Soft Computing Based Metaheuristic Algorithms for Resource Management in Edge Computing Environment

Nawaf Alhebaishi,Abdulrhman M.Alshareef,Tawfiq Hasanin,Raed Alsini,Gyanendra Prasad Joshi,Seongsoo Cho and Doo Ill Chul

1Department of Information Systems,Faculty of Computing and Information Technology,King Abdulaziz University,Jeddah,Saudi Arabia

2Department of Computer Science and Engineering,Sejong University,Seoul,Korea

3School of Software,Soongsil University,Seoul,06978,Korea

4Artificial Intelligence Education,Hankuk University of Foreign Studies,Dongdaemun-gu,Seoul,02450,Korea

Abstract: In recent times,internet of things(IoT)applications on the cloud might not be the effective solution for every IoT scenario, particularly for time sensitive applications.A significant alternative to use is edge computing that resolves the problem of requiring high bandwidth by end devices.Edge computing is considered a method of forwarding the processing and communication resources in the cloud towards the edge.One of the considerations of the edge computing environment is resource management that involves resource scheduling,load balancing,task scheduling,and quality of service (QoS)to accomplish improved performance.With this motivation,this paper presents new soft computing based metaheuristic algorithms for resource scheduling(RS)in the edge computing environment.The SCBMARS model involves the hybridization of the Group Teaching Optimization Algorithm (GTOA)with rat swarm optimizer (RSO)algorithm for optimal resource allocation.The goal of the SCBMA-RS model is to identify and allocate resources to every incoming user request in such a way, that the client’s necessities are satisfied with the minimum number of possible resources and optimal energy consumption.The problem is formulated based on the availability of VMs,task characteristics,and queue dynamics.The integration of GTOA and RSO algorithms assist to improve the allocation of resources among VMs in the data center.For experimental validation,a comprehensive set of simulations were performed using the CloudSim tool.The experimental results showcased the superior performance of the SCBMA-RS model interms of different measures.

Keywords: Resource scheduling; edge computing; soft computing; fitness function;virtual machines

1 Introduction

The Internet of Things (IoT)interlinks devices and Internet to perform useful communication among people and objects.The linking procedure usually comprises control, sensing, and actuating devices.In addition,this device conforms to an essential standard compliant transmission protocol.IoT could understand the goal of smart discovering,controlling,identifying,and following things in several diverse and effective manners [1].Therefore, IoT becomes common in fields like transport,smart healthcare, industrial automation, retail, logistics, etc.Most IoT enabled applications are computationally intensive, like augmented reality (AR)and interactive gaming, and it is complex for the device itself to satisfy this task because of the power consideration and hardware limitations.Though the cloud server provides adequate computation assets,a huge number of traffics sent to the cloud will cause unpredictable delays and network congestion that fails to encounter the lower latency need and reduces the quality of experience(QoE).The developing Edge Computing(EC)technique faces the limitations of cloud computing(CC)[2].Mobile edge computing(MEC)allows several IoT services and applications executed at the network edge rather than being sent to the remote cloud that decreases the response time and lessens the burden on backhaul connection.But,the storage,network resources, and computation of the edge server are constrained, and therefore, Resource Scheduling(RS)is significant to increase the QoE.In recent times,EC was broadly utilized for computationally intensive tasks of artificial intelligence (AI)in the real IoT platform [3].As the edge nodes have the features of dynamic changes and resource constrained, it is important to implement a proper architecture to offload and schedule computation tasks in the edge.Besides,soft computing techniques can be widely employed to design RS techniques in EC environment.Fig.1 shows the architecture of MEC[4].

Figure 1:Architecture of mobile edge computing

Reference[5]implement a genetic method and particle swarm optimization(PSO)based technique for solving assets distribution issue.But, the dependence among subtask wasn’t deliberated in this work.Non dominated sorting genetic technique II has been adapted to understand multi objective optimization for reducing the energy consumption and execution time of EC devices.[6]was verified that resource allocation strategy is defined by computing task and the highest completion time of its instant predecessor.In [7], a smart collaborative automation (SCA)system is presented to enhance asset utilization.Reference [8] projected a technique that attained distribution decision through randomization and semidefinite relaxation, however, the transmission between sub tasks is neglected using this study.The researchers in[9]acquired dependency between sub tasks to account and presented a multistage greedy adjustment(MSGA)method for solving task distribution issues.

Reference [10] focuses on task offloading challenge for finding an optimum trade off among energy consumption and task completion latency.They projected a modified fast and elitist non dominated sorting genetic technique for solving offload issues.But,in this study,the task is deliberated to be undivided,that wastes the equivalent computing ability of EC server.Researchers in[11]made unexecuted task queues at MEC server and presented an online technique for allocating resources.[12]proposed a management and allocation resources technique for reducing the module difficulty of resource allocation technique.The researchers in[13]take the dependencies among task to account,and projected a deep reinforcement learning(DRL)method to create offload decision,the variance between tasks are neglected by this study.[14] adapt Non-dominated Sorting Genetic method II to reduce the allocating resource time of computing task and decrease the energy utilization of EC servers,however,this study didn’t deliberate the dependency between tasks.

This paper designs a new soft computing based metaheuristic algorithms for resource scheduling (RS), called SCBMA-RS in the EC environment.The SCBMA-RS model involves the hybridization of the Group Teaching Optimization Algorithm (GTOA)with rat swarm optimizer(RSO)algorithm to optimally allocate the resources in EC.Besides, the SCBMA-RS model derives a fitness function to identify and allocate resources to every incoming user request in such a way, that the client’s necessities are satisfied with the minimum number of possible resources and optimal energy consumption.In the SCBMA-RS method, the incorporation of the GTOA and RSO algorithms helps to improvise the allocation of resources among VMs in the data center.A series of simulations were performed by the use of the CloudSim tool and the results are inspected interms of different performance measures.

2 Literature Review

Li et al.[15]proposed a useroriented improved ISCM in this study.Depending upon enhanced k means technique,the ISCM method resolves the issue that clustering outcome is sensitive to primary value and comprehends the reclustering, that creates the attained clustering outcomes more stable.Lastly, the EC RS system is attained depending upon clustering outcomes.Li et al.[16] proposed a hybrid computing architecture and implemented a smart RS approach to satisfy the real world need in smart manufacturing with the support of EC.Initially,a 4 layers computing model from the smart manufacturing platform is given for supporting the AI task function with the network perception.Later, a 2 phase’s technique to schedule the computing resource in the edge layer is implemented depending upon threshold and greedy approaches using latency limitations.

Wang et al.[17]proposed an optimization approach to compute resource allocation of huge IoHT devices to consider privacy protection from cloud EC platforms.Initially,a 5G heterogeneous cloud EC network is created.As well, based on network conditions, the computing needs of devices are distributed to EC/local implementation.The communication, computing resource allocation, and delay of edge servers are modelled consequently.Lastly,a protection module for immediate messaging privacy data is implemented with the consideration of the threat of large scale privacy data leakage in IoHT.In Vimal et al.[18], the factor causing this delay are forecast with MEC resources and assess the efficiency from the neighboring client tool.The efficiency creates a cognitive agent module for assessing the communication network and resource allocation is determined for enhancing the QoS.The Reinforcement Learning technique MOACO method was employed for handling the precise resource allocation among the end user from the manner of making the cost map table creation and optimum allocation from MEC.

Zhang et al.[19] proposed 2 slow movement PSO methods for solving the resulting NP hard problem.Especially, they improve a position based mapping system to map particles to schedule solutions.For preventing the substantial modification in particle position,they additionally proposed a new particle upgrading approach to slow down particle movement, to examine higher quality solution in the guide of global optimal particle and personal optimal particle.Porkodi et al.[20]proposed a new fuzzy clustering with flower pollination method named FCM FPA as resource provision module for computing fog.Initially,the resource attributes are normalized and standardized.Finally, the proposed resource provision method depending upon optimized fuzzy clustering was developed.

Arani et al.[21] presented a task scheduling method depending upon moth flame optimization method for assigning optimum sets of tasks to fog node for meeting the fulfilment of QoS needs of CPS application in this manner that the overall execution time of task is minimalized.Sheng et al.[22]leverage DRL to resolve time scheduling(viz.,task execution order)and resource allocation(viz.that VM the task is allocated to),consider the diversity of tasks and heterogeneity of accessible resource.The policy based REINFORCE method is presented to scheduling task problems,and an FCN is used for extracting features.

Li et al.[23]presented a RS technique for computing fog in this study.Initially,they normalize and standardize the resource attributes.Next,they integrate fuzzy clustering approaches with PSO for dividing the resource,and the scale of resource searching is decreased.Lastly,they proposed a novel RS method depending upon optimized fuzzy clustering.In Vimal et al.[24],the RL methods MOACO technique was employed to handle a precise resource allocation among the end user from the manner of making the cost mapping table creation and optimum allocation in MEC.

3 The Proposed Resource Management Technique

Fig.2 shows the overall working process of SCBMA-RS model.The physical machinemin the cloud datacenter is denoted as the set of P={p1,p2,...pm},with q VM indicated as V={v1,v2,...vn}and k tasks Tk={t1,t2,...tk}.Users submit the task to the cloud broker.It can be represented as a set of variables such as ti={ari,lni,dli,fti},whereas aridenotes arrival times,siindicates the length/size of tasks,dlirepresents time limits/deadline to the tasks execution,and ftidenotes the finish time of task.The submitted task tiis mapped for VM vjusing broker.The broker maps the submitted task to VM.In this method,they mostly emphasize the usage of the VM,the energy utilization,cost of datacenter,and completion time of tasks(makespan).Consider Pkas the processing time of entire tasks.

Figure 2:Overall process of SCBMA-RS model

3.1 Process Involved in GTOA

The group teaching optimization technique is developed on the basis of a group teaching module.In this method, the students, the knowledge of students, and subjects provided to the student,correspondingly,are denoted as fitness value,population,and decision variables.The 2 groups consist of student population,viz.,average and outstanding groups.A single teacher is chosen for teaching 2 groups in the teaching stage.At the same time,the teacher would assume the difference of the learning capability of 2 groups hence creating distinct teaching strategies and group learning strategies for the student includes student interaction and self-learning.The 6 phases create the architecture of the GTO technique contains teacher allocation phase,initialization,student phase,teacher phase,termination,and reconstruct population.Beforehand the key optimization loop imitates, the value of maximal iterations amountMaxItmust be fixed, and the present iteration amountTcurshould be initiated= 0.Also, the population X is produced based on amount of populationN, the dimension of the parametersD,and the upper bound ub&lower bound lb is given by:

Whereasranddenotes arbitrary amount among [0, 1],lbjandubjrepresents lower and upper bounds of thejth parameter,correspondingly.

The teacher Allocation Phase can be defined as follows.

Whereas Ttindicates chosen teacher in the present iterationt.,and,correspondingly,indicates present 1st,2nd,3rd optimum students.f(·)represents fitness function[25].

At the teacher Phase,the outstanding group can be determined using following equations.

Whereasdenot es knowledge of ith student from the remaining group learned in the teacher in present iterationst,Xitrepresents knowledge of ith student from the remaining groups.a,b,andcindicate arbitrary amounts in the range of zero and one.Fdenotes coefficient whether equivalents to one/two.Mtdenotes mean vector of the remaining groups in present iterationst.Fig.3 demonstrates the flowchart of GTOA.

Figure 3:Flowchart of GTOA

Next,at the teacher Phase,the average group can be determined using following equations.

Whereasdenotes knowledge of ith student in average group learned in the teacher in present iterationst.

Finally,the student phase can be defined as follows.

Whereaseandgindicates arbitrary amounts with the range of zero and one,represents knowledge of ith student learning in the student phase in iterationst+1.

where Xt+1indicates upgraded population,anddenotes upgraded outstanding and average group afterward an iteration,correspondingly.The optimization procedure would be ended when the iteration amountTcurgoes beyond the value ofN×MaxIt.

3.2 Overview of RSO

Rats are medium sized and long tailed rodents that are distinct based on their size and weight.It has 2 major species:Brown and Black rats.Generally,the female ones are called as does and the male ones are known as bucks.They are commonly socially intelligent in nature.Rats groom one another and include in several activities like chasing, jumping, boxing, and tumbling.They are territorial animal that lives in a set of females and males.Rat behaviors are highly aggressive in several instances that might lead to mortality of few animals.These behaviors are the major inspiration of this study when fighting and chasing prey.In this study,the fighting and chasing behaviors of rats are numerically modeled to implement RSO technique and execute optimization.

This section defines the rat behaviors viz.,fighting&chasing.Later the presented RSO technique is summarized.

3.2.1 Chasing the Prey

In general,rat is a social animal that chases the prey in groups by their social agonistic behavior.For defining this behavior numerically, they consider an optimum search agent (SA)implies the information regarding the place of the preys.The rest of the SAs could upgrades the places regarding an optimum SA reached until now[26].The succeeding formulations are provided below.

Here,x=0,1, 2, ···,MaxIteration

Thus,RandCdenotes arbitrary amount among[0,2]&[1,5],correspondingly.The variablesAandCare in charge for optimum exploitation and exploration on the iteration course.

3.2.2 Fighting with Prey

The fighting procedure of rats using prey can be determined using the succeeding formula:

The flowchart and steps(Fig.4)of RSO algorithm are listed as follows:

i)Initiate the rat populationPiwhereasi=1, 2,...,n.

ii)Elect primary variables of RSO:A,C,andR.

iii)Here,estimate the fitness values of every SA.

iv)An optimum SA is identified from the given searching area.

v)Updating SA place SA by Eq.(16).

vi)Check either any SA surpasses the edge limits of search space and later alters it.

vii)Next, estimate the updated SA fitness values and updatee the vectorPrwhen there is an optimum solution compared to prior optimum ones.

viii)Stop,if ending criteria is fulfilled.Or else,returns to Step 5.

ix)Display optimum solutions achieved so far.

3.3 Hybridization of GTOA and RSO for Resource Scheduling

In this study,an SCBMA-RS technique is utilized for scheduling resources in EC environment.The lower optimization accuracy and easier of falling to local optimal are the disadvantages in fruit fly optimization.The primary reason to develop the SCBMA-RS technique is to conquer the deficiencies of original GTOA optimization technique.The process of the projected method contains 2 phases.GTOA is utilized initial phase.The next phase incorporates RSO algorithm for updating the present location and solution to strengthen the problem of GTOA over early convergences, because of its exploitation and exploration capability.The suggested method enhances the optimization accuracy and convergence rate consequently.

Letpeindicates processing component,penum_pjrepresent processor amount,pemips-pjdenotes MIPS of entire processors,vbwv,-jrepresents bandwidth of Vj, the VM capacity can be determined as follows.

Figure 4:Flowchart of RSO

LetLVMdenotes load of a specific VM,Num(T,t)represents overall amount of tasks at time t,Service_rate(Vj,t)indicated service rates of VM Vjat t.

The load of entire VMs can be represented using the following equations:

HerelTrepresents length or size of task T and the fraction of CPU efficiency can be defined usingc(Vj).

Consider sttijas the starting time of tasks tirepresents VM vj,and ftiindicates finish time of tasks tion VM vj.ftishould be estimated as follows[27]:

χijdenotes decision parameter utilized as every task must be allocated to single VM.Pijdenotes processing time of tasks Tiassigned to VM Vj.

Thus,the objective functions of load balancing(LB)module given by:

Whereasf tijdenotes finishing time of tasktion VMvj.

Makespan in the EC determines the whole finishing time of tasks T in VM Vj.Therefore, this objective functions are utilized for reducing makespan of tasks.Consider econsijdenotes energy consumption created using task tioperating on VM vj,econs_ratejdenotes energy consumption rate of VM,andexect(T)represents task execution time.

The energy consumption is estimated as follows

The overall energy consumptions can be defined using Eq.(27):

Thus,the objective function is determined by

The cost of datacenter is estimated by Eq.(27)

where c denotes cost of one kW power.Also,the objective functions of the cost is determined as:

The objective function for LB module is determined by

Eq.(31)denotes that the single task must be assigned to vj,limitation(32)represents execution of every task must be lesser than deadline,limitation(33)reveals that the SD of load must be lesser than upper threshold value Thupper.

In the response time is time required in the tasks enter the scheme and time of tasks are scheduled.It can be estimated by

Finally,the degree of imbalance can be derived as follows.

Whereas Max(T)denotes maximal amount of task,Min(T)indicates minimal amount of task and Avg(T)represents average of task(T)

In this presented method,the removed tasks are deliberated as fly that seeks for appropriate VM depending upon mufti objective functions.The fundamental limitations are follows including the load of the VM,afterward allocating the task shouldn’t be higher compared to upper threshold value for selecting an appropriate VM to the removed task.When there is huge amount of VM is presented,then the deadline limitation is deliberated.The task deadline is essential to transfer the task from heavier load VM to lower load VM.When the deadline dliof removal task is higher,afterward the VM have minimal of high deadline tasks are chosen.When the task deadline is medium,then the VM has lesser amount of medium and high deadline tasks are chosen.The VM grouping is depending upon the present load LVM(t)of VM.They assume 2 kinds of groups like underloaded VM group(findVMListul)and overloaded VM group(findVMListol).The task is removal in findVMListoland assigned for VM in findVMListuldepending upon objective functions.The procedure of removal task in findVMListolis continual,until findVMListolis NULL.This task emphasis on LB,it also concentrates on storing the energy utilized from the datacenter for reducing cost.The radical procedure of energy preservation is depending upon creating the VM to ON&OFF state that isn’t in usage.When the load of specific VM is less compared to low threshold value Thlower, next VM donates to sleep mode, and when the load of VM is higher compared to Thupper,then awake the VM in sleep mode.When the VM load is NULL,afterward the VM is removal in VMList for saving energy.

4 Performance Validation

This section validates the experimental analysis of the presented SCBMA-RS technique under diverse aspects.The results are examined interms of Average Task Satisfaction Degree (ATSD)and Task Successful Ratio(TSR)under varying task arriving rate(TAR)and population skewness(PS).The values of the ATSD and TSR should be higher for effective allocation of resources in EC environment.

Tab.1 investigates the comparative outcomes analysis of the SCMBA-RS model interms of ATSD and TSR under different TAR.A brief ATSD analysis of the SCBMA-RS technique with other techniques under distinct TAR is provided in Fig.5.The figure demonstrated that the SCBMA-RS technique has obtained better performance over the other methods with the maximum ATSD.For instance,with the TAR of 4,the SCBMA-RS model has gained an increased ATSD of 3.17 whereas the DRL, Greedy-FCFS, and Greedy-SJF models have obtained a decreased ATSD of 3.08, 1.30,and 1.70 respectively.Besides,with the TAR of 7,the SCBMA-RS model has accomplished improved performance with the ATSD of 2.23 whereas the DRL,Greedy-FCFS,and Greedy-SJF models have resulted in reduced performance with the ATSD of 2.07,1.04,and 1.18 respectively.

Table 1: Result analysis of SCBMA-RS in terms of ATSD and TSR vs.TAR

Figure 5:ATSD analysis of SCBMA-RS model in task arriving rate

A detailed TSR analysis of the SCBMA-RS method with other approaches under different TAR is given in Fig.6.The figure showcased that the SCBMA-RS method has reached optimum performance over the other techniques with the maximal TSR.For instance, with the TAR of 4, the SCBMARS manner has gained an improved TSR of 0.56 whereas the DRL,Greedy-FCFS,and Greedy-SJF techniques have attained a lesser TSR of 0.53,0.31,and 0.43 correspondingly.

Figure 6:TSR analysis of SCBMA-RS model in task arriving rate

Followed by,with the TAR of 7,the SCBMA-RS technique has accomplished higher performance with the TSR of 0.30 whereas the DRL,Greedy-FCFS,and Greedy-SJF algorithms have resulted in minimum performance with the TSR of 0.28,0.22,and 0.24 correspondingly.

Tab.2 examines the comparative outcomes analysis of the SCMBA-RS technique with respect to ATSD and TSR under distinct PS.

Table 2: Result analysis of existing with proposed SCBMA-RS in terms of ATSR and TSR vs.popularity skewness

A examining ATSD analysis of the SCBMA-RS approach with other algorithms in different PS is provided in Fig.7.The figure exhibited that the SCBMA-RS method has achieved optimal performance over the other approaches with the maximal ATSD.For sample, with the PS of 0.1,the SCBMA-RS method has attained a higher ATSD of 2.38 whereas the DRL,Greedy-FCFS,and Greedy-SJF models have achieved a decreased ATSD of 2.05, 1.29, and 1.05 correspondingly.Also,with the PS of 0.9,the SCBMA-RS method has accomplished maximum performance with the ATSD of 4.85 whereas the DRL, Greedy-FCFS, and Greedy-SJF methodologies have resulted in lesser performance with the ATSD of 4.54,2.12,and 1.74 respectively.

Figure 7:ATSD analysis of SCBMA-RS model in popularity skewness

A brief TSR analysis of the SCBMA-RS method with other algorithms under different PS is given in Fig.8.The figure outperformed that the SCBMA-RS approach has attained good performance over the other techniques with superior TSR.For instance, with the PS of 0.1, the SCBMA-RS manner has reached a maximal TSR of 0.42 whereas the DRL, Greedy-FCFS, and Greedy-SJF techniques have reached a lesser TSR of 0.34,0.19,and 0.10 correspondingly.Moreover,with the PS of 0.9,the SCBMA-RS technique has accomplished higher performance with the TSR of 0.97 whereas the DRL,Greedy-FCFS, and Greedy-SJF algorithms have resulted in minimal performance with the TSR of 0.94,0.66,and 0.51 correspondingly.

In order to further validate the improved performance of the SCBMA-RS model,another ATSD analysis is made under different number of VMs in Tab.3 and Fig.9.The resultant experimental results depicted that the SCBMA-RS technique has showcased better performance with the maximum ATSD under all VMs.For instance, under 2 VMs, a higher ATSD of 1.724 has been obtained by the SCBMA-RS technique whereas the DRL, Greedy-FCFS, and Greedy-SJF techniques have demonstrated a lower ATSD of 1.518, 0.907, and 0.975 respectively.Eventually, under 4 VMs, the SCBMA-RS technique has resulted in an improved ATSD of 5.091 whereas the DRL,Greedy-FCFS,and Greedy-SJF techniques have attained a reduced ATSD of 4.728,1.996,and 2.231 respectively.

Figure 8:TSR analysis of SCBMA-RS model in popularity skewness

Table 3: Results comparison of existing with proposed SCBMA-RS in terms of ATSD

Figure 9:Result analysis of SCBMA-RS model interms of ATSD

To further validate the enhanced performance of the SCBMA-RS method,another TSR analysis is developed in distinct number of VMs in Tab.4 and Fig.10.The resultant experimental outcomes exhibited that the SCBMA-RS approach has outperformed optimum performance with the maximal TSR under all VMs.For sample, under 2 VMs, a superior TSR of 0.348 has been attained by the SCBMA-RS model whereas the DRL, Greedy-FCFS, and Greedy-SJF approaches have showcased a minimum TSR of 0.336,0.172,and 0.174 correspondingly.Finally,under 4 VMs,the SCBMA-RS method has resulted in a higher TSR of 0.587 whereas the DRL, Greedy-FCFS, and Greedy-SJF methodologies have reached a lesser TSR of 0.569,0.460,and 0.561 correspondingly.

Table 4: Results comparison of existing with proposed SCBMA-RS in terms of TSR

From the above mentioned tables and figures,it can be depicted that the proposed model is found to be an efficiency tool for scheduling resources from the EC environment.

5 Conclusion

This paper has developed an effective SCBMA-RS technique to effectively schedule resources in EC environment.The SCBMA-RS model derives a fitness function to identify and allocate resources to every incoming user request in such a way,that the client’s necessities are satisfied with the minimum number of possible resources and optimal energy consumption.In the SCBMA-RS method, the incorporation of the GTOA and RSO algorithms helps to improvise the allocation of resources among VMs in the data center.A series of simulations were performed by the use of the CloudSim tool and the outcomes are inspected interms of distinct performance measures.The experimental results showcased the superior performance of the SCBMA-RS model interms of different measures.In future, the secure data transmission from the EC environment can be accomplished by the use of lightweight cryptographic techniques.

Funding Statement:This research was supported by Hankuk University of Foreign Studies Research Fund of 2021.Also, This research was supported by the MIST (Ministry of Science, ICT), Korea,under the National Program for Excellence in SW),supervised by the IITP(Institute of Information&communications Technology Planing&Evaluation)in 2021”(2019-0-01816).

Conflicts of Interest:The authors declare that they have no conflicts of interest to report regarding the present study.