Location Aware Workflow Migration Based on Deep Reinforcement Learning in Mobile Edge Computing

Abstract

Prompted by the remarkable progress in mobile edge computing, there is an increasing need for executing complex applications on the edge server. These complex applications can be described using workflows which is a set of interdependent tasks. Existing studies focus on offloading the workflow tasks to the nearby edge servers in order to achieve high quality of service, However, the original edge server with the offloaded workflow tasks may be far away from the users due to the high mobility of users in mobile edge computing (MEC). Therefore, it is a key challenge to make good decisions on where and when the workflow tasks are migrated in the light of user’s mobility. In this paper, we propose a workflow task migration algorithm based on deep reinforcement learning with the goal of optimizing the cost of workflow migration under delay-guarantee constraints. The proposed algorithm firstly utilizes the Recurrent Neural Network (RNN) based model to predict the mobile location of users, and then applies a dynamic programming algorithm to calculate the completion time of workflow. Finally, an improved Deep Q Network (DQN) algorithm is adopted to find the optimal workflow migration strategy. In order to assess the performance of the proposed algorithm, extensive simulations are carried out for four well-known scientific workflows. The experimental results show that the proposed algorithm can meet threshold at lower costs in comparison with the state-of-the-art approaches applied to similar problems.

Supported in part by the National Natural Science Foundation of China under Grant 61662052, in part by the Natural Science Foundation of Inner Mongolia Autonomous Region under Grant 2021MS06002, in part by he Science and Technology Planning Project of Inner Mongolia Autonomous Region under Grant 2021GG0155, and in part by the Major Research Plan of Inner Mongolia Natural Science Foundation under Grant 2019ZD15.