MultiClassifier: A combination of DPI and ML for application-layer classification in SDN

In traditional campus network, application-layer classification is often achieved by using specific devices that support application-layer classification. Since different vendors have different realizations, even the same flow may have different results with different devices. Thus it´s hard to set a global consistent application-layer management policy for the whole network. The idea of separating the control plane and the data plane comes up with Software Defined Network have opened a gate for solving this problem. In the SDN paradigm, the control plane have a global view over the whole network, thus it can do application-layer classification and set policies globally. In this paper, we identify problems with the current application-layer classification in campus network and analyze the advantage of doing application-layer classification with SDN. And based on SDN, we show a new approach to do application-layer classification combining different classifiers: Deep Packet Inspection and Machine Learning based Packet Classification. Our experiments show that with this approach, we can archive a high classification speed while maintain an acceptable accuracy rate.