Intelligent IP Traffic / Flow Classification System

In QoS-aware networks, such as DiffServ-enabled IP networks, UMTS, or IEEE 802.11e, the QoS-aware applications that run over them can identify service classes to their flows. The flows are then treated by the networks differently with respect to their classes. In contrast, legacy applications are not aware of the concept of QoS and do not specify any classes to their flows. Thus they cannot benefit from the QoS-support provided by the networks. To this end, this paper brings forth a new intelligent flow classification system (FCS) that can automatically identify the service classes of legacy flows. The proposed FCS employs a new flow-level characteristics or “features” allowing it to identify flows on-the-fly and no packet-level data are required. Equipped with a machine learning technique, it is also adaptive and self-updatable. Moreover, the FCS is evaluated using packet traces from a sizeable network. The results show that our FCS works remarkably well with average accuracy of 99.66%.