A Review on Computational Intelligence Techniques in Cloud and Edge Computing

Cloud computing (CC) is a centralized computing paradigm that accumulates resources centrally and provides these resources to users through Internet. Although CC holds a large number of resources, it may not be acceptable by real-time mobile applications, as it is usually far away from users geographically. On the other hand, edge computing (EC), which distributes resources to the network edge, enjoys increasing popularity in the applications with low-latency and high-reliability requirements. EC provides resources in a decentralized manner, which can respond to users' requirements faster than the normal CC, but with limited computing capacities. As both CC and EC are resource-sensitive, several big issues arise, such as how to conduct job scheduling, resource allocation, and task offloading, which significantly influence the performance of the whole system. To tackle these issues, many optimization problems have been formulated. These optimization problems usually have complex properties, such as non-convexity and NP-hardness, which may not be addressed by the traditional convex optimization-based solutions. Computational intelligence (CI), consisting of a set of nature-inspired computational approaches, recently exhibits great potential in addressing these optimization problems in CC and EC. This paper provides an overview of research problems in CC and EC and recent progresses in addressing them with the help of CI techniques. Informative discussions and future research trends are also presented, with the aim of offering insights to the readers and motivating new research directions.

READ FULL TEXT VIEW PDF

Authors

page 1

page 15

08/18/2021

Resource Scheduling in Edge Computing: A Survey

With the proliferation of the Internet of Things (IoT) and the wide pene...
04/18/2022

Actions at the Edge: Jointly Optimizing the Resources in Multi-access Edge Computing

Multi-access edge computing (MEC) is an emerging paradigm that pushes re...
06/12/2018

Opportunistic Edge Computing: Concepts, Opportunities and Research Challenges

The growing need for low-latency access to computing resources has motiv...
02/16/2022

A Survey on Scheduling Techniques in the Edge Cloud: Issues, Challenges and Future Directions

After the advent of the Internet of Things and 5G networks, edge computi...
02/13/2019

Optimization problems with low SWaP tactical Computing

In a resource-constrained, contested environment, computing resources ne...
11/11/2020

Joint Management of Compute and Radio Resources in Mobile Edge Computing: a Market Equilibrium Approach

Edge computing has been recently introduced as a way to bring computatio...
07/30/2021

An Edge-Based Resource Allocation Optimization for the Internet of Medical Things (IoMT)

As the number of Internet of Medical Things (IoMT) increases, the need f...
This week in AI

Get the week's most popular data science and artificial intelligence research sent straight to your inbox every Saturday.

List of Abbreviations

ACO Ant Colony Optimization
AI Artificial Intelligence
AIS Artificial Immune System
BA Bee Algorithm
BFA Bacterial Foraging Algorithm
CC Cloud Computing
CI Computational Intelligence
C-RAN Cloud-Radio Access Network
CSA Cuckoo Search Algorithm
DE Differential Evolution
EAs Evolutionary Algorithms
EC Edge Computing
FA Firefly Algorithm
FC Fog Computing
FI Fuzzy Inference
FL Fuzzy Logic
FS Fuzzy System
GA Genetic Algorithm
HS Hybrid System
IaaS Infrastructure-as-a-Service
IoT Internet of Things
LBS Learning Based System
MCC Mobile Cloud Computing
MDP Markov Decision Process
MEC Mobile Edge Computing
NN Neural Networks
PSO Particle Swarm Optimization
QoS Quality of Service
SIAs Swarm Intelligence Algorithms

I Introduction

Cloud computing (CC) is a paradigm of computing technologies that provides on demand services to its clients. The vision of CC is to offer computing, storage, and network resources centrally in the remote clouds, which is related to data centers, backhaul networks, and core networks [8, 179]. It is an architecture for allowing appropriate, pervasive, and on request access to a shared pool of configurable resources [90]. The large number of resources available in the central cloud can then be leveraged to deliver elastic computing capacity and storage to support resource-constrained end devices. It has been driving the rapid growth of many Internet companies [88]. For example, the cloud business has risen to be the most profitable sector for Amazon [159], and Dropbox’s success depends highly on the cloud service of Amazon. It can intensify collaboration, scalability, nimbleness, and availability for enterprises as well as users. CC offers services on a pay-as-you-go basis and reduces hardware and software costs, energy, and carbon footprints by providing an optimized and sufficient computing environment [15]. In CC, users first offload tasks to the central cloud, and then the central cloud executes the tasks on behalf of users and sends back results to users [24]. Although CC provides a vast number of resources, easy back and recovery, high accessibility as well as an eco-friendly environment for users (i.e., can be accessed anytime from anywhere on demand), it may not be capable of fulfilling the requirements of real-time applications with low latency and high reliability as the central cloud is far away from users.

Fig. 1: Architecture of CC and EC.

On the other hand, edge computing (EC), which deploys the resources, data, and services from the central cloud to the network edge, enjoys increasing popularity recently. It enables analytics and knowledge generation to occur at or close to users’ end. In EC, users first offload tasks to the nearby edge cloud, and then the edge cloud executes their tasks and sends outcomes back to users. Fig. 1 presents a typical architecture of CC and EC, where edge clouds can provide computing and network resources between users and the central cloud. Compared with CC, EC can provide services with low latency, high security, and high mobility but with limited computing capacities. CC and EC facilitate a wide range of applications including virtual reality/augmented reality, online game, online video, etc.

In CC and EC, several metrics are related to the quality of service (QoS), such as security and privacy, latency, resource utilization, cost, energy consumption, throughput, and makespan. In order to improve these metrics, we need to consider several issues, such as job scheduling, resource allocation, and task offloading. For example, security and privacy can be improved by designing efficient task offloading schemes [25]. In addition, latency and energy consumption can be optimized by designing proper resource allocation [82] and task offloading schemes [27, 141]. Similarly, other metrics can be improved by considering job scheduling, resource allocation, and task offloading. To this end, many optimization problems have been formulated for tackling these issues. However, compared with optimization problems in other areas, these optimization problems are normally highly complex and NP-hard, which may include mixed/strongly-coupled variables, nonlinear constraints, multiple objectives, and bilevel structures. Therefore, they may not be solved by the traditional convex optimization-based methods. In addition, their optimal solutions should be found within a reasonable amount of time.

As a class of nature-inspired computational approaches, computational intelligence (CI) exhibits great potential in addressing complex optimization problems, which has attracted much attention from both academia and industry. CI includes evolutionary algorithms (EAs), swarm intelligence algorithms (SIAs), fuzzy system (FS), learning based system (LBS), and hybrid system (HS). These CI techniques have the capability to process imprecise information and search for approximate yet good enough solutions while ensuring robustness and computational tractability [112].

The aim of this paper is to present a review of CI techniques to address issues in CC and EC. Several survey papers and books have been devoted to CC and EC. For example, Hoang et al. [53] presented a survey on architecture, applications, and approaches in mobile CC (MCC). A short review on task scheduling in CC is carried out in [131]. QoS in CC is surveyed in [50] and [63]. Wu [161] studied multi-objective decision making for offloading decision in MCC. An extensive survey on MCC is given in [41]. A survey on challenges and opportunities in CC is presented in [40]. Zhou et al. [182] published a survey on artificial intelligence (AI) in EC. In [129], visions and challenges of EC are discussed. A survey on EC for Internet of things (IoT) is investigated in [171]. Liu et al. [77] surveyed systems and tools in EC. Peng et al. [107] studied service adaptation and provision in mobile EC (MEC). Mao et al. [88] elaborated on the communication perspective of MEC. Lin et al. presented a survey on computation offloading in EC. Wang et al. [151] focused on offloading algorithms, issues, methods, and perspectives in edge cloud. Moreover, some other surveys devoted to MEC are [140, 111, 110, 85, 3, 121]. However, none of the aforementioned surveys considers CI techniques. In addition, some survey papers review CI techniques in CC or EC. For example, evolutionary approaches for resource management in CC are reviewed in [176] and [47]. Chopra and Bedi [30]

reviewed the applications of fuzzy logic (FL) in CC. Deep learning (DL) in EC is studied in

[19]. It is worth noting that the above-mentioned surveys only consider one particular type of CI for CC or EC.

This survey attempts to give a detailed review of the state-of-the-art CI techniques in CC and EC. Our paper is an ambitious effort to capture the interplay among CI, CC, and EC, instead of delving into one particular CI technique in CC and EC exclusively. The motivation behind this survey is to provide researchers with a glance of mutual relationship between CI and both CC and EC at a higher level.

Fig. 2: Paper Organization.

The works introduced in this survey are taken from relevant journals, workshops, conference proceedings, and theses. This survey focuses on reviewing issues in CC and EC tackled by CI techniques, rather than a detailed study of CC/EC/CI techniques.

The rest of this paper is organized as follows. Section II presents an overview of the concepts related to CC and EC along with important metrics and critical issues. Section III introduces CI techniques used for addressing critical issues in CC and EC. Section IV reviews the existing applications of CI techniques in CC and EC. Finally, Section V offers a detailed discussion on the use of CI techniques in CC and EC along with future research trends, followed by the conclusion in Section VI. For clarity, the organization of this paper is given in Fig. 2.

Ii CC and EC Paradigms and Their Related Metrics and Critical Issues

Ii-a CC and EC Paradigms

CC and EC are complementary fields and their features are summarized in Table I.

Feature CC EC
Latency High Low
Mobility No Yes
Architecture Centralized Distributed
Location Awareness No Yes
Security Less Secure More Secure
Service Access Through Core At User End
Availability High High
Geographic Distribution No Yes
Scalability Average High
Reliability Low High
No. of devices connected In Millions In Billions
Resources Huge Limited
TABLE I: Features of CC and EC

Ii-A1 Cc

In CC, different types of paradigms have been designed. Next, we introduce two typical paradigms of CC: MCC and cloud-radio access network (C-RAN).

  • MCC is proposed to enrich resources for users. It merges CC, mobile computing, and wireless networks to enable service providers to help users to conduct computation-intensive tasks [41]. It provides several benefits to users like extending battery lifetime, improving storage capacity and processing power, and providing computation-intensive applications [53].

  • C-RAN provides centralized processing, collaborative radio, and energy efficient infrastructure for RAN [162]. In this architecture, network computation-based tasks are performed in the central baseband unit, and the radio signals from distributed antennas are gathered through remote radio heads and transmitted to the cloud by optical transmission network. It offers better services without affecting coverage of network by reducing the number of cell sites and capital expenditures and operating costs.

Ii-A2 Ec

EC includes many paradigms, e.g., cloudlet, fog computing (FC), and MEC, which are discussed below.

  • Cloudlet, introduced by Satyanarayanan et al. [125, 126], is a decentralized and widely distributed infrastructure, i.e., providing storage and processing close to users. It is also referred to as a “data center in a box”. It is indeed proposed to fulfill the demands of users to reduce response time for new latency-sensitive applications without going to the main central cloud [43]. Although cloudlet has many benefits due to being close to users, it does not have enough computing capacity as the central cloud; thus, it is limited in providing computation-intensive services.

  • FC was first discussed by Cisco in 2012 [16] and is referred to as an extension of services from the central cloud to the network edge. It reduces the amount of data sent to the central cloud for processing and storage [28]. It is a paradigm that furnishes computing, storage, and networking services close to users. FC provides better support for real-time applications, dense geographical distribution, low latency, and location awareness.

  • The architecture of MEC was proposed by European Telecommunications Standards Institute in 2014, enabling the provision of resources close to users via RAN [169]. It can provide services with low latency and high bandwidth at the edge of radio network. In addition, MEC makes it possible to measure and improve network performance by setting up services like software defined network and network function virtualization [56].

Although EC can effectively overcome some problems of network congestion and long latency in CC, it has limited computation and communication capabilities [120]. On the other hand, although CC has rich computing resources, it may suffer from high latency. To address the above-mentioned problems, researchers have made some attempts to investigate the hierarchical computing (i.e., the collaboration between CC and EC), such as [120, 127, 70], where the tasks of users can be partially or jointly processed at both edge and central clouds.

Ii-B Metrics

Over the past two decades, many researchers focused on diverse aspects of CC and EC, including security and privacy, energy consumption, resource utilization, etc. Some important metrics are given in Fig. 3 and discussed below.

Fig. 3: Metrics in CC and EC.

Ii-B1 Energy Consumption

EC may suffer from battery life problem. Even though CC is normally connected to the power grid, energy efficiency is still very important as it is relevant to the profit and economy of CC operators. The energy spent on end devices during task execution is a primary aspect to be considered. Total energy consumption includes static energy consumption and dynamic energy consumption [4]. Energy consumed by the system without considering any workload is called the static energy, while the dynamic energy is the energy consumed on the current cloud/edge resources by virtual machines (VMs). Usually, an edge cloud can be a small device like laptop and mobile, and applies the battery. However, CC normally connects to power grid, and has much more power consumption than EC. Therefore, saving energy in EC is to keep edge cloud alive, whereas in CC, saving energy contributes to green society. For users, due to the long distance transmission, CC usually needs more energy than EC for offloading tasks.

Ii-B2 Security and Privacy

Security and privacy is one of the important metrics to be considered for providing secure and trusted services to users. Many hackers try to interrupt or steal data while users offload and process their tasks at central/edge clouds. Security and privacy is a set of control-based technologies and policies designed to protect sensitive information, data, applications, and infrastructures [11]. CC is more vulnerable to be attacked by hackers due to the long distance transmission between users and the central cloud. In contrast, EC is a decentralized architecture, which is more secure than CC. Locally sharing, storing, exchanging, and analyzing data among edge clouds make it harder for hackers to get access to data. Moreover, real-time processing and response of EC make it difficult for malicious attackers to detect sensitive information of users. However, EC still inherits many security problems such as privacy leakage, forgery, tampering, spam, jamming, and impersonation [99].

Ii-B3 Resource Utilization

If CC and EC infrastructures place VMs for resource utilization without any specific scheme, some resources may run out while others are idle. Thus, an efficient resource utilization scheme becomes more important to get the maximum profit of resource utilization by utilizing resources properly [5]. Moreover, load balancing also needs to be considered for better performance and resource utilization. It is essential to distribute workloads across multiple nodes of the system to maintain stability, minimize latency, and improve resource utilization ratio in CC and EC. In CC, proper utilization of resources is needed to save energy, host more users, and make more profits, whereas in EC, it is important to meet high-reliability and low-latency requirements.

Ii-B4 Latency

A mass of novel mobile applications is emerging, most of them are latency-sensitive. Typically, latency is defined as the delay between a user’s request and response of a service provider [64]. It has a high effect on the usability and enjoyability of end devices. CC is a large system and hosts a huge number of users, thus leading to routing problems such as VMs’ connection problem. In addition, the long physical distance between users and the central cloud also leads to high latency. In contrast, edge clouds are placed at the network edge and thus have a better capability to perform latency-sensitive tasks than CC.

Ii-B5 Makespan

Makespan is the maximum required time to complete all assigned tasks [116, 47], including response time, execution time, waiting time, etc. It can be expressed as:

(1)

where is the completion time of task , and is the number of tasks.

Compared with EC, there are a huge number of tasks needed to be processed in CC. Therefore, many researchers focused on makespan in CC only.

Ii-B6 Cost

Since users and service providers usually belong to different entities, users have to pay for availing specific resources [64] such as computing or communication resources. The cost can consist of computing cost, running cost, and setup cost. The purpose of this metric is to reduce the cost of services provided to users. One of the most significant ways is to generate a program for services, which can manage the changing behavior of buyers and optimize costs for infrastructure maintenance and order.

Ii-B7 Throughput

Throughput measures the rate at which data is successfully transferred between two endpoints. It concerns the ability of a service provider to handle the demands of users as it cannot directly track users’ experience [64]. It is affected by latency and limitations existing at users’ devices and central/edge clouds. Usually, EC has a higher throughput compared with CC, as it takes less time in responding to users’ requests or transferring data to users.

Ii-C Issues in CC and EC

In order to improve the above-mentioned metrics in CC and EC, we need to address the following issues.

Ii-C1 Job Scheduling

The rising demands for services in CC and EC may lead to the imbalanced resource usage and drastically affect the performance of service providers; hence, job scheduling is a necessary prerequisite for improving QoS in CC and EC. The main aim of job scheduling is to order tasks in a specific way. There are two types of job scheduling: static and dynamic scheduling. In static scheduling, all jobs arrive at the same time and are assigned to VMs in a static way, while in dynamic scheduling, all the jobs once arrive, they are scheduled instantly.

Fig. 4: Main paradigms of CI family used in CC and EC.

Ii-C2 Resource Allocation

Improper distributions of resources in central/edge clouds lead to performance degradation of applications. It is becoming more and more difficult to allocate resources accurately to meet the demands of users and the service level agreements. The main aim of resource allocation is to assign the available resources to the demanded cloud/edge applications over the Internet or wireless networks. Resource allocation should be provided to both users and service providers and, as a result, users can access good-quality services through cloud/edge service providers.

Ii-C3 Task Offloading

The rapid growth of Internet services has yielded a variety of computation-intensive applications such as virtual reality applications. If applications are executed in the end devices, it leads to high computing costs, while if they are executed in central/edge clouds, it may take high transmission costs. Thus, users have to decide whether to offload their applications to central/edge clouds or not. Task offloading is the process that maps users’ tasks to suitable resources in the form of VMs to execute. Usually, in CC, users only need to consider whether to offload their tasks or not as there is only one data center. But in EC, users have to further consider where to offload their tasks as there are many edge clouds. Task offloading is a critical issue to improve QoS in CC and EC.

Ii-C4 Joint Issues

A growing number of applications are multi-sensitive. For instance, some tasks are both resource-intensive and delay-sensitive, e.g., face recognition

[133] and augmented reality applications [6]. Researchers started addressing multiple issues together in CC and EC, which are termed as joint issues.

Iii CI for CC AND EC

CI is a developed paradigm of intelligent systems, which is a set of nature-inspired computational methodologies and approaches to address complex real-world problems. It has attracted much attention from researchers and practitioners around the world. We focus on five categories of CI techniques applied in CC and EC, i.e., EAs, SIAs, FS, LBS, and HS, which are given in Fig. 4.

Iii-a EAs

EAs try to mimic the process of natural evolution to find suitable solutions to optimization problems [37]. EAs follow evolutionary mechanisms that vary considerably, however the basic ideas of these mechanisms are the same. Some commonly used EAs for CC and EC are given below.

  • Genetic algorithm (GA) was proposed by Holland in inspired by Darwin’s principle of survival of the fittest [54]. GA operates on a set of individuals called a population. Genetic operators, i.e., crossover and mutation, are applied to produce offspring from the parent population. Then, better individuals are selected for the next generation through a selection operator based on the fitness values. This process continues until the stopping criterion is met. Finally, the best individual is returned as the best solution found for an optimization problem. GA has been widely applied to address issues in CC and EC, such as [55, 2, 78, 18, 12].

  • Artificial immune system (AIS) is a kind of intelligent computational model inspired by the principles of human immune system with the characteristics of self organization, learning, memory, adaptation, robustness, and scalability [144]. In the initialization phase, candidate antibodies are randomly produced to form a population. The affinity of each antibody is evaluated by the fitness function. In each iteration, each antibody is cloned to a number of offspring. Then, these offspring experience the mutation operator. Only the one with the highest affinity is selected. Unlike GA, there is no crossover operator in AIS. AIS has been used for addressing issues in CC and EC, such as [20, 123, 39].

  • Differential evolution (DE) was developed by Storn and Price in [135]. Due to the simple structure, ease of implementation, and robustness, it has been successfully applied to complex optimization problems, such as resource allocation [33] and job scheduling [157]

    in CC and EC. In DE, the first step is to initialize a population randomly. After initialization, a mutant vector is generated by adding the weighted difference vector of individuals to another individual using a mutation operator. Then, a crossover operator is implemented to obtain a trial vector. Afterward, a selection operator is applied to select better individuals for the next generation. Finally, the best individual is returned as the solution of an optimization problem.

Iii-B SIAs

SIAs are a type of nature-inspired algorithms based on interaction among living organisms [106]. They can be described by the collective behaviors of living organisms working under specific rules. Some commonly used SIAs in CC and EC are discussed in the following:

  • Particle swarm optimization (PSO) was proposed by Kennedy and Eberhart [68] in 1995. It is inspired by the social behaviors of the groups of population in nature such as animal herds, bird flocking, or schooling of fish. In PSO, a population of particles is first generated. In each iteration, each particle has an adaptable velocity, according to which it moves in the search space to update position and velocity. Moreover, each particle has a memory, remembering the best position it has ever visited. Thus, its movement is an aggregated acceleration toward its best previously visited position and the best position ever detected/visited by the swarm [104]. This process continues until the stopping criterion is met. PSO has been used to tackle issues in CC [38, 1, 84].

  • Ant colony optimization (ACO), introduced in the early by Dorigo et al. [34, 35], is a technique for optimization. The inspiring source of ACO is the foraging behavior of real ant colonies[14]. In ACO, firstly a finite set of solution components is derived. Secondly, one has to define a set of pheromone values called the pheromone model, which is one of the central components of ACO. In each iteration, ACO assigns higher pheromone values to good solution components probabilistically. ACO has been applied to address job scheduling and resource allocation in CC and EC [89, 154, 103, 60].

  • Bee algorithm (BA), introduced by Pham[109] in 2005, is inspired by the natural foraging behavior of honey bees. In BA, the agent population is divided into a small set of scouts and a larger set of foragers. The scouts randomly sample in the solution space, and evaluate the visited flower patches (i.e., locations). In each iteration, BA compares the new solutions discovered by the scouts with the best-so-far solution. The solutions are ordered according to their fitness values, and the highest ranking ones are selected for local search. BA performs exploitative neighborhood search and random explorative search. BA has been applied to job scheduling in CC and EC [51, 148, 13].

  • Cuckoo search algorithm (CSA) is inspired by the living behaviour of a kind of bird named cuckoo and was developed by Yang and Deb in [167]. In CSA, the adult cuckoos lay eggs in the habitat of other birds. If these eggs are not found and thrown by host birds, they grow and become mature cuckoos. CSA begins with a primary population i.e., a group of cuckoos. In each iteration, CSA performs the following steps. First, individuals are generated by making modifications in existing individuals. After that, the new individuals are evaluated by the fitness function. The new individuals are compared with existing individuals, if they are better than existing individuals, they replace existing individuals in the population. Job scheduling in CC and EC has been addressed by using CSA [137, 98].

  • Firefly algorithm (FA) was developed by Yang [168] in 2008. It is inspired by the flashing pattern and behavior of fireflies. In FA, a population of fireflies is generated randomly. In each iteration, one firefly can be fascinated by other fireflies regardless of its sex. The less brighter firefly is fascinated by the brighter one in the population. The brightness of a firefly is set on the landscape of the fitness function [166]. In case there is no brighter one, the fireflies move randomly and form a new population for the next iteration. The light intensity of fireflies is updated by evaluating the new population. Resource allocation and job scheduling in CC and EC have been tackled by using FA [66, 67, 49].

  • Bacterial foraging algorithm (BFA) is a swarm optimization algorithm inspired by colonies of escherichia coli and proposed by Passino [105]. In BFA, individuals are first generated randomly. Then, for all individuals, the chemotaxis process is carried out. The process in which an organism moves with the gradient of a substance concentration is called chemotaxis. After the chemotaxis process, individuals are reproduced and dispersed. Elimination-dispersion is applied as a final step, in which many random individuals are eliminated and many new individuals are generated. BFA has been used for job scheduling in CC [146, 62].

Iii-C Fs

FS is a type of CI family that resembles human reasoning. FS is a unique system that can handle numerical data and linguistic knowledge at the same time. We have found two types of FS for CC and EC, which are described below.

  • FL was initiated by Zadeh in [174] during the development of theory of fuzzy sets. It works on approximate reasoning rather than exact approximation. The main difference between FL and traditional logic is that FL represents natural language and human thinking. In traditional logic, each member of a binary set has a logical value either 0 or 1, which shows that either it fully belongs to the set or not. There is no concept of partial membership in the traditional set theory [173]. Fuzzy modeling provides the best alternative for fuzziness and obscurities. FL has been applied to address issues in CC and EC, for instance, job scheduling [7, 48, 134], resource allocation [155, 48], and task offloading [122].

  • Fuzzy inference (FI) is a model that can infer a crisp value as an input from a set of inputs, their fuzzy set, membership function, and a set of inference rules [139]. FI is usually constructed as follows. Firstly, inputs for the model are fuzzified through a membership function linked with the predefined fuzzy set. Linguistics terms are used to label the fuzzy set. Afterward, the firing strength is calculated for each rule, showing that some rules are more important than the others in approaching the conclusion. All firing strengths are then aggregated and weighted for all rules, thus producing a fuzzy value. Finally, this fuzzified value is defuzzified by using a proper method. FI has been used for resource allocation [142] and task offloading [71, 113] in CC and EC.

Issue EA Metric Paradigm Reference
Job Scheduling GA execution time CC [55]
GA response time CC [2]
GA energy consumption, cost CC [78]
GA makespan CC [175]

GA
makespan, resource utilization CC [69]
GA makespan, computing cost CC [128]
GA execution time, cost FC [12]
AIS response time FC [156]
Resource Allocation DE resource utilization, makespan CC [33]

GA makespan CC [75]
GA energy consumption, latency MEC [82]
Task Offloading AIS security and privacy CC [20]
GA execution time, energy consumption MCC [31]
GA energy consumption MCC [102]
GA latency FC [18]
AIS security and privacy EC [123]
AIS security and privacy FC [39]
GA latency EC [27]
GA energy consumption MEC [141]
GA execution time, energy consumption MCC [44]
GA energy consumption, latency MEC [17]
Joint issues DE energy consumption MEC [157]
GA energy consumption MEC [45]
GA makespan MEC [73]
TABLE II: EAs for Addressing Issues in CC and EC.

Iii-D Lbs

LBS is inspired by the learning behaviour of living organisms. It uncovers the relationship between the set of nominal features and states or objects [160]. In the following, we present the three most prominent types of LBS.

  • DL is inspired by human’s thinking ability, i.e., the mechanisms of human brain and neurons for processing signals, and was proposed by Hinton

    et al. [52] in 2006. It can cataract layers to extract features from the input data and eventually form a decision [87]

    . DL consists of input layer, output layer, and one or more hidden layers. In the input layer, the state of environment is represented in a suitable numerical form, which can be used as inputs of the network. Afterward, an activation function (e.g., Sigmoid function, Tanh function, or rectified linear unit function) is applied to represent or interrupt the recollection of results. Subsequently, all weights among layers are updated. Finally, the structure of DL is constructed. DL has been used for task offloading in EC

    [25, 58, 118].

  • Reinforcement learning (RL) [138, 130] determines an optimal policy dictating which actions to take at certain states to achieve the highest possible reward. The problem such as resource allocation [97] in CC and EC is often formulated as a Markov decision process (MDP), where there is a set of states and actions. In RL, an agent under a given state selects a suitable action from the set of actions, receives a reward, and moves into a new state where it chooses another action from the updated set of actions. This process continues until a specific stopping criterion is met. RL has been applied to address issues in CC and EC [152, 108, 97, 36, 80, 32].

  • Deep RL (DRL) can be seen as a class of new efficient learning algorithm by combining DL with RL [170, 94, 9]. It is a powerful model that implements DL architectures such as deep neural networks with RL algorithms like Q-learning to scale and solve problems in different areas [42]. It has been effectively used for addressing issues in CC and EC, for instance, job scheduling [74, 178], resource allocation [165], task offloading [91, 23, 177, 100], and joint issues [59, 81, 101]. Two commonly used DRL algorithms are deep Q-learning and deep Q-network.

Iii-E Hs

HS is a system combining two or more CI techniques to overcome their individual shortcomings. For example, genetic-fuzzy system is proposed for automating maritime risk assessment [143], which is a hybrid system of GA and FL. Also, EAs and SIAs are often combined as both have similar behaviors. Similarly, fuzzy adaptive theory [145] is sometimes hybridized with FL and neural networks (NN) [112]. Recently, some hybrid algorithms have been applied to address critical issues in CC [86, 76, 119, 92].

Iv CI in Addressing Critical Issues in CC and EC

This section reviews the CI techniques introduced in Section III to solve the critical issues in CC and EC: job scheduling, resource allocation, task offloading, and joint issues.

Iv-a EAs in CC and EC

This subsection is relevant to the works on the use of EAs in CC and EC, which are summarized in Table II.

Iv-A1 EAs for Job Scheduling

Hu et al. [55] proposed an improved adaptive GA based on a priority mechanism for task scheduling in CC. This algorithm ensures the least execution time for job scheduling and guarantees the QoS requirements of users. It outperforms an adaptive GA and some other GAs in terms of convergence speed, feasibility and effectiveness. Agarwal and Srivastava [2] proposed a GA for task scheduling in CC, which distributes the loads among VMs effectively to optimize the overall response time. Experimental results show that the proposed algorithm outperforms some existing techniques such as greedy-based techniques and First Come First Serve in terms of overall response time. Liu et al. [78] developed a job scheduling model in CC based on multi-objective GA, which aims to minimize the energy consumption and maximize the profit of service. The proposed model has several components to analyze the applications and to allocate the appropriate resources to the applications. Experimental results show that the proposed model can obtain a higher profit, while consuming lower energy. Zarina et al. [175] investigated GA-based optimal job scheduling and load balancing in CC. They combined a GA with three traditional scheduling techniques: min-min, max-min and suffrage. In the first phase, these three traditional scheduling techniques are applied to obtain the minimum completion time for a given job at each VM. After that, GA is applied to attain better QoS by utilizing available resources. Experimental results reveal that the proposed algorithm outperforms min-min, max-min, suffrage, and First Come First Serve in terms of makespan. Kumar and Verma [69] presented an improved GA by using min-min and max-min techniques in the original GA for scheduling tasks in CC. In this improved GA, the initial population is generated by using min-min and max-min techniques. Makespan is considered as the fitness function. This improved GA is applied for two cases. In the first case, the number of VMs is kept constant and the number of cloudlets is varied, while in the second case, the number of cloudlets is fixed and the number of VMs is varied. Experimental results show that it performs better than the simple GA in terms of makespan and resource utilization. Shaminder et al. [128] proposed a modified GA by merging two existing scheduling algorithms with standard GA for scheduling tasks in CC. In the proposed algorithm, the initial population is generated by using the output schedules of two algorithms (i.e., longest cloudlet to fastest processor and smallest cloudlet to fastest processor) and

random schedules. To achieve time minimization and compare it with existing heuristics, a fitness function is formulated for single-user jobs. Experimental results show that the proposed algorithm performs better than standard GA in terms of response time.

Binh et al. [12] proposed a GA-based algorithm for solving task scheduling in cloud-FC. The proposed algorithm tries to achieve the optimal tradeoff between the execution time and operating costs by addressing different tasks in cloud-FC. Experimental results demonstrate that the proposed algorithm outperforms BA in terms of time-cost tradeoff. Wang et al. [156] studied an immune scheduling network-based method for task scheduling in FC. The proposed method uses forward and backward propagation in the ad hoc network along with the power of distributed schedulers to generate the optimized scheduler strategies to deal with computing node overloaded and achieve the optimal task finishing time reducing. Experimental results reveal that the proposed method performs better than the modified critical path, dynamic critical path, dominant sequence clustering, and GA.

Iv-A2 EAs for Resource Allocation

Dolly [33] used DE for resource allocation in CC. The proposed method avoids premature convergence effectively, reduces the makespan, and increases the resource utilization. The proposed method outperforms PSO in terms of makespan, resource utilization, and load balancing. Lin and Zhong [75] presented a GA-based strategy for computing resource allocation in CC. They incorporated enhancements and local search into GA for solving computing resource allocation in CC. The computation time is focused in this work, which is an important factor in cloud manufacturing for fast response to users’ requests. The proposed method is compared with other algorithms like PSO, BA, ACO, etc. Experimental results show the effectiveness of the proposed method in terms of makespan. Luo et al. [82] proposed a GA-based cashing placement strategy to minimize energy consumption in MEC. A joint optimization problem is formulated by considering energy consumption, backhaul capacities, and content popularity distributions. A GA is applied to solve this complicated joint optimization problem. Simulation results show that the proposed algorithm effectively determines the near-optimal caching placement and obtains better performance in terms of energy efficiency compared with the conventional caching placement strategies.

Iv-A3 EAs for Task Offloading

Chen and Yang [20] presented a data security strategy based on AIS in CC. They discussed the main factors affecting data security, introduced a Hadoop distributed file system, and applied AIS with negative selection and dynamic selection in CC. Simulation results show that the proposed strategy performs better than GA, PSO, and simulated annealing. Deng et al. [31] proposed a GA-based computation task offloading approach for MCC. This approach is designed for robust offloading decision to optimize execution time and energy consumption of mobile services. Numerical results demonstrate that with near-linear algorithmic complexity, the proposed approach can produce near-optimal solutions. Kaushik and Kumar [102] applied GA to a framework designed for elastic applications to reduce communication and computation energy by finding the optimum offloading solution. They focused on augmented execution of mobile applications and formulated an optimization problem by minimizing the cost function, which is the combination of communication energy and computation energy. Simulation results demonstrate that the proposed approach outperforms all mobile-side execution and all cloud-side execution.

Issue SIA Metric Paradigm Reference
Job Scheduling
PSO
throughput, response time CC [38]
PSO makespan, cost CC [1]
ACO makespan CC [89]
ACO energy consumption, resource utilization MCC [154]
ACO resource utilization CC [181]
ACO energy consumption MCC [158]
BA makespan CC [51]
BA response time, execution time CC [148]
CSA setup cost, running cost CC [137]


FA
energy consumption, resource utilization CC [66]
FA makespan CC [67]
BFA makespan, computation cost, resource utilization CC [146]
BFA cost, makespan CC [62]

BA execution time FC [13]
CSA response time, cost CC & FC [98]
ACO energy consumption, waiting time CC [103]
ACO energy consumption, resource utilization CC [79]
Resource Allocation
FA
resource utilization, cost CC & FC [49]

Task Offloading
PSO energy consumption, latency CC [84]
ACO response time MCC [10]
Joint Issues ACO energy consumption MEC [60]
TABLE III: SIAs for Addressing Issues in CC and EC.

Canali et al. [18] proposed a GA for service placement in FC. They studied mapping data streams over fog nodes and presented an optimization model. Then, a scalable heuristic based on GA is adopted to tackle this model. Experimental results verify the stability and performance of the proposed heuristic. Roman et al. [123] proposed an AIS-based strategy for the IoT system using edge technologies. In the proposed strategy, the requirements of immune system for IoT are analyzed and a security architecture is proposed to meet the demands of users. It makes a decision on the number and type of VMs deployed at the edge infrastructure. Farhoud et al. [39] proposed a new distributed and lightweight intrusion detection system in FC based on an AIS model. In this paper, the intrusion detection system is distributed to three layers: cloud, fog, and edge layers. Intrusion detection system, primary network traffic clustering, and detectors’ training are performed in the cloud layer. Intrusion alerts are analyzed by using smart data concept in the fog layer. Detectors are deployed in edge clouds in the edge layer. Experimental results show the efficiency of the proposed system. Cheng et al. [27] proposed a fast heuristic algorithm based on GA for just-in-time code offloading for wearable computing. The proposed algorithm is compared with offloading nothing, offloading all to cloud, and simple greedy offloading. Numerical results show that the proposed algorithm outperforms the three competitors significantly. Tang et al. [141] suggested a task caching and migration strategy in MEC based on GA to satisfy the completion time constraints and the goal of minimizing energy consumption. They adopted a fine-grained task partitioning migration model to transform users’ tasks into directed graphs with multiple subtasks. The task caching is proposed to further reduce the delay and energy consumption. Simulation results reveal that the proposed strategy can greatly reduce the energy consumption of end devices compared with all tasks being executed in the edge or end devices. Goudarzi et al. [44] proposed a GA-based algorithm for multi-site computation offloading in MCC. They modified genetic operators to reduce ineffective solutions and evaluated the efficiency of the proposed algorithm using graphs of real mobile applications. Experimental results demonstrate that the proposed algorithm outperforms other existing algorithms in terms of execution time, energy consumption, and weighted cost in a timely manner. Bozorgchenani et al. [17] proposed a multi-objective EA, i.e., non-dominated sorting GA, to find the optimal offloading decisions in MEC. They modeled the task offloading as a constrained multi-objective optimization problem that jointly minimizes the task processing delay and the energy consumption of mobile devices. Experimental results reveal that the proposed approach outperforms other existing approaches in terms of energy consumption and task processing delay.

Iv-A4 EAs for Joint Issues

Wang et al. [157] studied a multi-unmanned aerial vehicle (UAV) enabled MEC, where the delay-sensitive task can be executed on the local device or one of UAVs. This paper designs a two-layer optimization method for jointly optimizing the deployment of UAVs and task scheduling, where DE and the greedy method are adopted at the upper layer and the lower layer, respectively. Guo et al. [45] presented a computation offloading model in the multi-access and multi-channel interference MEC. The offloading decision, channel allocation, and computation resource allocation are formulated as a mixed-integer nonlinear programming problem. A suboptimal algorithm, i.e., a GA-based computation algorithm, is proposed to solve this large-scale and NP-hard problem. Simulation results demonstrate that the proposed algorithm outperforms a PSO-based computation algorithm, a random computation algorithm, and a local computation algorithm in terms of energy consumption. Li and Zhu [73] proposed a joint optimization method based on GA to optimize offloading proportion, channel bandwidth, and mobile edge servers’ computing resource allocation in MEC. Simulation results demonstrate that the proposed algorithm can effectively reduce the task completion time and guarantee fairness among users.

Iv-B SIAs in CC and EC

This subsection discusses the works on the use of SIAs for solving issues in CC and EC, as summarized in Table III.

Iv-B1 SIAs for Job Scheduling

Elina et al. [38] described a two-level cloud scheduler by using PSO, which operates under the IaaS model. They explored whether the use of a priority-based policy at the VM-level is suitable in an online cloud. Experimental results show that the proposed algorithm offers a good balance between throughput and response time. Guo et al. [1] proposed a PSO algorithm by incorporating a small position value rule to minimize processing cost in CC. Experimental results show that the proposed algorithm outperforms two other PSO-based algorithms in terms of convergence, processing time, and processing cost. Medhat et al. [89] presented an ACO algorithm for task scheduling in CC. At first, parameters with better values are determined for ACO through experiments. Then, ACO is evaluated on a set of instances with up to 1000 tasks. Wang et al. [154] investigated MCC-assisted execution of a multi-task scheduling problem in a hybrid MCC architecture, and formulated it as an optimization problem with time constraints. Cooperative multi-task scheduling based on ACO is used to solve this optimization problem by considering task profit, task deadline, task dependence, node heterogeneity, and load balancing. Zhang and Zhang [181]

applied an ACO algorithm for load balancing in a complex network named ACCLB for CC. The purpose of this study is to cope with the dynamic load balancing problem in open CC federation. Experimental results show that the proposed method performs better than SearchMax-SearchMin and classic ACO in terms of minimizing standard deviation and gains a more suitable distribution of the workloads on the whole cloud federation. Wei

et al. [158] studied application scheduling in MCC with load balancing by using ACO. They presented a hybrid local mobile cloud model by extending the cloudlet architecture. The proposed model can select applications with the maximum profit and minimum energy consumption in a heavy load environment.

Hesabian et al. [51] presented a scheduling method in CC based on BA to dedicate the sources optimally. The proposed method behaves like a load balancing BA for small-scale systems in terms of makespan, while outperforms it in large-scale systems. Walaa et al. [148] proposed a BA for load balancing in CC, which distributes workloads of multiple network links to avoid under utilization and over utilization of the resources. The proposed algorithm can minimize the response time and data center processing time by distributing workloads on different VMs based on the availability and load of each VM. It performs better than modified throttled and round robin algorithms in terms of average response time and execution time. Supacheep et al. [137] applied CSA for job scheduling in CC. They considered only static cases of job scheduling. The results of CSA are better than those of GA in most cases and it can find larger number of feasible solutions than GA. Kansal and Chana [66] proposed an energy-aware VM migration technique based on FA for CC, which migrates the maximally loaded VM to the least loaded active node while maintaining the performance and energy efficiency of the data centers. This technique is compared with ACO-based and FFD-based techniques in the Cloudsim simulator. Kaur and Sherma [67] applied FA to address workflow scheduling in CC, which reduces the execution time by efficient scheduling of workflow. The proposed FA performs better than other SIAs for workflow scheduling in terms of makespan. Verma et al. [146] proposed an improved BFA for scheduling tasks in CC. The performance of this algorithm is evaluated by using CloudSim toolkit. Experimental results demonstrate that it outperforms BFA and two other algorithms in terms of makespan, computation cost, and resource utilization. Jacob et al. [62] presented a BFA for resource scheduling in CC. In this work, a hyper-heuristic-based scheduling algorithm is used in the cloud system to map the resources in an efficient way. The proposed BFA reduces the cost and makespan and outperforms some existing algorithms like GA, ACO, a priority-based algorithm, and a Berger model-based algorithm.

Issue FS Metric Paradigm Reference
FL response time, throughput, resource utilization CC [7]
Job Scheduling FL latency CC [134]
FL latency, cost CC [115]

FL
latency CC [61]
FL resource utilization CC [93]
Resource Allocation FL resource utilization, cost CC [48]
FL cost CC [155]
FI resource utilization CC [142]

Task Offloading
FL security and privacy CC [122]
FI security and privacy CC [113]
FI security and privacy, execution time MEC [71]
Joint Issues FL resource utilization, response time EC & CC [132]
FL response time, processing time CC [183]

TABLE IV: FS for Addressing Issues in CC and EC.

Bitam et al. [13] proposed a BA for job scheduling in FC, in which two factors are considered: the CPU execution time and the allocated memory required by the overall tasks. The reliability and efficiency of the proposed BA are evaluated by performing a set of tests. It is superior to GA and PSO in terms of allocated memory and execution time. Nazir et al. [98] proposed a CSA for balancing load of users’ requests in residential areas in cloud-FC. The performance of CSA is compared with that of some existing techniques like round robin and throttled algorithms. Pang et al. [103] designed an ACO algorithm to address dynamic energy management in cloud data center. They used Petri net for analyzing scheduling process, and then a task-oriented resource allocation method is proposed to optimize the running time and energy consumption of the system. Liu et al. [79] proposed an algorithm called OEMAC based on ACO to minimize energy for VM placement in CC. OEMAC minimizes the number of active servers. Experimental results show that OEMAC outperforms conventional heuristics and some other EAs in terms of saving energy, improving resource utilization, minimizing the number of active servers, and balancing different resources.

Iv-B2 SIAs for Resource Allocation

Kanza et al. [49] utilized a FA for efficient resource allocation in cloud-FC, in which load balancing and cost reduction are considered.

Iv-B3 SIAs for Task Offloading

Deng et al. [84] adopted discrete PSO to search the optimal offloading policy in cloud-enhanced small cell networks, with the aim of minimizing energy consumption under strict delay constraints. The energy-efficient task offloading problem is formulated as a constrained 0-1 programming problem. Experimental results show that the proposed approach outperforms local execution and conventional rough-granularity offloading policy in terms of energy saving. Bao et al. [10] proposed an ACO-based method for addressing computation offloading problem in MCC. Different rules are adopted to provide services to a large number of service requests.

Iv-B4 SIAs for Joint Issues

Huang et al. [60] studied an ACO-based bilevel optimization approach for joint offloading decision and resource allocation in cooperative MEC to attain energy-efficient task execution under delay constraints. The effectiveness of the proposed approach is demonstrated by comparing it with four other algorithms.

Issue LBS Metric Paradigm Reference
Job Scheduling RL makespan CC [108]
DRL latency CC [74]
DRL energy consumption MEC [178]
Resource Allocation RL response time, resource utilization CC [36]
RL latency FC [97]
RL energy consumption, latency EC [80]
DRL latency MEC [165]
DL response time, resource utilization EC [83]
DRL response time MEC [150]
DRL resource utilization, makespan MEC [164]
Task Offloading DRL resource utilization, latency MCC [114]
RL resource utilization, latency MCC [136]
DL security and privacy MEC [25]
DL energy consumption, makespan Cloudlet [118]
RL energy consumption MEC [32]
DRL energy consumption, latency MEC [91]
DRL latency, cost MEC [23]
RL processing time, latency MEC [177]
DRL energy consumption FC [100]
RL energy consumption MEC [117]
DRL latency MEC [149]
DRL latency, energy consumption MEC [22]
DRL latency, energy consumption, computation cost MEC [26]
Joint Issues DL energy consumption, cost MEC [58]
DRL cost, energy consumption, latency MEC [59]
DRL latency, resource utilization EC [81]
DRL energy consumption, execution time EC [101]
RL security and privacy, energy consumption MEC [163]
DRL latency, energy consumption MEC [72]
DRL latency MEC [57]
TABLE V: LBS for Addressing Issues in CC and EC.

Iv-C FS in CC and EC

The works devoted to the use of FS in CC and EC are discussed in this subsection and summarized in Table IV.

Iv-C1 FS for Job Scheduling

Anindita [7] proposed a new dynamic task scheduling approach in CC based on FL, which takes two inputs: the time required to complete the tasks in each VM and the number of requests received from each VM. These inputs produce an output i.e., id of VM which is assigned to the host. Experimental results show that the proposed algorithm outperforms First Come First Serve, round robin, and BA in terms of throughput, response time, and resource utilization. Srinivas et al. [134] proposed an efficient load balancing algorithm by using FL based on round robin load balancing technique for attaining better resource utilization in CC. For evaluating the balanced load through FL, two parameters are used: processor speed and assigned load of a VM. Numerical results reveal that the proposed algorithm outperforms round robin load balancing technique in terms of minimizing the processing and response time. Ragmani et al. [115] proposed an improved scheduling strategy based on FL in CC to evaluate processing time, response time, and total cost. FL uses the number of VMs, bandwidth, the number of processors, processor speed, the size of data, and request per user per hour as inputs. Then, the fuzzy controller calculates the global performance indicator. Issawi et al. [61] proposed an efficient adaptive load balancing algorithm in CC based on FL and round robin/random assignment, which consists of three main components: burst detector, load balancing algorithm, and fuzzifier. When a request is received in the data center, it is detected by the burst detector as normal or burst workload state. Then, a load balancing algorithm (round robin in burst and random assignment in non-burst state) is selected, which assigns the received task to an appropriate VM based on the information provided by the fuzzifier. Experimental results show that the proposed algorithm reduces the response and processing time. Mondal et al. [93] used FL to improve QoS in CC by balancing the imposed load in the system. This paper considers the speed of processor and load as inputs and load balancing as output.

Iv-C2 FS for Resource Allocation

Haratian et al. [48] proposed an adaptive and fuzzy resource management approach called AFRM for allocating resources in CC. In AFRM, the last resource values of each VM are collected through the environment sensors and sent to a fuzzy controller. Then, AFRM analyzes the received information to make a decision on how to reallocate the resources in each iteration of a self-adaptive control cycle. Experimental results show that AFRM outperforms rule-based and static-fuzzy approaches in terms of resource allocation efficiency, utility, the service level agreement violations, and cost. Wang et al. [155] proposed a strategy for resource allocation based on improved fuzzy clustering in CC. They first divided the set of resources in CC into different resource pools on the service level of users by using improved fuzzy clustering, and then generated a resource scheduling scheme by using a scheduling algorithm. The proposed strategy shows certain advantages in terms of the number of iterations and the accuracy of classification compared with other algorithms. Tao et al. [142] designed a containerized test environment in CC and developed a node selection algorithm based on FI for container deployment, where FI is applied to dynamically predict the most proper node for the deployment of the selected containers. Experimental results show that the proposed algorithm performs better than existing container deployment algorithms.

Iv-C3 FS for Task Offloading

Ritu and Jain [122] presented a trust model in CC by using FL. The proposed model makes use of turnaround time, availability, and reliability for evaluating trust in CC. Qu and Buyya [113] proposed a cloud trust evaluation system by using hierarchical FI for service selection in CC. To facilitate service selection, this system evaluates the trust of clouds according to users’ fuzzy QoS requirements and services’ dynamic performance. Simulations and case studies demonstrate the effectiveness and efficiency of this system. Li et al. [71] designed an architecture in MEC to decouple the security functions with physical resources and developed a FI-based algorithm to find the optimal order of the required security functions. Numerical results show that the proposed algorithm performs better than a widely used fuzzy-based simple additive weighting algorithm in terms of inverted generational distance values and execution time.

Iv-C4 FS for Joint Issues

Sonmez et al. [132] proposed an FL-based approach for workload orchestration in EC, where execution locations for incoming tasks from mobile devices are decided within an EC infrastructure. Simulation results demonstrate that the proposed approach performs better than other algorithms for the cases studied in terms of resource utilization and response time. Zulkar et al. [183] presented a FL-based dynamic load balancing algorithm in virtualized data centers, which can efficiently predict the VM where the next job is scheduled. They modeled the requirements of memory, bandwidth, and disk space using FL. Simulation results demonstrate that the proposed algorithm outperforms other scheduling algorithms in terms of response and processing time in the data centers.

Iv-D LBS in CC and EC

In this subsection, we review the works on the use of LBS in CC and EC as summarized in Table V.

Iv-D1 LBS for Job Scheduling

Peng et al. [108] proposed a RL-based mixed job scheduler scheme for CC, which considers accurate scaled CC environment and efficient job scheduling under VM resource and service level agreement constraints. The proposed scheme outperforms fast-fit, best-fit, min-min, and max-min scheduling schemes in terms of makespan. Lin et al. [74] presented a multi-resource cloud job scheduling strategy in CC based on current popular DRL and deep Q-network, which aims to reduce the average job completion time and average job slowdown. In this strategy, the convolutional NN is adopted to perceive the system resources, job state features, and RL decision-making capabilities to solve online awareness decision problems in CC. Based on the experimental results, the proposed strategy performs better than classical heuristic algorithms and converges faster than a standard policy gradient algorithm. Zhang et al. [178] combined a stacked auto-encoder with a Q-learning model to design a deep Q-learning model for energy-efficient scheduling in real-time systems. The main function of the stacked auto-encoder is to replace the Q-function for learning the Q-value. The proposed model can save 4.2% energy compared with hybrid dynamic voltage and frequency scaling scheduling based on Q-learning for different sets of tasks.

Iv-D2 LBS for Resource Allocation

Dutreilh et al. [36] studied RL for autonomic resource allocation in CC, in which proper initialization is adopted at the early stages and convergence speedups are applied in the learning phases. Nassar and Yilmaz [97] proposed a RL-based resource allocation algorithm in fog radio access networks. They formulated the resource allocation problem as an MDP in two alternative formulations: infinite-horizon MDP and finite-horizon MDP. Experimental results show that the proposed algorithm outperforms the fixed-threshold method. Liu et al. [80] proposed a RL approach based on -greedy Q-learning for resource allocation in EC. Experimental results show the effectiveness of the proposed approach in terms of minimizing energy consumption and latency. Yang et al. [165] proposed DRL-based resource allocation in MEC. They allocated computation resources by investigating different strategies in MEC networks that operate with finite block length codes to support low-latency communications. Simulation results show that the proposed algorithm outperforms the random and equal scheduling benchmarks. Luong et al. [83] proposed an optimal auction based on DL in the edge resource allocation. They constructed a multi-layer NN architecture based on analytical solution of the optimal auction. The NN first performs monotone transformations of the miners’ bids. Then, allocation and conditional payment rules are calculated for the miners. Simulation results show that the proposed scheme can quickly converge to a solution. Wang et al. [150] proposed a DRL-based approach for smart resource allocations including computing resource allocation and network resource allocation in MEC. They considered two aspects: average service time minimization and resource allocation balancing. Experimental results reveal that the proposed approach outperforms the traditional open shortest path first algorithm. Xiong et al. [164] proposed a DRL-based approach for resource allocation of IoT in EC. They formulated the resource allocation problem as an MDP. They also proposed an improved deep Q-network algorithm for policy learning, where multiple replay memories are applied to separately store the experiences with small mutual influence. Simulation results show that the proposed algorithm outperforms the original deep Q-network algorithm in terms of convergence, and that the corresponding policy performs better than other policies regarding completion time.

Iv-D3 LBS for Task Offloading

Quan et al. [114] proposed a two-layered RL algorithm for task offloading with a tradeoff between physical machine utilization rate and delay in MCC. The k-nearest neighbors algorithm divides the physical machines into many clusters. The first layer selects a cluster via learning the optimal policy, while the second layer learns an optimal policy to choose the optimal physical machine to execute the current offloaded task. Numerical results show that the proposed algorithm is faster than DRL when learning the optimal policy for task offloading. Sundar and Liang [136] proposed a game and learning approach for multi-user computation offloading in MCC. This paper discusses both online and offline computation task offloading from multiple users to a cloud or nearby cloud at the edge. The offline algorithm provides a better average solution while online algorithm is much faster.

Issue HS Metric Paradigm Reference
Job Scheduling GA-PSO makespan, cost, execution time CC [86]
GA-ACO execution time CC [76]
GA-ACO energy consumption, makespan, resource utilization MCC [119]
ACO-CSA energy consumption, makespan CC [92]
Resource Allocation RL-ACO throughput MEC [147]
Task Offloading NN-PSO security and privacy CC [124]
Joint Issues GA-PSO energy consumption MEC [46]
DRL-PSO-FL energy consumption MEC [65]
TABLE VI: HS for addressing issues in CC and EC.

Chen et al. [25]

proposed a DL-based model to detect security threats in MEC. This model uses unsupervised learning and location information to improve the detection process, which can detect malicious applications at the edge of a cellular network. Numerical results demonstrate that the proposed model outperforms softmax regression, decision tree, support vector machine, and random forest. Rani and Pounambal

[118] proposed DL-based dynamic task offloading in mobile cloudlet, which considers energy consumption and execution time as objective metrics. The task computed on cloudlet or cloud server is divided into subtasks. Experimental results show that the proposed algorithm outperforms cloudlet-based dynamic task offloading in terms of energy consumption and completion time. Dinh et al. [32] proposed a RL-based computation offloading scheme in MEC to reduce energy consumption. They studied multi-user multi-edge-node computation offloading problem, and formulated it as a non-cooperative game where each user maximizes its own utility. Experimental results show that the proposed scheme outperforms local processing and random assignment. Meng et al. [91] proposed a DRL-based task offloading algorithm to minimize the mean slowdown of tasks and energy consumption of MEC. A new reward function is designed to optimize the tradeoff between average slowdown and average energy consumption. Numerical results reveal that the proposed algorithm performs better than the baseline algorithms such as all in MEC server, all in mobile device, and random in terms of average energy consumption and average slowdown.

Chen et al. [23] presented two DRL algorithms for optimizing computation offloading performance in virtual EC, where the stochastic computation offloading problem is formulated as an MDP. Numerical results show that the proposed algorithms outperform mobile execution, server execution, and greedy execution in terms of computation offloading performance. Zhang et al. [177] proposed a DRL-based task offloading scheme for vehicular edge computing networks (VECNs), in which the central cloud server is considered as a backup server due to its powerful computation capacities. The task offloading problem is formulated as a processing time minimization problem with delay constraints. Ning et al. [100] studied DRL for intelligent Internet of vehicles. They constructed an offloading framework consisting of three layers (i.e., cloudlet, RodeSide Units, and fog nodes) to minimize the total energy consumption under the delay constraint. The proposed method outperforms the baseline algorithms in terms of average energy consumption. Ranadheera et al. [117]

proposed a computation offloading approach based on game theory and RL in MEC to reduce energy consumption.

Wang et al. [149] proposed a DRL-based method for computation offloading in MEC. The offloading problem in MEC is formulated as an MDP and the S2S neural network is designed to represent the offloading policy. Simulation results show that the proposed method performs better than two heuristic baselines in terms of latency, and can obtain nearly optimal results while having polynomial time complexity. Chen et al. [22] proposed a DRL-based approach for performance optimization in MEC. They considered MEC for a representative mobile user in an ultra-dense network, where one of multiple base stations can be selected for computation offloading. They modeled an optimal computation offloading policy as an MDP and developed a deep Q-network-based strategic computation offloading algorithm to learn the optimal policy without having any priori knowledge of the dynamic statistics. Chen and Wang [26] proposed a DRL-based approach for decentralized computation offloading in MEC to minimize the long-term average computation cost in terms of power consumption and buffering delay. They adopted continuous action space-based DRL to learn efficient computation offloading policies independently at each mobile user. Experimental results demonstrate that the proposed approach outperforms conventional deep Q-network-based discrete power control strategy and some other greedy strategies in terms of computation cost.

Iv-D4 LBS for Joint Issues

Huang et al. [58] proposed distributed DL-based offloading in MEC. They formulated joint offloading decision and bandwidth allocation as a mixed-integer programming problem. Multiple parallel deep NNs are used to generate offloading decisions. Numerical results reveal that the proposed algorithm outperforms deep Q-network in terms of total cost and energy consumption. Huang et al. [59] further studied the same joint issue in multi-user MEC by adopting the idea of deep Q-network. Experimental results show that the proposed algorithm performs better than the MUMTO algorithm in [21] in terms of overall cost, energy consumption, and delay. Liu et al. [81] formulated offloading and resource allocation in VECNs as a semi-Markov process, by considering stochastic vehicle traffic, dynamic computation requests, and time-varying communication conditions. Two RL methods, i.e., a Q-learning based method and a DRL method, are designed to get the optimal policies for computation offloading and resource allocation. Ning et al. [101]

constructed an intelligent offloading system for VECNs by using DRL. They used Markov chains to model communication and computation states. To improve users’ quality of experience, task scheduling and resource allocation are formulated as a joint optimization problem. To schedule offloading requests and allocate network resources, this paper designs a two-sided matching scheme and a DRL approach. Experimental results show that the proposed method outperforms Q-learning, a greedy method, local computing, and deep Q-network in terms of quality of experience and execution time. Xiao

et al. [163] applied RL to provide secure offloading to the edge nodes against jamming attacks in MEC. Lightweight authentication and secure collaborative cashing schemes are presented for securing data privacy. Simulation results show that the proposed RL-based secure solution can effectively enhance the security and user privacy of MEC and protect it against various smart attacks with low overhead. Li et al. [72] proposed a DRL-based computation offloading and resource allocation scheme for MEC. In order to minimize the sum cost of delay and energy consumption for all user equipments in MEC, they jointly optimized the offloading decision and computing resource allocation. Simulation results reveal that the proposed scheme performs better than other baselines. Huang et al. [57] proposed a DRL-based method for online computation offloading in wireless powered MEC networks to maximize the weighted sum computation rate with binary computation offloading. They optimally adapted task offloading decisions and wireless resource allocations to the time-varying wireless channel conditions. Simulation results demonstrate that the proposed algorithm achieves similar near-optimal performance as existing benchmark methods but reduces the CPU execution latency by more than an order of magnitude.

Iv-E HS in CC and EC

This section discusses the works on the use of HS for solving issues in CC and EC, which are summarized in Table VI.

Iv-E1 HS for Job Scheduling

Manasrah and Ali [86] proposed a hybrid GA-PSO algorithm for workflow task scheduling in CC. It outperforms GA, PSO, and some other CI techniques in terms of total execution time of the workflow task. Liu et al. [76] proposed a hybrid GA-ACO task scheduling algorithm in CC, which takes advantage of fast convergence from ACO and global search ability of GA. Simulation results show that it outperforms GA and ACO in terms of task execution time. Rashidi and Sharifian [119] presented a hybrid GA-ACO algorithm for task scheduling in MCC. Experimental results show that it outperforms queue-based round robin, queue-based random, and queue-based weighted round robin assignment algorithms in terms of average completion time, total energy consumption of mobile devices, and the number of dropped tasks. Moganarangan et al. [92] combined ACO with CSA to address job scheduling in CC. The proposed algorithm performs better than ACO in terms of energy consumption and makespan.

Iv-E2 HS for Resource Allocation

Vimal et al. [147] proposed a hybridization of RL and multi-objective ACO to enhance resource allocation in MEC for industrial IoT. The proposed algorithm allocates resources accurately and optimally for users in MEC. Experimental results show that the proposed algorithm outperforms GA and BA in terms of throughput.

Iv-E3 HS for Task Offloading

Saljoughi et al. [124] presented a hybrid NN-PSO algorithm to detect intrusions and attacks in CC. For extraction of the optimal weights of NN, the weights of NN are optimized by using PSO. The proposed method obtains better outcomes than the simple NN on NSL-KDD and KDD-CUP databases.

Iv-E4 HS for Joint Issues

Guo et al. [46] studied the energy-efficient computation offloading management scheme based on hybrid GA-PSO in MEC with small cell networks. They jointly optimized computation offloading, spectrum, power, and computation resource to minimize the energy consumption of all users’ equipments. They presented the computation offloading model and formulated the problem as a mixed-integer nonlinear programming problem. Simulation results show that the proposed algorithm performs better than other baseline algorithms. Jiang et al. [65] presented a hybrid DL-PSO-FL algorithm for online offloading to minimize the energy consumption of users’ equipments in a hybrid MEC network. In the proposed algorithm, FL is used to locate the ground vehicles and UAVs, PSO is used to solve the mixed-integer nonlinear programming problem and provides high-quality samples to DNN, and DL is applied to make the task admission and resource allocation decision in real time. Simulation results reveal that the proposed algorithm reduces the CPU time by more than several orders of magnitude.

Fig. 5: Overall percentage distributions of CI techniques for addressing issues in CC and EC.
Fig. 6: Suitability of CI techniques for addressing issues in CC and EC.
Fig. 7: Distribution of EAs in CC and EC.
Fig. 8: Distribution of SIAs in CC and EC.
Fig. 9: Distribution of FS in CC and EC.
Fig. 10: Distribution of LBS in CC and EC.

V Discussion and Future Research Trends

Recent literature shows that researchers have paid much attention to the innovative use of CI techniques to address issues in CC and EC, such as job scheduling, resource allocation, task offloading, and joint issues. However, there are still open challenges for researchers to tackle. This section provides the statistics which reflect the status-quo of CI for CC and EC, and points out future research trends.

V-a Discussion

Fig. 5 presents the overall percentage distributions of CI techniques mentioned in Fig. 4 for solving issues in CC and EC. Fig. 6 presents all issues, CI techniques, and the number of papers found for each issue in CC and EC. Furthermore, we elaborate on the main research findings derived from Section IV.

V-A1 EAs for CC and EC

From Fig. 5, EAs are the second most commonly used techniques among all types of CI techniques. Most of the works found in this survey are on the use of EAs for job scheduling and task offloading as shown in Fig. 7. Furthermore, the most frequently used EA in CC and EC is GA as shown in Fig. 7. Fig. 7 also presents the distribution of EAs, which indicates the suitability of each kind of EAs for each issue in CC and EC.

V-A2 SIAs for CC and EC

SIAs have been used in 22% papers compared with other CI techniques as shown in Fig. 5. We can observe from Fig. 8 that job scheduling is the hottest issue addressed by SIAs and ACO is the most frequently used SIA in CC and EC. Fig. 8 also shows the distribution of SIAs in the context of issues in CC and EC.

V-A3 FS for CC and EC

It can be seen from Fig. 5 that 14% papers employ FS to address issues in CC and EC. FS has been frequently used for job scheduling, resource allocation, and task offloading as shown in Fig. 9. Fig. 9 presents the distribution of FS in the context of issues in CC and EC.

V-A4 LBS for CC and EC

LBS is the most commonly studied CI technique in CC and EC as shown in Fig. 5. In particular, it has been used more intensively for task offloading, resource allocation, and joint issues in CC and EC from Fig. 10. In addition, DRL is the most frequently used LBS as shown in Fig. 10. Fig. 10 gives the distribution of LBS in CC and EC.

V-A5 HS for CC and EC

HS has not been utilized frequently (only 8% papers) in CC and EC as depicted in Fig. 5. Specifically, half of papers related to HS studies on job scheduling as shown in Fig. 6.

Remark: Based on the above discussion, we would like to give more details to understand what kinds of CI techniques are good at solving what kinds of issues:

  • GA is the best choice for solving job scheduling and task offloading among all EAs [27, 12, 175, 46]. The reasons are twofold: 1) it is the most popular and frequently used EA paradigm; and 2) it has various crossover and mutation operators that can deal with different optimization problems, such as discrete and continuous optimization problems.

  • ACO is the best SIA for job scheduling among all the issues studied in this paper [181, 158, 154], due to the following reasons: 1) it can consider/incorporate multiple scheduling targets/metrics, such as load balancing, energy consumption, makespan, and cost; 2) each variable in the solution obtained by ACO is generated one by one; thus, jobs can be assigned to VMs one by one to optimize the scheduling process; and 3) it can ensure the fast convergence and good performance by balancing the exploration of new solutions and exploitation of accumulated experience about the problem [79].

  • Compared with FI, FL is better for solving job scheduling, resource allocation, and joint issues [7, 134], since it is flexible, easy to understand, compatible with the uncertainty of CC and EC parameters as well as users’ behaviors, and can deal with imprecise data and complex problems with several variables.

  • DRL performs the best among all LBS for resource allocation, task offloading, and joint issues [72, 83, 164, 101]. The superiority of DRL can be summarized as follows: 1) it can automatically adapt and customize itself according to users’ requirements [150]; 2) it can discover/learn new knowledge from large databases; 3) it can develop models that are difficult and expensive to be designed manually due to the requirements of specific skills; and 4) it has a strong ability to handle complex problems by efficiently learning from experiences.

V-B Future Research Trends

V-B1 From CC and EC Perspective

  • Many-Metric Formulation: More metrics need to be considered at the same time in CC and EC to create the trade-off among them. For example, in the future, we may consider the trade-off among latency, cost, energy consumption, security, and privacy to satisfy QoS of users.

  • Edge Node Allocation: Due to the distributed nature of EC, edge clouds that offer services across diverse geolocation and regions are difficult to be allocated [153]. In the future, efficient service discovery protocols are needed to design, such that users can identify and locate the relevant service providers to meet their demands.

  • Real-Time Optimization: For many edge application scenarios, the service environments are of high dynamics [18] and it is hard to correctly foresee future events. Thus, it would require the remarkable capabilities of online edge resource orchestration and provisioning to continuously handle massive dynamic workloads and tasks. In the future, real-time resource allocation is required to fulfill dynamic task demands.

  • Decentralized Trust: The open nature of EC leads to the decentralized trust, e.g., services provided by different edge entities must be secured and trustworthy. Thus, efficient security mechanisms are required to ensure users’ authentication, data integrity, and mutual platform verification for EC [182]. Also, novel secure routing schemes and trust network topologies are critical for EC. In addition, end devices would generate a large volume of data at the network edge, which can be privacy-sensitive since they may contain users’ location data, health status, personal activities records or other sensitive information [99]. Therefore, feasible paradigms are needed to secure data sharing to minimize the privacy leakage.

  • Hardware Constraints in EC: Due to hardware constraints, unlike CC, EC cannot support heavy-weight software [99]. Big data analysis and data warehousing will never be feasible with existing EC because of the increasing number of duplication of small software in market. Therefore, users are facing difficulties in looking for a trusted edge provider among edge providers with a lack of standardized framework. Thus, developing software and hardware for handling computation offloading from the central cloud is a critical issue to be addressed.

  • Migration of EC Applications: Migration of EC applications among different edge clouds is a challenging issue that can help to balance loads or accommodate users’ movements and minimize the end-to-end latency of users [19]. In the future, efficient techniques are required based on system measurements and experiments to handle the migration challenges.

  • Interoperability and Collaboration of CC and EC: Thanks to EC, users are able to process latency-sensitive applications at the network edge [96]. However, handling an increasing number of multiple services is still a big challenge. Some tasks may be redirected to the central clouds for further processing. Thus, it is important to develop effective architectures and efficient algorithms to facilitate the above collaboration.

  • Heterogeneity: The future edge and cloud networks may present a huge level of heterogeneity, i.e., we may come across various kinds of users, ranging from smartphones, smartwatches, sensor nodes, and intelligent cameras [180]. This may cause the instability of the networks and, therefore, efficient solutions are highly required [95].

  • Edge Intelligence: There is a growing trend to process data at edge clouds because of privacy and other concerns [182]

    . However, due to the limited resources in edge clouds, some computational-intensive tasks such as machine learning models may not be trained at edge clouds without any proper modification. Therefore, in the future, it is important to study how to execute or train machine learning models in edge clouds. Federated learning has been proposed to address this issue but more attempts are expected in the future.

  • Blockchain-Assisted CC and EC: Blockchain has attracted much attention from both academia and industry [172, 29]. It is also interesting to design the blockchain-based networks for CC and EC. Three aspects can be considered here: 1) blockchain can be applied to enhance the privacy in the communication and computing resource sharing; 2) blockchain can be very useful in terms of improving the price and security of CC and EC; and 3) as blockchain may need huge computing resources to run, CC and EC can provide computing resources to blockchain-related tasks or applications.

V-B2 From CI Technique Perspective

  • Execution Time: Most CI techniques like EAs need a long time for searching the optimal solution in CC and EC. In the future, we may find a way to decrease the execution time.

  • Convergence: Sometimes, it is not easy to prove the convergence in CI techniques. However, in CC and EC, it is important to have an algorithm with convergence guarantee. Therefore, a future trend is to design some CI techniques with good convergence performance.

  • Multi/Many-Objective Optimization: Two or more objectives need to be addressed in CC and EC to meet the service level agreements of users. Therefore, a future trend may be to design multi/many-objective CI techniques to address multi/many-objective optimization problems in CC and EC.

  • Constraint-Handling Issue: Since most CI techniques cannot handle constraints effectively, we can incorporate some constraint-handling techniques into CI techniques to address constrained optimization problems in CC and EC.

  • CI Techniques for Security and Privacy: Directly sharing data among multiple edge nodes may run a high risk of privacy leakage. Therefore, federated learning paradigms can be applied to secure data sharing by training distributed data such that the original data sets can be kept in their source nodes and the edge AI model parameters can be shared among different nodes.

  • AI Edge Models: Instead of utilizing the existing resource-intensive AI models in CC and EC, we can design a resource-aware edge AI model. For example, methods like CI techniques can be adopted to efficiently search over the AI model design parameter space by taking into account the impact of hardware resource constraints on the performance metrics such as execution latency, security, and energy overhead.

  • CI Techniques for Migration of EC Applications: Sharing and migrating applications among edge clouds is a challenging task, as we have to consider the load balancing and reduce the latency for all tasks. LBS and FS may be suitable candidates for migration of EC applications.

  • CI Techniques for Interoperability and Collaboration of CC and EC: Multiple latency-sensitive requests may arrive at the same time. Therefore, fast CI techniques (such as LBS and FS) along with prediction algorithms are potential for the collaboration between the central and edge clouds.

  • Hybrid Versions: Hybrid CI techniques have shown great potential in solving complex optimization problems; however, few papers have studied them for addressing issues in CC and EC. A future trend may be to design hybrid CI techniques. For instance, FS is suitable for job scheduling while LBS is competitive for task offloading. In the future, we may combine FS and LBS to address joint issues. Similarly, more hybrid CI techniques can be developed to deal with energy consumption, completion time, and security in CC and EC.

Vi Conclusion

In this survey paper, we reviewed the applications of CI techniques to four critical issues in CC and EC: job scheduling, resource allocation, task offloading, and joint issues. We commenced with rudimentary concepts of CC and EC along with critical issues and metrics, and then focused on five categories of CI techniques used in CC and EC: EAs, SIAs, FS, LBS, and HS. Subsequently, diverse designed approaches relying on CI techniques were reviewed in the context of CC and EC. In addition, the statistics about the status-quo of CI for CC and EC were provided based on the works collected in this survey paper. We found that LBS was intensively used in CC and EC, followed by EAs and SIAs. However, FS and HS were not fully utilized in CC and EC. Finally, we pointed out some challenges and future research trends in CI, CC, and EC.

References

  • [1] L. 2, S. Zhao, S. Shen, and C. Jiang (2012) Task scheduling optimization in cloud computing based on heuristic algorithm. Journal of Network 7, pp. 547–553. Cited by: 1st item, §IV-B1, TABLE III.
  • [2] M. Agarwal and G. M. S. Srivastava (2016-04) A genetic algorithm inspired task scheduling in cloud computing. In 2016 International Conference on Computing, Communication and Automation (ICCCA), Vol. , pp. 364–367. External Links: Document, ISSN null Cited by: 1st item, TABLE II, §IV-A1.
  • [3] A. Ahmed and E. Ahmed (2016-01) A survey on mobile edge computing. In 2016 10th International Conference on Intelligent Systems and Control (ISCO), Vol. , pp. 1–8. External Links: ISSN Cited by: §I.
  • [4] E. Ahvar, A. Orgerie, and A. L bre (2019, in press) Estimating energy consumption of cloud, fog and edge computing infrastructures. IEEE Transactions on Sustainable Computing (), pp. . External Links: Document, ISSN 2377-3790 Cited by: §II-B1.
  • [5] A. Al-Shaikh, H. Khattab, A. Sharieh, and A. Sleit (2016) Resource utilization in cloud computing as an optimization problem. International Journal of Advanced Computer Science and Applications 7 (6), pp. 336–342. Cited by: §II-B3.
  • [6] A. Al-Shuwaili and O. Simeone (2017-06) Energy-efficient resource allocation for mobile edge computing-based augmented reality applications. IEEE Wireless Communications Letters 6 (3), pp. 398–401. External Links: ISSN Cited by: §II-C4.
  • [7] K. Anindita (2015-03) A new approach for task scheduling of cloud computing using fuzzy. International Journal of Innovative Research in Computer Science & Technology (IJIRCST) 3, pp. 112–116. External Links: ISSN 2347-5552 Cited by: 1st item, §IV-C1, TABLE IV, 3rd item.
  • [8] M. Armbrust, A. Fox, R. Griffith, A. D. Joseph, R. H. Katz, A. Konwinski, G. Lee, D. A. Patterson, A. Rabkin, I. Stoica, and M. Zaharia (2009-02) Above the clouds: a berkeley view of cloud computing. Technical report Technical Report UCB/EECS-2009-28, EECS Department, University of California, Berkeley. Cited by: §I.
  • [9] K. Arulkumaran, M. P. Deisenroth, M. Brundage, and A. A. Bharath (2017) Deep reinforcement learning: a brief survey. IEEE Signal Processing Magazine 34 (6), pp. 26–38. Cited by: 3rd item.
  • [10] W. Bao, H. Ji, X. Zhu, J. Wang, W. Xiao, and J. Wu (2016-01) ACO-based solution for computation offloading in mobile cloud computing. Big Data and Information Analytics 1, pp. 1–13. Cited by: §IV-B3, TABLE III.
  • [11] R. Basedia and M. Kumbhkar (2016-04) Cloud computing security issues and challenges. International Journal of Innovative Research in Computer and Communication Engineering 4 (4), pp. 6733–6736. Cited by: §II-B2.
  • [12] H. T. T. Binh, T. T. Anh, D. B. Son, P. A. Duc, and B. M. Nguyen (2018) An evolutionary algorithm for solving task scheduling problem in cloud-fog computing environment. See DBLP:conf/soict/2018, pp. 397–404. Cited by: 1st item, TABLE II, §IV-A1, 1st item.
  • [13] S. Bitam, S. Zeadally, and A. Mellouk (2017-04) Fog computing job scheduling optimization based on bees swarm. Enterprise Information Systems 12, pp. 1–25. External Links: Document Cited by: 3rd item, §IV-B1, TABLE III.
  • [14] C. Blum (2005) Ant colony optimization: introduction and recent trends. Physics of Life Reviews 2 (4), pp. 353 – 373. External Links: ISSN 1571-0645 Cited by: 2nd item.
  • [15] I. Bojanova and A. Samba (2011) Analysis of cloud computing delivery architecture models. In Proceedings of the 2011 IEEE Workshops of International Conference on Advanced Information Networking and Applications, WAINA’11, Washington, DC, USA, pp. 453–458. External Links: ISBN 978-0-7695-4338-3 Cited by: §I.
  • [16] F. Bonomi, R. Milito, J. Zhu, and S. Addepalli (2012) Fog computing and its role in the Internet of things. In Proceedings of the First Edition of the MCC Workshop on Mobile Cloud Computing, MCC ’12, New York, NY, USA, pp. 13–16. External Links: ISBN 978-1-4503-1519-7 Cited by: 2nd item.
  • [17] A. Bozorgchenani, F. Mashhadi, D. Tarchi, and S. Salinas Monroy (2020, in press) Multi-objective computation sharing in energy and delay constrained mobile edge computing environments. IEEE Transactions on Mobile Computing (), pp. . Cited by: TABLE II, §IV-A3.
  • [18] C. Canali and R. Lancellotti (2019) GASP: genetic algorithms for service placement in fog computing systems. Algorithms 12 (10). External Links: ISSN 1999-4893 Cited by: 1st item, TABLE II, §IV-A3, 3rd item.
  • [19] J. Chen and X. Ran (2019-08) Deep learning with edge computing: a review. Proceedings of the IEEE 107 (8), pp. 1655–1674. External Links: ISSN Cited by: §I, 6th item.
  • [20] J. Chen and D. Yang (2013) Data security strategy based on artificial immune algorithm for cloud computing. Applied Mathematics & Information Sciences 7 (1L), pp. 149–153. Cited by: 2nd item, TABLE II, §IV-A3.
  • [21] M. Chen, B. Liang, and M. Dong (2016) Joint offloading decision and resource allocation for multi-user multi-task mobile cloud. In 2016 IEEE International Conference on Communications (ICC), pp. 1–6. Cited by: §IV-D4.
  • [22] X. Chen, H. Zhang, C. Wu, S. Mao, Y. Ji, and M. Bennis (2018) Performance optimization in mobile-edge computing via deep reinforcement learning. In 2018 IEEE 88th Vehicular Technology Conference (VTC-Fall), Vol. , pp. 1–6. Cited by: §IV-D3, TABLE V.
  • [23] X. Chen, H. Zhang, C. Wu, S. Mao, Y. Ji, and M. Bennis (2019-06) Optimized computation offloading performance in virtual edge computing systems via deep reinforcement learning. IEEE Internet of Things Journal 6 (3), pp. 4005–4018. External Links: ISSN 2327-4662 Cited by: 3rd item, §IV-D3, TABLE V.
  • [24] X. Chen (2015-04) Decentralized computation offloading game for mobile cloud computing. IEEE Transactions on Parallel and Distributed Systems 26 (4), pp. 974–983. External Links: Document, ISSN 2161-9883 Cited by: §I.
  • [25] Y. Chen, Y. Zhang, and S. Maharjan (2017) Deep learning for secure mobile edge computing. arXiv preprint arXiv:1709.08025. Cited by: §I, 1st item, §IV-D3, TABLE V.
  • [26] Z. Chen and X. Wang (2018) Decentralized computation offloading for multi-user mobile edge computing: a deep reinforcement learning approach. arXiv preprint arXiv:1812.07394. Cited by: §IV-D3, TABLE V.
  • [27] Z. Cheng, P. Li, J. Wang, and S. Guo (2015-03) Just-in-time code offloading for wearable computing. IEEE Transactions on Emerging Topics in Computing 3 (1), pp. 74–83. External Links: ISSN 2168-6750 Cited by: §I, TABLE II, §IV-A3, 1st item.
  • [28] M. Chiang and T. Zhang (2016) Fog and IoT: an overview of research opportunities. IEEE Internet of Things Journal 3 (6), pp. 854–864. Cited by: 2nd item.
  • [29] K. R. Choo, Z. Yan, and W. Meng (2020) Blockchain in industrial IoT applications: security and privacy advances, challenges, and opportunities. IEEE Transactions on Industrial Informatics 16 (6), pp. 4119–4121. Cited by: 10th item.
  • [30] P. Chopra and R. Bedi (2017) Applications of fuzzy logic in cloud computing: a review. International Journal of Scientific Research Engineering & Technology 6, pp. 1083–1086. Cited by: §I.
  • [31] S. Deng, L. Huang, J. Taheri, and A. Y. Zomaya (2015-12) Computation offloading for service workflow in mobile cloud computing. IEEE Transactions on Parallel and Distributed Systems 26 (12), pp. 3317–3329. External Links: ISSN 1045-9219 Cited by: TABLE II, §IV-A3.
  • [32] T. Q. Dinh, Q. D. La, T. Q. S. Quek, and H. Shin (2018-12) Learning for computation offloading in mobile edge computing. IEEE Transactions on Communications 66 (12), pp. 6353–6367. External Links: ISSN Cited by: 2nd item, §IV-D3, TABLE V.
  • [33] E. S. Dolly (2017) Efficiently resource allocation in cloud scheduling using differential evolution. International Journal of Innovative Research in Computer and Communication Engineering 5, pp. 10232–10240. Cited by: 3rd item, TABLE II, §IV-A2.
  • [34] M. Dorigo, V. Maniezzo, and A. Colorni (1991) Positive feedback as a search strategy. technical report 91-016. Technical report Politecnico di Milano, Milano, Italy. Cited by: 2nd item.
  • [35] M. Dorigo, V. Maniezzo, and A. Colorni (1996-02) Ant system: optimization by a colony of cooperating agents. IEEE Transactions on Systems, Man, and Cybernetics, Part B (Cybernetics) 26 (1), pp. 29–41. External Links: Document, ISSN 1083-4419 Cited by: 2nd item.
  • [36] X. Dutreilh, S. Kirgizov, O. Melekhova, J. Malenfant, N. Rivierre, and I. Truck (2011) Using reinforcement learning for autonomic resource allocation in clouds: towards a fully automated workflow. In ICAS 2011: The Seventh International Conference on Autonomic and Autonomous Systems, pp. 67–74. Cited by: 2nd item, §IV-D2, TABLE V.
  • [37] Agoston. E. Eiben and J. E. Smith (2003)

    Introduction to evolutionary computing

    .
    Springer, Berlin, Heidelberg. External Links: ISBN 978-3-662-05094-1 Cited by: §III-A.
  • [38] P. Elina, M. Cristian, and G. Carlos (2014-04) Dynamic scheduling based on particle swarm optimization for cloud-based scientific experiments. Latin American Center for Computer Studies; CLEI Electronic Journal, CLEI Electronic Journal 14 (1), pp. 1–14. Cited by: 1st item, §IV-B1, TABLE III.
  • [39] H. Farhoud, A. Payam, P. Juha, H. m. l. i. Timo, and T. Hannu (2016-12) An intrusion detection system for fog computing and IoT based logistic systems using a smart data approach. International Journal of Digital Content Technology and its Applications(JDCTA) 10 (5), pp. 34–46. External Links: ISSN 1975-9339 Cited by: 2nd item, TABLE II, §IV-A3.
  • [40] F. Fatemi Moghaddam, M. Ahmadi, S. Sarvari, M. Eslami, and A. Golkar (2015-05) Cloud computing challenges and opportunities: a survey. In 2015 1st International Conference on Telematics and Future Generation Networks (TAFGEN), Vol. , pp. 34–38. External Links: Document, ISSN null Cited by: §I.
  • [41] N. Fernando, S. W. Loke, and W. Rahayu (2013) Mobile cloud computing: a survey. Future Generation Computer Systems 29 (1), pp. 84 – 106. Note: Including Special section: AIRCC-NetCoM 2009 and Special section: Clouds and Service-Oriented Architectures External Links: ISSN 0167-739X Cited by: §I, 1st item.
  • [42] V. Fran ois-Lavet, P. Henderson, R. Islam, M. G. Bellemare, and J. Pineau (2018) An introduction to deep reinforcement learning. Vol. , now, UK. Cited by: 3rd item.
  • [43] Y. Gao, W. Hu, K. Ha, B. Amos, and M. Satyanarayanan (2015) Are cloudlets necessary?. Technical report School of Computer Science, Carnegie Mellon University, Pittsburgh, PA, USA. Cited by: 1st item.
  • [44] M. Goudarzi, M. Zamani, and A. Toroghi Haghighat (2016) A genetic-based decision algorithm for multisite computation offloading in mobile cloud computing. International Journal of Communication Systems 30 (10), pp. e3241:1–e3241:17. Cited by: TABLE II, §IV-A3.
  • [45] F. Guo, H. Zhang, H. Ji, X. Li, and V. C. M. Leung (2018-05) Energy efficient computation offloading for multi-access MEC enabled small cell networks. In 2018 IEEE International Conference on Communications Workshops (ICC Workshops), Vol. , pp. 1–6. External Links: ISSN 2474-9133 Cited by: TABLE II, §IV-A4.
  • [46] F. Guo, H. Zhang, H. Ji, X. Li, and V. C. M. Leung (2018) An efficient computation offloading management scheme in the densely deployed small cell networks with mobile edge computing. IEEE/ACM Transactions on Networking 26 (6), pp. 2651–2664. Cited by: §IV-E4, TABLE VI, 1st item.
  • [47] M. Guzek, P. Bouvry, and E. Talbi (2015-05) A survey of evolutionary computation for resource management of processing in cloud computing. IEEE Computational Intelligence Magazine 10 (2), pp. 53–67. External Links: ISSN 1556-603X Cited by: §I, §II-B5.
  • [48] P. Haratian, F. Safi-Esfahani, L. Salimian, and A. Nabiollahi (2019-10) An adaptive and fuzzy resource management approach in cloud computing. IEEE Transactions on Cloud Computing 7 (4), pp. 907–920. External Links: Document, ISSN 2372-0018 Cited by: 1st item, §IV-C2, TABLE IV.
  • [49] K. Hassan, N. Javaid, F. Zafar, S. Rehman, M. Zahid, and S. Rasheed (2019) A cloud fog based framework for efficient resource allocation using firefly algorithm. In Advances on Broadband and Wireless Computing, Communication and Applications, L. Barolli, F. Leu, T. Enokido, and H. Chen (Eds.), Cham, pp. 431–443. External Links: ISBN 978-3-030-02613-4 Cited by: 5th item, §IV-B2, TABLE III.
  • [50] A. A. Helen and B. V. RebeccaJeya (2015-09) A survey on quality of service in cloud computing. International Journal of Computer Trends and Technology 27 (1), pp. 58–63. Cited by: §I.
  • [51] N. Hesabian, H. Haj, and S. Javadi (2015-06) Optimal scheduling in cloud computing environment using the Bee algorithm. International Journal of Computer Networks and Communications Security 3 (6), pp. 253–258. External Links: ISSN 2410-0595 Cited by: 3rd item, §IV-B1, TABLE III.
  • [52] G. E. Hinton, S. Osindero, and Y. Teh (2006-07) A fast learning algorithm for deep belief nets. Neural Computation 18 (7), pp. 1527–1554. External Links: ISSN 0899-7667 Cited by: 1st item.
  • [53] D. T. Hoang, C. Lee, D. Niyato, and P. Wang (2013) A survey of mobile cloud computing: architecture, applications, and approaches. Wireless Communications and Mobile Computing 13, pp. 1587–1611. Cited by: §I, 1st item.
  • [54] J. Holland (1974) Erratum: genetic algorithms and the optimal allocation of trials. SIAM Journal on Computing 3 (4), pp. 326–326. Cited by: 1st item.
  • [55] B. Hu, X. Sun, Y. Li, and H. Sun (2012-12) An improved adaptive genetic algorithm in cloud computing. In 2012 13th International Conference on Parallel and Distributed Computing, Applications and Technologies, Vol. , pp. 294–297. External Links: Document, ISSN 2379-5352 Cited by: 1st item, TABLE II, §IV-A1.
  • [56] W. Hu, Y. Gao, K. Ha, J. Wang, B. Amos, Z. Chen, P. Pillai, and M. Satyanarayanan (2016) Quantifying the impact of edge computing on mobile applications. In Proceedings of the 7th ACM SIGOPS Asia-Pacific Workshop on Systems, APSys ’16, New York, NY, USA, pp. 5:1–5:8. External Links: ISBN 978-1-4503-4265-0 Cited by: 3rd item.
  • [57] L. Huang, S. Bi, and Y. J. Zhang (2019, in press) Deep reinforcement learning for online computation offloading in wireless powered mobile-edge computing networks. IEEE Transactions on Mobile Computing (), pp. . Cited by: §IV-D4, TABLE V.
  • [58] L. Huang, X. Feng, A. Feng, Y. Huang, and L. P. Qian (2018-11-29) Distributed deep learning-based offloading for mobile edge computing networks. Mobile Networks and Applications, pp. 1–8. External Links: ISSN 1572-8153 Cited by: 1st item, §IV-D4, TABLE V.
  • [59] L. Huang, X. Feng, C. Zhang, L. Qian, and Y. Wu (2019) Deep reinforcement learning-based joint task offloading and bandwidth allocation for multi-user mobile edge computing. Digital Communications and Networks 5 (1), pp. 10 – 17. External Links: ISSN 2352-8648 Cited by: 3rd item, §IV-D4, TABLE V.
  • [60] P. Huang, Y. Wang, K. Wang, and Z. Liu (2019) A bilevel optimization approach for joint offloading decision and resource allocation in cooperative mobile edge computing. IEEE Transactions on Cybernetics (), pp. 1–14. External Links: ISSN 2168-2267 Cited by: 2nd item, §IV-B4, TABLE III.
  • [61] S. F. Issawi, A. Al Halees, and M. Radi (2015) An efficient adaptive load balancing algorithm for cloud computing under bursty workloads. Engineering, Technology & Applied Science Research 5 (3), pp. 795–800. Cited by: §IV-C1, TABLE IV.
  • [62] L. Jacob and P. Sengottuvelan (2014-04) Resource scheduling in cloud using bacterial foraging optimization algorithm. International Journal of Computer Applications 92 (1), pp. 14–20. Cited by: 6th item, §IV-B1, TABLE III.
  • [63] K. Jayashree (2018) Survey on QoS metrics and ranking in cloud services. International Journal of Engineering & Technology 7 (1.3), pp. 146–149. Cited by: §I.
  • [64] M. Jelassi, C. Ghazel, and L. A. Sa dane (2017-Sep.) A survey on quality of service in cloud computing. In 2017 3rd International Conference on Frontiers of Signal Processing (ICFSP), Vol. , pp. 63–67. External Links: Document, ISSN null Cited by: §II-B4, §II-B6, §II-B7.
  • [65] F. Jiang, K. Wang, L. Dong, C. Pan, W. Xu, and K. Yang (2019, in press) Deep learning based joint resource scheduling algorithms for hybrid MEC networks. IEEE Internet of Things Journal (), pp. . Cited by: §IV-E4, TABLE VI.
  • [66] N. J. Kansal and I. Chana (2016) Energy-aware, virtual machine migration for cloud computing - a firefly optimization approach. Journal of Grid Computing 14, pp. 327–345. Cited by: 5th item, §IV-B1, TABLE III.
  • [67] A. Kaur and S. Sharma (2017) Workflow scheduling in cloud computing environment using firefly algorithm. International Journal of Computer Science and Technology 8, pp. 73–76. Cited by: 5th item, §IV-B1, TABLE III.
  • [68] J. Kennedy and R. Eberhart (1995-11) Particle swarm optimization. In Proceedings of ICNN’95 - International Conference on Neural Networks, Vol. 4, pp. 1942–1948 vol.4. External Links: ISSN Cited by: 1st item.
  • [69] P. Kumar and A. Verma (2012) Scheduling using improved genetic algorithm in cloud computing for independent tasks. In Proceedings of the International Conference on Advances in Computing, Communications and Informatics, ICACCI ’12, New York, NY, USA, pp. 137–142. External Links: ISBN 978-1-4503-1196-0 Cited by: TABLE II, §IV-A1.
  • [70] Y. Lan, X. Wang, C. Wang, D. Wang, and Q. Li (2019) Collaborative computation offloading and resource allocation in cache-aided hierarchical edge-cloud systems. Electronics 8 (12). Cited by: §II-A2.
  • [71] G. Li, H. Zhou, B. Feng, G. Li, T. Li, Q. Xu, and W. Quan (2017) Fuzzy theory based security service chaining for sustainable mobile-edge computing. Mobile Information Systems 2017, pp. 1–13. Cited by: 2nd item, §IV-C3, TABLE IV.
  • [72] J. Li, H. Gao, T. Lv, and Y. Lu (2018) Deep reinforcement learning based computation offloading and resource allocation for MEC. In 2018 IEEE Wireless Communications and Networking Conference (WCNC), Vol. , pp. 1–6. Cited by: §IV-D4, TABLE V, 4th item.
  • [73] Z. Li and Q. Zhu (2020) Genetic algorithm-based optimization of offloading and resource allocation in mobile-edge computing. Information 11 (2), pp. 1–11. Cited by: TABLE II, §IV-A4.
  • [74] J. Lin, Z. Peng, and D. Cui (2018) Deep reinforcement learning for multi-resource cloud job scheduling. In Neural Information Processing, L. Cheng, A. C. S. Leung, and S. Ozawa (Eds.), Cham, pp. 289–302. External Links: ISBN 978-3-030-04182-3 Cited by: 3rd item, §IV-D1, TABLE V.
  • [75] Y. Lin and C. S. Chong (2017-06-01) Fast GA based project scheduling for computing resources allocation in a cloud manufacturing system. Journal of Intelligent Manufacturing 28 (5), pp. 1189–1201. External Links: ISSN 1572-8145 Cited by: TABLE II, §IV-A2.
  • [76] C. Liu, C. Zou, and P. Wu (2014-11) A task scheduling algorithm based on genetic algorithm and ant colony optimization in cloud computing. In 2014 13th International Symposium on Distributed Computing and Applications to Business, Engineering and Science, Vol. , pp. 68–72. External Links: ISSN Cited by: §III-E, §IV-E1, TABLE VI.
  • [77] F. Liu, G. Tang, Y. Li, Z. Cai, X. Zhang, and T. Zhou (2019-08) A survey on edge computing systems and tools. Proceedings of the IEEE 107 (8), pp. 1537–1562. External Links: Document, ISSN 1558-2256 Cited by: §I.
  • [78] J. Liu, X. Luo, X. Zhang, F. Zhang, and B. Li (2013) Scheduling model for cloud computing based on multi-objective genetic algorithm. IJCSI International Journal of Computer Science Issues 10 (3), pp. 134–139. Cited by: 1st item, TABLE II, §IV-A1.
  • [79] X. Liu, Z. Zhan, J. D. Deng, Y. Li, T. Gu, and J. Zhang (2018-02) An energy efficient ant colony system for virtual machine placement in cloud computing. IEEE Transactions on Evolutionary Computation 22 (1), pp. 113–128. External Links: ISSN 1089-778X Cited by: §IV-B1, TABLE III, 2nd item.
  • [80] X. Liu, Z. Qin, and Y. Gao (2019) Resource allocation for edge computing in IoT networks via reinforcement learning. arXiv preprint arXiv:1903.01856. Cited by: 2nd item, §IV-D2, TABLE V.
  • [81] Y. Liu, H. Yu, S. Xie, and Y. Zhang (2019-11) Deep reinforcement learning for offloading and resource allocation in vehicle edge computing and networks. IEEE Transactions on Vehicular Technology 68 (11), pp. 11158–11168. External Links: Document, ISSN 1939-9359 Cited by: 3rd item, §IV-D4, TABLE V.
  • [82] Z. Luo, M. LiWang, Z. Lin, L. Huang, X. Du, and M. Guizani (2017) Energy-efficient caching for mobile edge computing in 5G networks. Applied Sciences-basel Journal 7, pp. 1–13. Cited by: §I, TABLE II, §IV-A2.
  • [83] N. C. Luong, Z. Xiong, P. Wang, and D. Niyato (2018-05) Optimal auction for edge computing resource management in mobile blockchain networks: a deep learning approach. In 2018 IEEE International Conference on Communications (ICC), Vol. , pp. 1–6. External Links: ISSN 1938-1883 Cited by: §IV-D2, TABLE V, 4th item.
  • [84] M. Deng, H. Tian, and B. Fan (2016-05) Fine-granularity based application offloading policy in cloud-enhanced small cell networks. In 2016 IEEE International Conference on Communications Workshops (ICC), Vol. , pp. 638–643. External Links: ISSN Cited by: 1st item, §IV-B3, TABLE III.
  • [85] P. Mach and Z. Becvar (2017-thirdquarter) Mobile edge computing: a survey on architecture and computation offloading. IEEE Communications Surveys & Tutorials 19 (3), pp. 1628–1656. External Links: ISSN 1553-877X Cited by: §I.
  • [86] A. M. Manasrah and H. B. Ali (2018-07) Workflow scheduling using hybrid GA-PSO algorithm in cloud computing. Wireless Communications and Mobile Computing 2018, pp. 1–16. Cited by: §III-E, §IV-E1, TABLE VI.
  • [87] Q. Mao, F. Hu, and Q. Hao (2018-Fourthquarter) Deep learning for intelligent wireless networks: a comprehensive survey. IEEE Communications Surveys & Tutorials 20 (4), pp. 2595–2621. External Links: ISSN 1553-877X Cited by: 1st item.
  • [88] Y. Mao, C. You, J. Zhang, K. Huang, and K. B. Letaief (2017-Fourthquarter) A survey on mobile edge computing: the communication perspective. IEEE Communications Surveys & Tutorials 19 (4), pp. 2322–2358. External Links: ISSN 1553-877X Cited by: §I, §I.
  • [89] T. Medhat, E. Ashraf, K. Arabi, and T. Fawzy (2015-03) Cloud task scheduling based on ant colony optimization. The International Arab Journal of Information Technology 12 (2), pp. 129–137. Cited by: 2nd item, §IV-B1, TABLE III.
  • [90] P. Mell and T. Grance (2011-09) The NIST definition of cloud computing. Technical report Technical Report 800-145, National Institute of Standards and Technology (NIST)National Institute of Standards and Technology (NIST), Gaithersburg, MD. Cited by: §I.
  • [91] H. Meng, D. Chao, and Q. Guo (2019) Deep reinforcement learning based task offloading algorithm for mobile-edge computing systems. In Proceedings of the 2019 4th International Conference on Mathematics and Artificial Intelligence, ICMAI 2019, New York, NY, USA, pp. 90–94. External Links: ISBN 978-1-4503-6258-0 Cited by: 3rd item, §IV-D3, TABLE V.
  • [92] N. Moganarangan, R.G. Babukarthik, S. Bhuvaneswari, M.S. S. Basha, and P. Dhavachelvan (2016-01) A novel algorithm for reducing energy-consumption in cloud computing environment: web service computing approach. Journal of King Saud University - Computer and Information Sciences 28 (1), pp. 55–67. Cited by: §III-E, §IV-E1, TABLE VI.
  • [93] H. S. Mondal, M. T. Hasan, T. K. Karmokar, and S. Sarker (2017-Sep.) Improving quality of service in cloud computing architecture using fuzzy logic. In 2017 4th International Conference on Advances in Electrical Engineering (ICAEE), Vol. , pp. 149–152. External Links: Document, ISSN 2378-2692 Cited by: §IV-C1, TABLE IV.
  • [94] A. Mosavi, P. Ghamisi, Y. Faghan, P. Duan, and S. Shamshirband (2020) Comprehensive review of deep reinforcement learning methods and applications in economics. arXiv preprint arXiv:202003.0309.v1. Cited by: 3rd item.
  • [95] M. Mukherjee, L. Shu, and D. Wang (2018-thirdquarter) Survey of fog computing: fundamental, network applications, and research challenges. IEEE Communications Surveys & Tutorials 20 (3), pp. 1826–1857. Cited by: 8th item.
  • [96] R. K. Naha, S. Garg, D. Georgakopoulos, P. P. Jayaraman, L. Gao, Y. Xiang, and R. Ranjan (2018) Fog computing: survey of trends, architectures, requirements, and research directions. IEEE Access 6 (), pp. 47980–48009. Cited by: 7th item.
  • [97] A. T. Nassar and Y. Yilmaz (2018) Reinforcement-learning-based resource allocation in fog radio access networks for various IoT environments. arXiv preprint arXiv:1806.04582. Cited by: 2nd item, §IV-D2, TABLE V.
  • [98] S. Nazir, S. Shafiq, Z. Iqbal, M. Zeeshan, S. Tariq, and N. Javaid (2019) Cuckoo optimization algorithm based job scheduling using cloud and fog computing in smart grid. In Advances in Intelligent Networking and Collaborative Systems, F. Xhafa, L. Barolli, and M. Greguš (Eds.), Cham, pp. 34–46. External Links: ISBN 978-3-319-98557-2 Cited by: 4th item, §IV-B1, TABLE III.
  • [99] J. Ni, K. Zhang, X. Lin, and X. S. Shen (2018-Firstquarter) Securing fog computing for Internet of things applications: challenges and solutions. IEEE Communications Surveys & Tutorials 20 (1), pp. 601–628. External Links: ISSN 1553-877X Cited by: §II-B2, 4th item, 5th item.
  • [100] Z. Ning, P. Dong, X. Wang, L. Guo, J. J. P. C. Rodrigues, X. Kong, J. Huang, and R. Y. K. Kwok (2019-12) Deep reinforcement learning for intelligent internet of vehicles: an energy-efficient computational offloading scheme. IEEE Transactions on Cognitive Communications and Networking 5 (4), pp. 1060–1072. External Links: Document, ISSN 2372-2045 Cited by: 3rd item, §IV-D3, TABLE V.
  • [101] Z. Ning, P. Dong, X. Wang, J. J. Rodrigues, and F. Xia (2019-01) Deep reinforcement learning for vehicular edge computing: an intelligent offloading system. ACM Transactions on Intelligent Systems and Technology 10 (6), pp. 1–24. External Links: ISSN Cited by: 3rd item, §IV-D4, TABLE V, 4th item.
  • [102] K. Nitesh and K. Jitender (2014) A computation offloading framework to optimize energy utilization in mobile cloud computing environment. International Journal of Computer Applications & Information Technology 5, pp. 61–69. External Links: ISSN 2278-7720 Cited by: TABLE II, §IV-A3.
  • [103] S. Pang, W. Zhang, T. Ma, and Q. Gao (2017) Ant colony optimization algorithm to dynamic energy management in cloud data center. Mathematical Problems in Engineering 2017, pp. 1–10. Cited by: 2nd item, §IV-B1, TABLE III.
  • [104] K.E. Parsopoulos and M.N. Vrahatis (2002-06-01) Recent approaches to global optimization problems through particle swarm optimization. Natural Computing 1 (2), pp. 235–306. External Links: ISSN 1572-9796 Cited by: 1st item.
  • [105] K. M. Passino (2002-06) Biomimicry of bacterial foraging for distributed optimization and control. IEEE Control Systems Magazine 22 (3), pp. 52–67. External Links: ISSN Cited by: 6th item.
  • [106] W. Pedrycz (2015) Springer handbook of computational intelligence. Springer, Berlin, Heidelberg. External Links: ISBN 978-3-662-43505-2 Cited by: §III-B.
  • [107] K. Peng, V. Leung, X. Xu, L. Zheng, J. Wang, and Q. Huang (2018) A survey on mobile edge computing: focusing on service adoption and provision. Wireless Communications and Mobile Computing 2018, pp. 8267838:1–8267838:16. Cited by: §I.
  • [108] Z. Peng, D. Cui, Y. Ma, J. Xiong, B. Xu, and W. Lin (2016-06) A reinforcement learning-based mixed job scheduler scheme for cloud computing under SLA constraint. In 2016 IEEE 3rd International Conference on Cyber Security and Cloud Computing (CSCloud), Vol. , pp. 142–147. External Links: ISSN Cited by: 2nd item, §IV-D1, TABLE V.
  • [109] D.T. Pham, A. Ghanbarhh, E. Koc, S. Otri, S. Rahim, and M. Zaidi (2005) The bees algorithm, technical note. Technical report Manufacturing Engineering Center, Cardiff University, Cardiff, UK. Cited by: 3rd item.
  • [110] Q. Pham, F. Fang, V. N. Ha, M. Le, Z. Ding, L. B. Le, and W. Hwang (2019) A survey of multi-access edge computing in 5G and beyond: fundamentals, technology integration, and state-of-the-art. arXiv preprint arXiv:1906.08452. Cited by: §I.
  • [111] P. Porambage, J. Okwuibe, M. Liyanage, M. Ylianttila, and T. Taleb (2018-Fourthquarter) Survey on multi-access edge computing for Internet of things realization. IEEE Communications Surveys & Tutorials 20 (4), pp. 2961–2991. External Links: ISSN 1553-877X Cited by: §I.
  • [112] N. Primeau, R. Falcon, R. Abielmona, and E. M. Petriu (2018-Fourthquarter) A review of computational intelligence techniques in wireless sensor and actuator networks. IEEE Communications Surveys & Tutorials 20 (4), pp. 2822–2854. External Links: ISSN 1553-877X Cited by: §I, §III-E.
  • [113] C. Qu and R. Buyya (2014-05) A cloud trust evaluation system using hierarchical fuzzy inference system for service selection. In 2014 IEEE 28th International Conference on Advanced Information Networking and Applications, Vol. , pp. 850–857. External Links: Document, ISSN 2332-5658 Cited by: 2nd item, §IV-C3, TABLE IV.
  • [114] L. Quan, Z. Wang, and F. Ren (2018) A novel two-layered reinforcement learning for task offloading with tradeoff between physical machine utilization rate and delay. Future Internet 10 (7), pp. 1–17. External Links: ISSN 1999-5903 Cited by: §IV-D3, TABLE V.
  • [115] A. Ragmani, A. El Omri, N. Abghour, K. Moussaid, and M. Rida (2016) An improved scheduling strategy in cloud computing using fuzzy logic. In Proceedings of the International Conference on Big Data and Advanced Wireless Technologies, BDAW ’16, pp. 22:1–22:9. External Links: ISBN 978-1-4503-4779-2 Cited by: §IV-C1, TABLE IV.
  • [116] R. Raju, R. G. Babukarthik, D. Chandramohan, P. Dhavachelvan, and T. Vengattaraman (2013-02) Minimizing the makespan using hybrid algorithm for cloud computing. In 2013 3rd IEEE International Advance Computing Conference (IACC), Vol. , pp. 957–962. External Links: Document, ISSN null Cited by: §II-B5.
  • [117] S. Ranadheera, S. Maghsudi, and E. Hossain (2017) Mobile edge computation offloading using game theory and reinforcement learning. arXiv preprint arXiv:1711.09012. Cited by: §IV-D3, TABLE V.
  • [118] D. S. Rani and M. Pounambal (2019, in press) Deep learning based dynamic task offloading in mobile cloudlet environments. Evolutionary Intelligence. Cited by: 1st item, §IV-D3, TABLE V.
  • [119] S. Rashidi and S. Sharifian (2017) A hybrid heuristic queue based algorithm for task assignment in mobile cloud. Future Generation Computer Systems 68, pp. 331–345. External Links: ISSN 0167-739X Cited by: §III-E, §IV-E1, TABLE VI.
  • [120] J. Ren, G. Yu, Y. He, and G. Y. Li (2019) Collaborative cloud and edge computing for latency minimization. IEEE Transactions on Vehicular Technology 68 (5), pp. 5031–5044. Cited by: §II-A2.
  • [121] J. Ren, D. Zhang, S. He, Y. Zhang, and T. Li (2019-10) A survey on end-edge-cloud orchestrated network computing paradigms: transparent computing, mobile edge computing, fog computing, and cloudlet. ACM Comput. Surv. 52 (6), pp. 125:1–125:36. External Links: ISSN 0360-0300 Cited by: §I.
  • [122] Ritu and S. Jain (2016-05) A trust model in cloud computing based on fuzzy logic. In 2016 IEEE International Conference on Recent Trends in Electronics, Information Communication Technology (RTEICT), Vol. , pp. 47–52. External Links: Document, ISSN null Cited by: 1st item, §IV-C3, TABLE IV.
  • [123] R. Roman, R. Rios, J. A. Onieva, and J. Lopez (2019-06) Immune system for the Internet of things using edge technologies. IEEE Internet of Things Journal 6 (3), pp. 4774–4781. External Links: ISSN 2327-4662 Cited by: 2nd item, TABLE II, §IV-A3.
  • [124] A. S. Saljoughi, M. Mehvarz, and H. Mirvaziri (2017-Decembe) Attacks and intrusion detection in cloud computing using neural networks and particle swarm optimization algorithms. Emerging Science Journal 1 (4), pp. 179–191. Cited by: §IV-E3, TABLE VI.
  • [125] M. Satyanarayanan, P. Bahl, R. Caceres, and N. Davies (2009-10) The case for VM-based cloudlets in mobile computing. IEEE Pervasive Computing 8 (4), pp. 14–23. External Links: ISSN 1536-1268 Cited by: 1st item.
  • [126] M. Satyanarayanan, P. Simoens, Y. Xiao, P. Pillai, Z. Chen, K. Ha, W. Hu, and B. Amos (2015) Edge analytics in the Internet of things. IEEE Pervasive Computing 14 (2), pp. 24–31 (eng). External Links: ISSN 1536-1268 Cited by: 1st item.
  • [127] H. Shah-Mansouri and V. W. S. Wong (2018) Hierarchical fog-cloud computing for iot systems: a computation offloading game. IEEE Internet of Things Journal 5 (4), pp. 3246–3257. Cited by: §II-A2.
  • [128] K. Shaminder and V. Amandeep (2012) An efficient approach to genetic algorithm for task scheduling in cloud computing environment. I.J. Information Technology and Computer Science 10, pp. 74–79. Cited by: TABLE II, §IV-A1.
  • [129] W. Shi, J. Cao, Q. Zhang, Y. Li, and L. Xu (2016-10)

    Edge computing: vision and challenges

    .
    IEEE Internet of Things Journal 3 (5), pp. 637–646. External Links: ISSN 2327-4662 Cited by: §I.
  • [130] Y. Shoham, R. Powers, and T. Grenager (2003) Multi-agent reinforcement learning: a critical survey, technical report. Technical report Computer Science Department, Stanford University. Cited by: 2nd item.
  • [131] R. M. Singh, S. Paul, and A. Kumar (2014) Task scheduling in cloud computing: review. International Journal of Computer Science and Information Technologies 5 (6), pp. 7940–7944. Cited by: §I.
  • [132] C. Sonmez, A. Ozgovde, and C. Ersoy (2019) Fuzzy workload orchestration for edge computing. IEEE Transactions on Network and Service Management 16 (2), pp. 769–782. Cited by: §IV-C4, TABLE IV.
  • [133] T. Soyata, R. Muraleedharan, C. Funai, M. Kwon, and W. Heinzelman (2012-07) Cloud-vision: real-time face recognition using a mobile-cloudlet-cloud acceleration architecture. In 2012 IEEE Symposium on Computers and Communications (ISCC), Vol. , pp. 59–66. External Links: ISSN Cited by: §II-C4.
  • [134] S. Srinivas, S. Anupama, and K. Suvendu (2012-07) Efficient load balancing in cloud computing using fuzzy logic. IOSR Journal of Engineering (IOSRJEN) 2, pp. 65–71. External Links: ISSN 2250-3021 Cited by: 1st item, §IV-C1, TABLE IV, 3rd item.
  • [135] R. Storn and K. Price (1997) Differential evolution a simple evolution strategy for fast optimization. Dr Dobb s 22 (4), pp. 18–24. Cited by: 3rd item.
  • [136] S. Sundar and B. Liang (2017) Gaming and learning approaches for multi-user computation offloading. In Proceedings of the IEEE Vehicular Technology Conference, Toronto, Canada, pp. 24–27. Cited by: §IV-D3, TABLE V.
  • [137] A. Supacheep and M. Toshiyuki (2015) Cuckoo search algorithm for job scheduling in cloud systems. IEICE Transactions on Fundamentals of Electronics, Communications and Computer Sciences E98.A (2), pp. 645–649. Cited by: 4th item, §IV-B1, TABLE III.
  • [138] R. S. Sutton and A. G. Barto (1998) Introduction to reinforcement learning. 1st edition, MIT Press, Cambridge, MA, USA. External Links: ISBN 0262193981 Cited by: 2nd item.
  • [139] T. Takagi and M. Sugeno (1985) Fuzzy identification of systems and its applications to modeling and control. IEEE Transactions on Systems, Man, and Cybernetics SMC-15, pp. 116–132. Cited by: 2nd item.
  • [140] T. Taleb, K. Samdanis, B. Mada, H. Flinck, S. Dutta, and D. Sabella (2017-thirdquarter) On multi-access edge computing: a survey of the emerging 5G network edge cloud architecture and orchestration. IEEE Communications Surveys & Tutorials 19 (3), pp. 1657–1681. External Links: ISSN 1553-877X Cited by: §I.
  • [141] L. Tang, B. Tang, L. Kang, and L. Zhang (2019) A novel task caching and migration strategy in multi-access edge computing based on the genetic algorithm. Future Internet 11 (8), pp. 1–14. External Links: ISSN 1999-5903 Cited by: §I, TABLE II, §IV-A3.
  • [142] Y. Tao, X. Wang, X. Xu, and Y. Chen (2017-03) Dynamic resource allocation algorithm for container-based service computing. In 2017 IEEE 13th International Symposium on Autonomous Decentralized System (ISADS), Vol. , pp. 61–67. External Links: ISSN Cited by: 2nd item, §IV-C2, TABLE IV.
  • [143] A. Teske, R. Falcon, R. Abielmona, and E. Petriu (2017) Automating maritime risk assessment with genetic fuzzy systems. In Proc. of the 2nd International Symposium on Fuzzy and Rough Sets (ISFUROS), pp. 1–10. Cited by: §III-E.
  • [144] J. Timmis, A. Hone, T. Stibor, and E. Clark (2008) Theoretical advances in artificial immune systems. Theoretical Computer Science 403 (1), pp. 11–32. External Links: ISSN 0304-3975 Cited by: 2nd item.
  • [145] S.G. Tzafestas and K.C. Zikidis (2001) NeuroFAST: on-line neuro-fuzzy ART-based structure and parameter learning TSK model. IEEE Transactions on Systems, Man and Cybernetics, Part B (Cybernetics) 31 (5), pp. 797–802. Cited by: §III-E.
  • [146] J. Verma, S. Sobhanayak, S. Sharma, A. Kumar-Turuk, and B. Sahoo (2017) Bacteria foraging based task scheduling algorithm in cloud computing environment. International Conference on Computing, Communication and Automation (ICCCA), pp. 777–782. Cited by: 6th item, §IV-B1, TABLE III.
  • [147] S. Vimal, M. Khari, N. Dey, R. G. Crespo, and Y. H. Robinson (2020) Enhanced resource allocation in mobile edge computing using reinforcement learning based moaco algorithm for IIOT. Computer Communications 151 (), pp. 355–364. Cited by: §IV-E2, TABLE VI.
  • [148] H. Walaa, N. Heba, and R. Rawya (2017) Honey bee based load balancing in cloud computing. KSII Transactions on Internet and Information Systems 11 (12), pp. 5694–5711. Cited by: 3rd item, §IV-B1, TABLE III.
  • [149] J. Wang, J. Hu, G. Min, W. Zhan, Q. Ni, and N. Georgalas (2019) Computation offloading in multi-access edge computing using a deep sequential model based on reinforcement learning. IEEE Communications Magazine 57 (5), pp. 64–69. Cited by: §IV-D3, TABLE V.
  • [150] J. Wang, L. Zhao, J. Liu, and N. Kato (2019, in press) Smart resource allocation for mobile edge computing: a deep reinforcement learning approach. IEEE Transactions on Emerging Topics in Computing (), pp. . Cited by: §IV-D2, TABLE V, 4th item.
  • [151] J. Wang, J. Pan, F. Esposito, P. Calyam, Z. Yang, and P. Mohapatra (2018) Edge cloud offloading algorithms: issues, methods, and perspectives. arXiv preprint arXiv:1806.06191. Cited by: §I.
  • [152] L. Wang, P. Huang, K. Wang, G. Zhang, L. Zhang, N. Aslam, and K. Yang (2019-06) RL-based user association and resource allocation for multi-UAV enabled MEC. In 2019 15th International Wireless Communications Mobile Computing Conference (IWCMC), Vol. , pp. 741–746. External Links: Document, ISSN 2376-6492 Cited by: 2nd item.
  • [153] S. Wang, Y. Zhao, J. Xu, J. Yuan, and C. Hsu (2019) Edge server placement in mobile edge computing. Journal of Parallel and Distributed Computing 127, pp. 160–168. External Links: ISSN 0743-7315 Cited by: 2nd item.
  • [154] T. Wang, X. Wei, C. Tang, and J. Fan (2018-07-01) Efficient multi-tasks scheduling algorithm in mobile cloud computing with time constraints. Peer-to-Peer Networking and Applications 11 (4), pp. 793–807. External Links: ISSN 1936-6450 Cited by: 2nd item, §IV-B1, TABLE III, 2nd item.
  • [155] X. Wang, Y. Wang, Z. Hao, and J. Du (2015-06) The research on resource scheduling based on fuzzy clustering in cloud computing. In 2015 8th International Conference on Intelligent Computation Technology and Automation (ICICTA), Vol. , pp. 1025–1028. External Links: ISSN Cited by: 1st item, §IV-C2, TABLE IV.
  • [156] Y. Wang, C. Guo, and J. Yu (2018) Immune scheduling network based method for task scheduling in decentralized fog computing. Wireless Communications and Mobile Computing 2018, pp. 1–8. Cited by: TABLE II, §IV-A1.
  • [157] Y. Wang, Z. Ru, K. Wang, and P. Huang (2019) Joint deployment and task scheduling optimization for large-scale mobile users in multi-UAV-enabled mobile edge computing. IEEE Transactions on Cybernetics (), pp. 1–14. External Links: ISSN Cited by: 3rd item, TABLE II, §IV-A4.
  • [158] X. Wei, J. Fan, Z. Lu, and K. Din (2013) Application scheduling in mobile cloud computing with load balancing. Journal of Applied Mathematics 2013 (), pp. 1–13. External Links: ISSN Cited by: §IV-B1, TABLE III, 2nd item.
  • [159] N. Wingfield (2016-07) Amazon’s profits grow more than 800 percent, lifted by cloud services. Technical report New York Times, New York, NY, USA. Cited by: §I.
  • [160] I. H. Witten, E. Frank, and M. A. Hall (2011) Data mining: practical machine learning tools and techniques. 3 edition, Morgan Kaufmann Series in Data Management Systems, Morgan Kaufmann, Amsterdam. External Links: ISBN 978-0-12-374856-0 Cited by: §III-D.
  • [161] H. Wu (2018) Multi-objective decision-making for mobile cloud offloading: a survey. IEEE Access 6 (), pp. 3962–3976. External Links: ISSN 2169-3536 Cited by: §I.
  • [162] J. Wu, Z. Zhang, Y. Hong, and Y. Wen (2015-01) Cloud radio access network (C-RAN): a primer. IEEE Network 29 (1), pp. 35–41. External Links: ISSN Cited by: 2nd item.
  • [163] L. Xiao, X. Wan, C. Dai, X. Du, X. Chen, and M. Guizani (2018-06) Security in mobile edge caching with reinforcement learning. IEEE Wireless Communications 25 (3), pp. 116–122. External Links: ISSN 1536-1284 Cited by: §IV-D4, TABLE V.
  • [164] X. Xiong, K. Zheng, L. Lei, and L. Hou (2020, in press) Resource allocation based on deep reinforcement learning in iot edge computing. IEEE Journal on Selected Areas in Communications , pp. . Cited by: §IV-D2, TABLE V, 4th item.
  • [165] T. Yang, Y. Hu, M. C. Gursoy, A. Schmeink, and R. Mathar (2018-08) Deep reinforcement learning based resource allocation in low latency edge computing networks. In 2018 15th International Symposium on Wireless Communication Systems (ISWCS), Vol. , pp. 1–5. External Links: ISSN Cited by: 3rd item, §IV-D2, TABLE V.
  • [166] X. S. Yang (2010) Firefly algorithm, stochastic test functions and design optimisation. International Journal of Bio-Inspired Computation 2 (2), pp. 78–84. Cited by: 5th item.
  • [167] X. Yang and S. Deb (2009-12) Cuckoo search via lévy flights. In 2009 World Congress on Nature Biologically Inspired Computing (NaBIC), Vol. , pp. 210–214. External Links: ISSN Cited by: 4th item.
  • [168] X. Yang (2008) Nature-inspired metaheuristic algorithms. Luniver Press. External Links: ISBN 1905986106, 9781905986101 Cited by: 5th item.
  • [169] S. Yi, Z. Qin, and Q. Li (2015) Security and privacy issues of fog computing: a survey. In Wireless Algorithms, Systems, and Applications, K. Xu and H. Zhu (Eds.), Cham, pp. 685–695. External Links: ISBN 978-3-319-21837-3 Cited by: 3rd item.
  • [170] F. Youssef and B. Houda (2018) Deep reinforcement learning overview of the state of the art. Journal of Automation, Mobile Robotics & Intelligent Systems 12 (3), pp. 20–39. Cited by: 3rd item.
  • [171] W. Yu, F. Liang, X. He, W. G. Hatcher, C. Lu, J. Lin, and X. Yang (2018) A survey on the edge computing for the Internet of things. IEEE Access 6 (), pp. 6900–6919. External Links: ISSN 2169-3536 Cited by: §I.
  • [172] Y. Yuan and F. Wang (2018) Blockchain and cryptocurrencies: model, techniques, and applications. IEEE Transactions on Systems, Man, and Cybernetics: Systems 48 (9), pp. 1421–1428. Cited by: 10th item.
  • [173] L.A. Zadeh (1965) Fuzzy sets. Information and Control 8 (3), pp. 338–353. External Links: ISSN 0019-9958 Cited by: 1st item.
  • [174] L. A. Zadeh (1988-04) Fuzzy logic. Computer 21 (4), pp. 83–93. External Links: ISSN 0018-9162 Cited by: 1st item.
  • [175] M. Zarina, A. M. Aminu, S. W. N. Wan Nur, M. A. Mohamed, and M. D. Mustafa (2018) A genetic algorithm for optimal job scheduling and load balancing in cloud computing. International Journal of Engineering and Technology 7, pp. 290–294. Cited by: TABLE II, §IV-A1, 1st item.
  • [176] Z. Zhan, X. Liu, Y. Gong, J. Zhang, H. S. Chung, and Y. Li (2015-07) Cloud computing resource scheduling and a survey of its evolutionary approaches. ACM Computing Surveys 47 (4), pp. 63:1–63:33. External Links: ISSN 0360-0300 Cited by: §I.
  • [177] J. Zhang, H. Guo, and J. Liu (2019) A reinforcement learning based task offloading scheme for vehicular edge computing network. In Artificial Intelligence for Communications and Networks, S. Han, L. Ye, and W. Meng (Eds.), Cham, pp. 438–449. External Links: ISBN 978-3-030-22971-9 Cited by: 3rd item, §IV-D3, TABLE V.
  • [178] Q. Zhang, M. Lin, L. T. Yang, Z. Chen, and P. Li (2019) Energy-efficient scheduling for real-time systems based on deep Q-learning model. IEEE Transactions on Sustainable Computing 4 (1), pp. 132–141. Cited by: 3rd item, §IV-D1, TABLE V.
  • [179] Q. Zhang, L. Cheng, and R. Boutaba (2010-05-01) Cloud computing: state-of-the-art and research challenges. Journal of Internet Services and Applications 1 (1), pp. 7–18. External Links: ISSN 1869-0238 Cited by: §I.
  • [180] Y. Zhang, B. Di, P. Wang, J. Lin, and L. Song (2020) HetMEC: heterogeneous multi-layer mobile edge computing in the 6 G era. IEEE Transactions on Vehicular Technology 69 (4), pp. 4388–4400. Cited by: 8th item.
  • [181] Z. Zhang and X. Zhang (2010-05) A load balancing mechanism based on ant colony and complex network theory in open cloud computing federation. In 2010 The 2nd International Conference on Industrial Mechatronics and Automation, Vol. 2, pp. 240–243. External Links: ISSN Cited by: §IV-B1, TABLE III, 2nd item.
  • [182] Z. Zhou, X. Chen, E. Li, L. Zeng, K. Luo, and J. Zhang (2019-08) Edge intelligence: paving the last mile of artificial intelligence with edge computing. Proceedings of the IEEE 107 (8), pp. 1738–1762. External Links: ISSN Cited by: §I, 4th item, 9th item.
  • [183] M. S. Q. Zulkar Nine, M. A. K. Azad, S. Abdullah, and R. M. Rahman (2013) Fuzzy logic based dynamic load balancing in virtualized data centers. In 2013 IEEE International Conference on Fuzzy Systems (FUZZ-IEEE), Vol. , pp. 1–7. Cited by: §IV-C4, TABLE IV.