Predicting Traffic Flow Size and Duration

• Autor(a) principal: Martins, Ricardo Alexandre Sacoto
• Data de Publicação: 2018
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10362/59918
• topic: Software Defined Networking; OpenFlow; Data Collecting Controller; flow; Domínio/Área Científica::Engenharia e Tecnologia::Engenharia Eletrotécnica, Eletrónica e Informática

Current networks suffer from poor traffic management that leads to traffic congestion, even when some parts of the network are still unused. In traditional networks each node decides how to forward traffic based only on local reachability knowledge in a setting where optimizing the cost and efficiency of the network is a complex task. Modern networking technologies like Software-Defined Networking (SDN) provide automation and programmability to Networks. In such networks control functions can be applied in a different manner to each specific traffic flow and a variety of traffic information can be gathered from several different sources. This dissertation studies the feasibility of an intelligent network that can predict traffic characteristics, when the first packets arrive. The goal is to know the duration and size of flow to improve scheduling, load balancing and routing capabilities. An OpenFlow application is implemented in an SDN Data Collecting Controller (DCC), that shows how the first few packets of a traffic flow can be gathered with scalability concerns and in a non-intrusive way. The use of different classifiers such as Random Forest, Naive Bayes, Support Vector Machines, Multi-layer Perceptron and K-Neighbour for effective flow duration and size classification is studied. The results of using each of these classifiers to predict flow size and duration using the DCC gathered data are presented and compared.