Efficient Computer Networks Peer-To-Peer (P2P) Traffic Management and Control Using Machine Learning with Open Source Tools

Peer-to-Peer (P2P) applications constitute a large proportion of today's Internet traffic because of the growing number of users (i.e. Internet-enabled devices) and applications (most of which are P2P) on the Internet. P2P applications are known to use a lot of bandwidth during transfer of files. The impact of P2P in educational institutions especially in developing countries if not adequately controlled may lead to congestion and subsequently considerable reduction in traffic flow. It is therefore necessary to correctly identify and manage P2P on a given network especially where bandwidth resource is limited and costly. To facilitate P2P feature detection, using a university campus network as case study, we propose two methods for P2P detection, classification and control namely: a logistic regression analysis in WEKA and was used to model the feature selection and detection, and a hybrid system comprising Self Organizing Map (SOM) and Multi-Layer Perceptron (MLP) Neural Network. The logistic regression model detected 98.78% of P2P activity. The hybrid technique could detect 99.93%, 99.75% and 100% P2P activity in 4840, 2001 and 1473 instances of network traffic data respectively. This suggests that when P2P applications take on new features in the future, an unsupervised learning algorithm such as SOM has the potential to identify new P2P features for effective control and management on any given network.