A High-performance IoT Middleware Based on Reinforcement Learning

This paper proposes a high-performance Internet of Things (IoT) middleware based on reinforcement learning. The adaptive decision-maker is realized by the K-armed bandit algorithm, which can dynamically resize the server accept queue. This paper adopts service-oriented architecture(SOA) to make the middleware architecture be loosely-coupled, reusable, and easily extensible. Through standardized service interfaces, users can transparently interact with heterogeneous IoT devices. In addition, the high availability of the middleware platform is achieved through non-blocking IO, caching, and message queuing technologies. Through the experiments over Apache JMeter, the results verify that our IoT middleware has a better performance under high concurrency scenarios.