Deep Learning for Community Network Traffic Classification
Community networks provide a wide range of application services such as VoIP, content distribution, on-demand and live streaming media, instant messaging as well as back-ups and software updates. The challenge often is that they have to provide these services on links and servers with limited capacity. Therefore, efficiently prioritizing network traffic is vital to ensuring a good user experience on the network. Traffic clssification is a vital part of the Quality of Service (QoS) system, which prioritises network traffic. The main goal of the project thus was to build and experimentally evaluate Deep Learning network traffic packet classifiers for the purposes of QoS and traffic engineering in community networks.
Data Aquisition
The first step for the project involves accessing the community network traffic data to be used for training and testing the models. We use a dataset that was collected from the Ocean View community network in Cape Town. The data consists of numerous PCAP files that were collected at the gateway of the network, capturing all traffic flowing between the network and the Internet from February 2019. The PCAP files have been copied to a data repository at the University of Cape Town, through which we access the data.
