Edge Resource Provisioning

Mobile edge computing has the advantages of reduced latency and relieved outsourced traffic to the core network since it provisions resources at the edge of network. However, edge nodes have inherently limited resource capacities, which cannot well scale with the time-varying mobile requests. Different from Chap. 5 , which seeks to accommodate the time-varying requests by dynamically scheduling among multiple edge nodes, this chapter emphasizes more on provisioning elastic resources in edge systems. By providing different types of cloud instances, cloud computing can help enhance the scalability of edge resources. Thus, this chapter introduces the Cloud-Assisted Mobile Edge (CAME) computing framework to provision scalable resources for edge platforms. In the CAME framework, the burstness of mobile requests at edge nodes can be counteracted by outsourcing excessive requests to the cloud at rush hours. To ensure the diversified quality of service requirements with minimal resource cost, the fixed edge resource capacity and elastic cloud resource capacity should be properly utilized in a cost-efficient manner. The CAME framework should address the following two questions in particular: (1) the optimal resource capacity for the edge, and (2) the types of cloud instances being tenanted and the instance number of each type. To answer the above questions, this chapter formulates the cost-efficient resource provisioning problem of the CAME system. By analyzing the property of the problem, this chapter presents algorithms to determine the optimal edge resource capacity and tune the cloud tenant strategy adaptively. Regarding various types of cloud instances, the Optimal Resource Provisioning (ORP) algorithms are designed for three cases, i.e., edge resource provisioning with on-demand instances, with reserved instances, and with hybrid instances. Extensive simulations are performed to evaluate with two widely used mathematical traffic models and Google cluster usage tracelogs. It is confirmed in the results that the ORP algorithms have the advantages of resource flexibility and cost-efficiency over the benchmarks.