A Mobility Prediction Based Adaptive Task Migration in Mobile Edge Computing

Abstract

During the past few years, mobile data traffic has exponentially increased due to emerging applications, such as social media, online gaming, and augmented/virtual reality. Although the capabilities of mobile devices are significantly improved, they are unable to execute computationally intensive tasks. To extend the computing capabilities of resource-constrained mobile devices, computation offloading is performed on edge servers. Due to user mobility, offloaded tasks often need to be migrated from one edge server to another. Mobility-aware task migration faces different challenges due to varying mobility characteristics of end-users. These challenges include latency, server utilization, and energy consumption. Existing techniques of task and machine (VM) migration do not consider the user movement trajectories while making migration decisions. Consequently, the task or VM is migrated to the edge server that may be far away from the mobile users' location that increases the response time. In this paper we proposed Mobility Migration Algorithm based on Linear Regression (MALR). After outsourcing the task, a recurrent neural network (RNN) and linear regression are used to forecast the user's present location. Using the distance between the user and the server, it gets a list of nearby servers, and then moves the task there. The proposed approach eliminates the job migration time with improvement in forecast accuracy as compared to the logistic regression and K-mean.