparably few flows contribute to this phenomenon. With the prospect of traffic engineering we examined the traffic s property of time persistence.