Task Allocation For Mobile Edge Computation Offloading

With the rapid development of mobile Internet and wireless communication technologies comes the latest structure of Mobile Edge Computing(MEC).The MEC architecture deploys computing-capable devices at the edge of the network near the mobile devices to shorten the distance between the user and the computing device.The mobile device can therefore offload tasks to computing device over a one-hop transmission distance,reducing propagation delay.Offloading in mobile edge computing is no longer a binary alternative of whether to perform locally or migrate to the far end.Instead,it becomes a Task Allocation Problem under multi-device environment.In this article,we mainly discuss the allocation problem of offloading for different delay requirement tasks and the load balancing problem of computing devices in the edge network.In order to solve the problem of allocation in the case of multi-server edge network,we measure the load of a node by its usage prediction on task computing time.A centralized two-step reservation mechanism is proposed to solve the allocation problem.We adapt a modified genetic algorithm(GA)to reserve a suitable server and time slot for each task.By means of population iteration and evolution,an optimal feasible solution is obtained which not only guarantees node load balancing but also minimizes the average task completion.Simulation results show the effectiveness of the two-step central reservation mechanism on task allocation,it can also restrain the load of each server in the cloudlet network.The two-step mechanism outperforms some of the current task allocation scheme with 9.5% drop of stand deviation for node usage.To accurately distinguish the delay requirements for different tasks,we propose a task allocation mechanism based on servers' historical statistics under multi-server edge computing scenario.A centralized controller is deployed at the base station to gather tasks from different users.The users first rate their tasks for different priority and upload them to the central controller's corresponding priority queue.The controller calculates the average occupancy rate of the current edge server cluster and compare it with the priority threshold of each queue.Queues with threshold higher than the occupancy rate are able to send some of their tasks to the server cluster.Queues whose threshold lower than the rate should take time for vacation and wait for another turn.Simulation results show that the algorithm based on statistical priority can effectively allocate tasks to propriae servers with low latency,especially for those high-priority tasks,whose completion rate goes above90%.