Delay and QoS Aware Low Complex Optimal Service Provisioning for Edge Computing

Abstract

Edge computing can be used as a distributed counterpart of the cloud computing, and task offloading with much shorter delays can be achieved. In general, edge servers are resource-limited, but the services are becoming further diversified to meet the ever-increasing user demand. When an edge server cannot process the received service request of a particular service type for not locally having the corresponding service execution image, the request needs to be forwarded to another on which the particular image is installed. In this article, an optimal scheduling method that forwards such requests is proposed by using the variant multiple-knapsack framework. Then, to find the optimal solution, we propose to reformulate it as an integer linear program. As the problem size increases, however, the problem easily becomes intractable, so we propose a distributed solution by using both Lagrangian relaxation and decomposition theory. To evaluate the effectiveness of the proposed approaches, we carry out simulations with different network layouts and with different request arrival scenarios. The evaluation results show that the centralized Lagrange dual solution can yield near-optimal solutions. In addition, the decomposed distributed solution can significantly reduce the computational and operational complexity at the expense of the total reward.