Simulation and Analysis of Mathematical Methods in Real-Time Engineering Applications. Группа авторов

      2

      Edge Computing Optimization Using Mathematical Modeling, Deep Learning Models, and Evolutionary Algorithms

       P. Vijayakumar*, Prithiviraj Rajalingam and S. V. K. R. Rajeswari

       ECE Department, SRMIST, Kattankulathur, Chennai, India

       Abstract

      The rapid growth of the Internet of Things (IoT) with advanced applications requires high speed and real-time computing power. Edge computing brings the computation of data closer to the machine where it is being collected. It leads to a decrease in latency, bandwidth usage, and resources for the server and its cost. The significant challenges in edge computing are 1) optimal offloading decision making, 2) resource allocation, 3) Meeting Quality-of-Service (QoS) and Experience (QoE). This chapter addresses the above challenges using mathematical models, Deep Learning and the Evolutionary algorithm. The deep learning algorithm solves the highly complex problem by developing a model from the training data or observation (reinforcement learning). The deep learning approach converts the optimization problem of edge computing into classification or regression or intelligent decision-making problems and solves them. The Evolution algorithm finds an optimum solution for the given problem through the natural process of evaluation, which is used to solve the edge computing multi-optimization problem. An evolution algorithm like a genetic algorithm and ant colony can solve a few research problems of edge computing like task scheduling.

      Keywords: Edge computing, deep learning, machine learning, evolutionary algorithm

2.1 Introduction to Edge Computing and Research Challenges

      Edge computing is a new distributed computing paradigm. The pattern of edge computing is closer to the location as a platform for computation and data storage before working with the cloud. In simpler terms, edge computing works with smaller and real-time data, whereas cloud works with big data. Edge computing helps in quick response times and also saves bandwidth [1, 2]. In the use case of cloud-based augmented reality applications, latency and processing limitations are key challenges to implementing the cloud system due to geographical distance from the infrastructure. Edge computing comes into the picture as an advancement of cloud gaming as it allows short-distance travel of data. Edge computing has the advantages of reducing lag times and latency [3]. Mostly edge computing has a role in helping cloud-based IoT systems to provide computational service. A small recap of the Cloud-Based IoT system is provided below.

      2.1.1 Cloud-Based IoT and Need of Edge Computing

      The Internet of Things (IoT) plays a vital role in human daily life by making all the devices connected through the internet, and it works ingeniously. Day by day, the IoT plays a crucial role in all the domains [4]. For example, IoT provides excellent service to medical applications like tracking patient status, heart rate, blood pressure, and sugar level can be monitored, and if a patient goes into a critical or unstable condition, the doctor can provide solutions through the report generated by the IoT application [6]. The IoT data can also be used to study different patients’ lifestyles and activities to prevent them from going into a critical situation. Therefore, the IoT has developed opportunities to provide brilliant solutions with many predictions and intelligence.

      IoT devices are correctly functioning because of several technologies like cloud computing that give many advantages to IoT devices, including storage infrastructure, processing the real-time data in IoT devices, and high-performance computing. It leads to cloud computing as a revolutionary part of IoT devices, which provides smart and self-predicted data [6]. Due to IoT devices’ evolution, cloud providers take an immense advantage to provide the communication or transfer of data between the IoT devices. This results in the Cloud of Things, which connects both cloud computing and IoT devices.

Large data centers are built by various cloud providers across the world, which have the capacity to serve users from around the world. Hence there exists a separation from the cloud centers to the users, which creates delays, latencies. Another disadvantage of cloud services due to distance of separation is location access can never be accurate. Information about users and their mobility is also another disadvantage of cloud computing. In augmented reality applications like cloud gaming as discussed for real-time tracking system applications like vehicular systems, a unique computing method is required, which is achieved by edge computing [7].

      Edge computing enables the deployment of cloud computing capabilities at the edge of the network. The infrastructure providers own the data centers and implement multi-tenant virtualization. The third-party customers, end users, and infrastructure providers can access these edge data centers. Edge computing services are automated, thus refraining from disconnecting from the cloud. This leads to the possibility of creating a hierarchical multi-tiered architecture. Edge computing leads to an open ecosystem where one trusted domain cooperates with other trusted domains and a multitude of customers are served. Though there are multiple edge paradigms with few differences, there are also similarities [8]. Edge Architecture’s outline is provided below for understanding before dealing with the challenges and the solution using mathematical models (Markov Chain Model and game theory), deep learning and evolutionary algorithm.

      2.1.2 Edge Architecture

      All the edge activities are controlled by two-tier architecture, so it works properly in time-sensitive systems. The design of two-dimensional structures focuses primarily on efficiency, application management, and edge management.

The edge network aims to reduce the data strain by providing computation away from data centers towards the network’s edge. To provide services for cloud computing, an edge computing network is created by using smart objects and network gateways and decentralizing the data centers. The edge network consists of three layers, among which Edge devices are the first layer. Edge devices are the data providers that include user gadgets, e.g., sensors, machines, smartphones, and wearables, as shown in Figure 2.1. They are responsible for data collection and delivery [7]. Edge nodes, the second layer of the network layer, are the router, switches, and small/macro base stations responsible for computing operations and data processing and data routing [9]. The third layer is the cloud that consists of data centers, servers, databases, and storage, and is responsible for data analytics using Artificial intelligence, visualization, and other high computational requirements [9].