Traffic classification and management based flow statistics netfpga

Abstract

The internet bandwidth increased significantly over the past years but the problem of network bandwidth management remained a key issue. One of the major problems associated with bandwidth management is network bottleneck, which is the overcapacity of network traffic due to abnormal application bandwidth usage. With the release of new applications every year, especially P2P applications that require high bandwidth, effective network management has become even more important. Congestion can be caused inside a network by numerous flows and high bandwidth applications that may dominate the total bandwidth allocation, affecting normal users. This report presents an approach to detect and manage high bandwidth traffic flows in a congested network, providing fair bandwidth usage to normal users and restricting bandwidth-heavy applications. Flow statistics information is used for classification of network traffic by applying k-means clustering. An inline rate-limiter technique based on queue management is used for controlling high bandwidth flows. The proposed traffic shapping method queues the header packets of flows that are classified as high bandwidth flows. These modules are integrated into the NetFPGA platform, where decision making is carried out with minimal intervention of network administrators by only updating the classifier model when accuracy falls below a threshold line. It ensure zero intrusion of user privacy and at the same time it is able to reduce the high bandwidth rate, providing fair network usage for home users.