Page 86 - Kaleidoscope Academic Conference Proceedings 2020

2020 ITU Kaleidoscope Academic Conference

Artificial Intelligence (AI) to provide ubiquitous connectivity and realize the fusion of
Cybe rs pa ce physical space and cyberspace. Thanks to its capabilities of
Vis u a liza tio n An a lys is De cis io n transmitting a huge volume of data, about 20 Gigabits in
every second, connecting one million devices in the area of
Big Data I d e n t ifica t io n Pre d ict io n Op t im iza t io n every square kilometer, and reducing transmission latency to
around one millisecond, 5G networks can upload a huge
5G Actu a tion volume of sensing raw data from the physical space to
cyberspace and transfer knowledge and control or actuation
Se ns ing
IoT Ph ys ica l s p a ce commands from cyberspace to the physical space.
AI and ML enable cyberspace to learn, infer, and make an
Robot
intelligent decision like humans. Additionally, AI/ML
techniques have the capability of processing a huge amount
Figure 1 - Cyber-physical system illustration of data (known as big data) in a short time, exceeding the
capability of humans. AI/ML techniques are now being used
Y.3074 [5], is a scalable directory service system, which can to solve numerous problems in various service domains such
store control information of a billion IoT devices in the form as healthcare, manufacturing, transport mobility, energy
of name records and provide a very fast lookup service with management, weather forecast and safety. The advancement
the latency of a few milliseconds. The proposed resource of IoT and the reduction of sensor electronics’ price have
control scheme dynamically adjusts the amount of allocated made it feasible to collect a huge volume of data of different
computational resource to the IoT-DS module according to granularity. Similarly, the availability of sophisticated
the fluctuating workload. It assures that enough computation and data storage facility in cyberspace has made
computational resource is provided so that the record lookup it possible to store, share and process big data by AI/ML
latency does not exceed the maximum tolerable limit, while techniques in a short time. The more the volume and
the allocated resource is utilized at a maximum possible level. diversity of big data are, the better the AI systems can be
This scheme is related to the ITU-T Study Group 13’s trained so that they can produce highly accurate results.
standardization work on ML-based network control and
management methods. Robots are the agents in the physical space that work on
behalf of humans. A robotic agent can exist separately, or be
attached to other machines. Humanoid robots helping with
The remainder of this paper is organized as follows. Section housework can exist separately. In Society 5.0, with the
2 provides an overview of cyber-physical systems and support of robots, humans can live more comfortably and
related works. Section 3 presents the proposed multiple devote more time to create innovative products and services,
regression-based resource adjustment scheme. Section 4 optimizing the entire social and organizational system.
discusses the performance evaluation results obtained from
a preliminary implementation in a small-scale experimental 2.2 Related work
system. The last section concludes the paper with some
remarks on related ITU-T standardization work.
This subsection outlines related work on the computing
2. CYBER-PHYSICAL SYSTEM OVERVIEW AND resource control of cyberspace. The computing resource is
allocated in the form of CPU cores and CPU time in a fine
RELATED WORK granularity of microseconds. ML and threshold rule-based
techniques have been utilized for virtual resource
This section presents an overview of cyber-physical systems management. ML-based techniques use various models such
and outlines related works on machine learning-based as Gaussian process, auto-regression, and supervised
resource control mechanisms. learning. A Gaussian process-based online workload and
latency prediction model is presented in [6] to adjust the
2.1 Cyber-physical systems number of CPU cores allocated to Docker containers for
executing tasks of latency-sensitive applications. An auto-
Figure 1 illustrates a cyber-physical system, whose regressive model is used in [7] to predict CPU resource
cyberspace contains big data, and AI and 5G network and requirements and allocate CPUs to virtualized servers
physical space contains IoT devices and robots. running in enterprise data centers. A supervised learning
approach is described in [8], where regression models are
IoT is a generic concept of extending Internet connectivity to trained offline and then employed for the dynamic resource
any physical object by embedding devices capable of sensing, adjustment decisions. The authors of [9] employed a rule-
controlling, computing and communication in the object. based vertical scaling of Docker containers running Graylog
Various types of sensors attached to IoT devices monitor the server applications to adjust the allocated CPU cycles every
environment they lie in and generate a huge volume of data 4 seconds.
in the form of video, image, sound, weather conditions, and
system operation logs to name a few. 5G mobile Prior work [8] proposed a scheme employing prediction
communication networks, whose deployment has started models based on multiple regressions to adjust resources
recently in many countries, are indispensable infrastructures allocated to latency sensitive VNFs dynamically. The

– 28 –