Traffic classification in server farm using supervised learning techniques

Abstract

Server farms used in web hosting and commercial applications connect multiple servers. Edge computing being a realm of cloud technology is orchestrated with server farms to enhance network efficiency. Edge computing increases the availability of cloud resources and Internet services. The higher availability of services and their ease of access deeply affect the user’s requesting behavior. The anomalous requesting behavior is creating malicious traffic, and enormous amount of such traffics at server farm denies the services to the legitimate users. Categorizing the incoming traffic into malicious and non-malicious traffic at server farm is the foremost criteria to eliminate the attacks, which in turn improves the QoS of the server farm. In the light of preventing the biased usage of the server farm, this paper proposes a SVM classifier based on requesting statistics. The proposed classifier discovers the attacks that deny services to legitimate users in two levels, based on the user’s request behavior. The pattern of arrival, its statistical characteristics and security misbehaviors are investigated at both levels. An incremental learning algorithm is proposed to enhance the learning plasticity of the proposed classifier. The experimental results illustrate that the performance of the proposed two-level classifier with respect to classification accuracy is competently improved with incremental learning.