Automation of Computational Resource Control of Cyber-Physical Systems with Machine Learning

Cyber-physical systems require the quality of service (QoS) guaranteed performance of service functions and processes executed in cyberspace. Because of low-latency requirements, most of such functions must be executed in edge computing infrastructure, where computational resources are limited. For efficient management of limited available resources in edge cloud to meet very low-latency requirements of services, this paper proposes a dynamic resource control scheme to adjust computational resources allocated to virtual network functions (VNFs). The scheme employs machine learning (ML) techniques composed of multiple regression models, which are continuously retrained online by using performance data collected from the running system. We demonstrate its effectiveness through experimental evaluation results obtained from an implementation of an IoT-directory service function in a resource virtualization platform provided by Docker containers in cyberspace. The IoT-directory service, whose architecture is based on Recommendation ITU-T Y.3074, is a scalable system that can store a huge amount of control information of a billion IoT devices in the form of name records and provides a very fast lookup service with the latency of a few milliseconds. The proposed scheme is related to ML-based network control and management methods currently being standardized in the ITU-T Study Group 13.