Overload prediction and avoidance for maintaining optimal working condition in a fog node

Abstract Fog computing uses fog nodes (FNs) near end devices to serve user applications in a highly virtualized environment. FNs are vulnerable to overloading because of the large number of user applications requesting service and their heterogeneous resource requirements. The proposed paper concentrates on maintaining FN in optimal working condition guaranteeing its operational efficiency and quality of service (QoS). This paper presents two methods for workload classification and overload prediction in FN, viz., a threshold-based technique adopting the criteria inferred from experimental results and a method based on Hidden Markov Model (HMM). The paper recommends virtual machine (VM) migration as a solution for avoiding the predicted overloading of FN through a hybrid multiple attribute decision making (MADM) method, for selecting the VM for migration, and an algorithm for finding the destination FN willing to host the migrating VM. Experimental results validate the proposed solution.