Page 85 - Kaleidoscope Academic Conference Proceedings 2020

AUTOMATION OF COMPUTATIONAL RESOURCE CONTROL OF CYBER-PHYSICAL
SYSTEMS WITH MACHINE LEARNING

Ved P. Kafle; Abu Hena Al Muktadir

National Institute of Information and Communications Technology, Tokyo, Japan

ABSTRACT extends Internet connectivity to physical objects such as
machines, people, animals and environments equipped with
Cyber-physical systems require the quality of service (QoS) devices capable of sensing, controlling, computing and
guaranteed performance of service functions and processes communication. 5G networks are used to connect physical
executed in cyberspace. Because of low-latency objects with the corresponding digital objects in cyberspace.
requirements, most of such functions must be executed in 5G networks will be capable to upload a huge volume of
edge computing infrastructure, where computational sensing data from physical space to cyberspace and transfer
resources are limited. For efficient management of limited knowledge or control commands in the reverse direction. For
available resources in edge cloud to meet very low-latency time-critical services, the cyber-physical interaction should
requirements of services, this paper proposes a dynamic be completed in a very small-time latency, usually within a
resource control scheme to adjust computational resources few milliseconds [3]. Namely, communication and
allocated to virtual network functions (VNFs). The scheme computing tasks to be completed in cyberspace should finish
employs machine learning (ML) techniques composed of within a very strict time limit.
multiple regression models, which are continuously
retrained online by using performance data collected from Computing, storage and communication resources in
the running system. We demonstrate its effectiveness through cyberspace are virtualized so that they can be controlled and
experimental evaluation results obtained from an utilized more efficiently. These resources can be allocated to
implementation of an IoT-directory service function in a cyber functional entities in a fine granularity and adjusted
resource virtualization platform provided by Docker dynamically. Functional entities in cyberspace exist in the
containers in cyberspace. The IoT-directory service, whose form of software programs, which are placed as nodes of a
architecture is based on Recommendation ITU-T Y.3074, is network and known as virtual network functions (VNFs). To
a scalable system that can store a huge amount of control satisfy the performance requirements of latency-sensitive
information of a billion IoT devices in the form of name services, such as augmented reality, telemedicine, online
records and provides a very fast lookup service with the multiparty video games, and self-driving vehicles, the VNFs
latency of a few milliseconds. The proposed scheme is are placed in cloud computing infrastructure available at the
related to ML-based network control and management end-user’s proximity, known as the edge cloud [4], so that
methods currently being standardized in the ITU-T Study communication latency incurred for exchanging information
Group 13. between the users and cyberspace can be reduced to a very
small value. Since edge cloud has limited computing
Keywords— Artificial intelligence, cyber-physical system, resources, the dynamic and fine tuning of resources allocated
edge computing, machine learning, network infrastructure, to VNFs is necessary to optimally utilize resources while
resource control, standardization satisfying the service’s performance requirements. To
address this issue, in this work we investigate a dynamic
1. INTRODUCTION resource control scheme to optimally adjust the
computational resource allocated to VNFs.
Japan’s Society 5.0 [1] and Europe’s Industry 4.0 [2] visions
have considered exploiting the potential of digital The proposed scheme employs machine learning (ML)
technology for the unprecedented transformation of society techniques for the fine control of computational resources.
and industry. They envision exploiting technological Multiple regression models of gradient boosting regression
innovations in the field of Internet of Things (IoT), cyber- and extremely randomized trees are used as ML techniques.
physical systems, 5G networks, big data, edge computing, These models are trained offline before their deployment and
artificial intelligence (AI) and robotics. These technologies retrained regularly online by using telemetry data collected
help in realizing a forward-looking society that overcomes from the running system. We demonstrate the effectiveness
the existing limitations of humans, society, industry, and of the proposed scheme through experimental results
systems. obtained from an implementation of an IoT-directory service
(IoT-DS) system as a VNF in the resource virtualization
Cyber-physical systems integrate the objects of physical platform created with Docker containers [12]. IoT-DS,
space with the digital objects or entities in cyberspace. IoT whose architecture is based on Recommendation ITU-T

978-92-61-31391-3/CFP2068P @ ITU 2020 – 27 – Kaleidoscope