Providing Quality-of-service Support to Legacy Applications Using Machine Learning

Quality-of-Service (QoS) support is an essential feature in recent and upcoming networking standards. Applications running on these networks can specify appropriate service classes for their connection flows so that the flows are treated accordingly. However, current internet applications (legacy applications) cannot benefit from this facility as they are designed using the best-effort scheme. Effective QoS support to applications with unspecified service classes can be provided through our proposed intelligent QoS Manager. This paper describes the important features that can be used to determine service classes and discusses how machine learning can be incorporated into a QoS Manager, enabling it to distinguish different types of application flows and assign them appropriate service classes. Experiments using 10-fold cross-validation (CV), 33% hold-out (HO) and the learner's specific features lead to the selection of PART (Frank & Witten, 1998) as the classifier of the framework. It achieves 91.55% and 93.29% prediction correctness in CV and HO experiments respectively.