Practical Service Allocation In Mobile Edge Computing Systems

Mobile edge computing is an emerging technology required to meet the very low latency requirement of 5G communication systems, where application resources are embedded within access networks. Edge computing resources can be deployed very close to end users, e.g., attachment points (AP). In such cases, the edge servers are likely to have a relatively smaller computing capacity than servers deeper in the network. This can be partially compensated by the density and scale of deployment of edge servers. To leverage such deployment, it is important to allocate services efficiently at the edge. For example, neighbor edge servers may serve users attached to a distant AP, as long as the network topology, usage, and edge application requirements, e.g., latency, enables it. In this context, user mobility is important. The user session with the edge should be migrated when needed to ensure that a mobile user continues to be served with a proper Quality of Service (QoS), typically linked to latency. In this paper, we study a service instance allocation algorithm focusing on maintaining QoS for end users, using service migration as a tool to achieve this goal. We propose a globally optimum allocation algorithm, derive a practical algorithm from it, and evaluate its use versus a heuristic and a reference algorithm.