Research on Fine-Grained Micro-Service Placement Strategy for IoT Networks with Restricted Computation Resources

In recent years, AI enabled smart applications such as image recognition, network operation have attracted much attention. However, it is usually difficult to support the AI enabled services in IoT network, since the computation and communication resources are strictly restricted. In this paper, we consider an IoT network scenario with several APs having edge computing function. We first put forward a micro-service placement strategy, with which the AI applications are implemented in terms of micro-services on a Kubernetes platform. Considering the characteristics of Kubernetes, we divide one whole service into different layers which are implemented in the containers, respectively. In this circumstance, different AI applications with the same basic micro-service can get the implementation result from one designative edge server (APs having edge computing function) avoiding replication computation. In addition, the higher layered micro-service layers can be performed coordinately by other IoT edge servers. With which, the computation redundancy can be highly decreased. Finally, we propose a deep reinforcement learning algorithm for the fine-grained micro-service placement optimization. In the simulation part, we demonstrate the advantage of the proposed scheme.