Computation Offloading over Fog and Cloud Using Multi-Dimensional Multiple Knapsack Problem

Computation offloading over fog and cloud is critical to improve service quality and efficiency of future networks. Mobile vehicles have also been considered as potential fog nodes by sparing their computation capability to nearby users. In this paper, we propose a multi-layer computation offloading architecture, consisting of the user layer, mobile fog layer, fixed fog layer and cloud layer. Multiple wireless roadside units (RSUs) are deployed in the network to collect computation tasks from user layer, and offload the tasks to other layers. Each layer has distinct multi-dimensional characteristics, such as different transmission rates and computation capabilities. The computation tasks may consume different communication and computation resources when they are uploaded to different layers. However, the available resources of each layer are limited. Consider that each user will pay for the offloaded computation tasks according to their sizes, we aim to maximize the total profits of computation offloading from the infrastructure perspective. Specifically, the offloading problem is formulated as a generalized multidimensional multiple knapsack problem (MMKP), in which each layer is considered as a large knapsack and the computation tasks are treated as items. We propose a modified branch-and-bound algorithm to obtain the optimal solution, and a heuristic greedy method to obtain approximate performance with much lower computational overhead. A comprehensive simulation is conducted to compare the proposed two algorithms. Simulation results demonstrate that the proposed computation offloading architecture together with the task allocation algorithms can achieve the purpose of maximizing the total profits of offloaded tasks.