Quasi Oppositional Dragonfly Algorithm for Load Balancing in Cloud Computing Environment

In cloud computing (CC), load balancing tasks remain a critical problem in distributing resources from a data center. Ensure that every virtual machine (VM) has a balanced load to maximize capacity utilization. In the CC world, load balancing is a Non-Polynomial (NP) problem resolved with metaheuristic algorithms. A new Quasi-Oppositional Dragonfly Algorithm for Load Balancing (QODA-LB) has been developed to obtain optimum resource scheduling in a CC configuration. The proposed QODA-LB algorithm uses three variables to calculate an objective function: execution time, execution cost, and charge. The QODA-LB algorithm assigns tasks to VM according to its potential and the resulting objective function. Also, the QODA-LB algorithm employs the Quasi-Oppositional Based Learning principle to increase the standard convergence rate of the Dragonfly (DA) algorithm. A complete series of experiments were conducted, and the results were analyzed in various ways to ensure the increased efficiency of the QODA-LB algorithm. Simulation results demonstrated an optimal load balancing efficiency and outperformed key approaches.


Introduction
Load balancing in CC is considered a complicated study to separate VM operations in data centers.CC is an essential method for the provision of on-demand Internet services in this situation.The cloud is an extensive interconnected system that uses all files and fields in various ways.To deliver the distribution of resources such as software, hardware, data, and files according to the requirements of alternative machines on the cloud, CC is combined with the concept of distributed and parallel computing.In the shared framework there is a "Pay as You Want" module.The user does not need a computing environment to measure an operation, but an Internet connection to allocate resources when spending money over a while.As a result, it limits the quantity of software purchased that is not required full-time, and CC offers dynamic resource features.VMs are CC processing units that measure and allocate resources dynamically when performing a transaction.
Huge VMs are connected in CC by allocating resources preventatively rather than preventatively.However, resources are not evenly distributed and only a few VMs can do the work.If a task is performed in the cloud, VMs should perform the process quickly and with minimal complexity, and all virtual machines should operate in parallel.Identify the need for job planning and implement with available resources.When multiple tasks are assigned to huge virtual machines, they are all performed simultaneously to complete the tasks.If the transaction is assigned to securities, the scheduler must ensure that all transactions are not invoked in the same VM and that other securities can be used.Therefore, all user tasks on all VMs in CC should be managed by the Scheduler role.To solve the problem of load balancing in all VMs, an intelligent load balancing model is needed.And improve the response time of assigned operations by maximizing the use of available resources, as shown in Fig. 1.

Fig. 1 Load balancing in CC platform
The input task is uniformly distributed across the VMs during load balancing.The primary objective of load balancing is to relieve all CCs of their pressure and report this to other weighted minimum VMs.It features system performance and throughput.The developers also used heuristic and metaheuristic models to resolve load balancing issues.On a distributed network, the protected equilibration is recorded [1].Also, three models were used to address load balancing and security concerns on a distributed network.For example, it provides a mechanism for a mobile agent to broadcast to all nodes on a distributed network.Subsequently, it provides a framework for peer-to-peer load reform to provide maximum efficiency, and, as a result, it provides network security.To solve the load sharing on the Grid IT platform, a hierarchical load balancing strategy is proposed.When dynamically sharing the input task from the VMs and the merits of this application, the overall response time restricts for the Grid application to specifics of the actual node are used.By comparing the speed of a method to the minimum completion time and perfect details at the finish, the speed of the method was validated.
Equilibration based on the nature of honey bees is resolved in non-empty stand-alone tasks on VMs [2].The new method resolves load balancing on VM to improve throughput, limiting the waiting time of tasks by assuming the preferences consequently.The associations performed with alternate models like weighted Round Robin (RR), First In First Out (FIFO), and Dynamic Load Balancing provide the effectiveness of a system for throughput and limit the response time of VMs.Dynamic load distribution across a virtual heat distribution platform [3].It projects two dynamic load balancing models initially it applies local and global load balancing in the distributed virtual platforms by using heat diffusion, and second, it examined two functional aspects like load balancing factor as well as convergence threshold.
An extended Particle Swarm Optimization (PSO) model is being developed to address the balancing problem in the CC system [4].It has led to faster and more efficient implementation.A Tabu Search (TS) approach is developed for resource management within the CC template.The optimized management of dynamic asset planning is deployed according to factors such as time constraints, cost constraints, and optimum solutions.The TS method is used to resolve resource allocation through prioritization and task grouping.The TS method is used to optimize the positions of cloud data center software units such as data routing and network linkage capabilities [5].
The scale and increased use of resources is limited by a simple planning approach in a grid computing platform [6].It provides a balancing model where the load is evenly shared and reduces response time.Fuzzy Logic (FL) is implicit for efficient load balancing and minimizes cost and power in Geo-Distributed multiple datacenters.It features the best offline spatial balancing using the FL inference system.The input data mapping is nonlinear such as the current application of renewable power, accessibility of electric cost as well as power utilization to produce the requests by the data center.
Load balancing is carried out for file systems distributed in cloud systems where it optimizes network traffic by improving the bandwidth of a system [7].The complete distribution of the balancing rebalances model is developed to address the load imbalance and therefore the newly developed approach is linked to the previous centralized approach.An algorithm in line with the Lyapunov optimization strategy is proposed for load-aware eco-management for cloud data centers [8].The main theme of this approach is to reduce the average eco-conscious energy cost of CC data center time when assuring Quality of Experience (QoE) constraints.The honey bee method is executed to limit the makespan and allocates the resource to expand the flow in the CC platform [9].It is considered a dynamical approach that has been applied to identify variations in nature.Among the dependency and independence operations, limit the makespan to all tasks using the task preferences.Loading a server is managed by the selfadjusted randomized optimization application.The energy consumption was reduced during the allocation of resources to each task of the CC platform.Load prediction and resource requirements are defined in the exponentially adjusted moving average.It is expected that the software calculation methods will solve the dynamic load balancing in the CC environment [10].
The load balancing method for CC is conducted using a Genetic Algorithm (GA).Resources are allocated to homogenous and heterogeneous platforms [11].The researchers presented the makespan and power application at the time the resources were allocated.A static load balancing approach is developed to deal with load balancing [12].It uses static data to balance the load without the influence of a cluster node load that contains a lower adaptive capacity.The Bayes system is run for extended load equilibration.An enhanced weighted approach to preliminary rounds is deployed to address load balancing for dependent tasks that do not require a vacuum [13].
The load balancing function is achieved by various modules and the results are related to flow and velocity [14].A variety of programming approaches are deployed for the Map-Reduce environment that involves load balancing for the CC network [15].An adaptive task allocation scheduler is projected to maximize the Map-Reduce function in a different cloudy network [16].The deadline reduction scheduler is developed to reduce the deadline of a task in which it is collapsed while computing the massive data like video and image in Map Reduce frameworks [17].An independent agent which depends upon the load balancing approach is presented for balancing a load by VMs with the help of 3 agents like Channel agent, load agent, and migration agent [18].
This document provides an effective QODA-LB within the CC environment for optimal resource planning.The proposed QODA-LB algorithm derives an objective function using three variables, namely ET, EC and task allocation load at VMs for its capacity.Moreover, the QODA-LB algorithm incorporates the Quasi-Oppositional Based Learning (QOBL) concept to improve the convergence rate of the classic DA.A series of simulations were performed to ensure the efficient performance of the QODA-LB algorithm and the results are examined in several aspects.
Firefly Algorithms (FA) proposed to achieve best average load and improve important indicators such as resource efficiency and task response time.This study also provided certain parameters to assess the success of the proposed hybrid methodology.Apart from similar approaches, the results showed the worst performance in while maximizing the average load through multi-objective optimization [19,20].
When effective load balancing is performed on the cloud, good resource efficiency is achieved.But in cloud computing, load balancing is a NP-hard optimization issue.A novel load balancing task scheduling algorithm in the cloud using the Adaptive Dragonfly Algorithm (ADA) is proposed to solve this problem.The ADA is a hybridization between dragonfly and firefly algorithms.But the issue is a multi-objective feature based on three parameters, namely completion time, processing costs, and load, is designed is not achieved because of heavy load [21,22].

The Proposed QODA-LB Technique
Figure 2 illustrates the system model of the load-sharing methodology presented.The primary objective of the projected model is to allocate all transactions to the VM to load capacity.Load balancing method guides to eliminate tasks from overloaded VMs and allocates to VMs under a loaded step.The proposed approach.
h is composed of massive data centers called Physical Machines (PM) and contains a few VMs to precede the user's tasks.Each CC user has several functions to calculate VM.Here, charges are allocated to VM under a load balancing approach.The framework presented checks a load of all securities in CC.The task of VM is based on the computation time of all loads.

Problem Definition with Solution Framework
Assume cloud C , that have " n " number of PM is comprised of m " number of VM.
where C indicates the cloud, PM 1 shows the initial and PM while PM n implies the nth PM which is represented in the following: where VM 1 is a first VM and VM m refers the final VM .Likewise, i count of users is loaded in cloud and user is composed of i count of task.A user is represented in the following; The main theme of these models is to limit the time and expense of the task and to complete a balanced load of entire VMs in the CC system.Therefore, it is exploited using 3 objectives namely, the limitation of the time of execution of the task, the reduction of the cost of execution and finally, to disperse the load for all VMs in CC.
The overall ET is defined with the help of Eq. ( 4). (1) Fig. 2 Proposed system architecture The EC is determined using Eq. ( 5).
In order to carry out the appropriate scheduling, load balancing must be carried out precisely.In the absence of a balanced load, a system requires more time and costs to execute the operation.To resolve the issue, an effective multi-objective load balancing system was deployed [23].The planned Multi-Objective Function (MOF) is described in Eq. ( 7).

Proposed Load Scheduling Algorithm
The key objective of the presented model is for allocating a task to VM under the application of QODA-LB to reduce the overall ET and EC at the time of balancing the load.Load is a vital attribute of scheduling, where the method of dispersing load across various nodes of an allocated model to improve the response time of resource application tasks.To overcome these issues, multiobjective-based load balancing has been deployed with the help of QODA-LB.DA is defined as a meta-heuristic approach that is evolved by the static and dynamic swarming nature of dragonflies.Dragonflies swarm for 2 targets such as hunting (static swarm) and migration (dynamic swarm).
First of all, the massive dragonflies swarm during roaming over longer distances to different territories causes in the exploration phase.For the static swarm, the dragonfly changes in higher swarms rolling by near deployments and immediate changes in the direction of flight, leading to the operational phase.To improve exploration capability and eliminate the local optimum, the QOBL is incorporated into the DA.The organization chart for the DA approach is presented in Fig. 3.The periodic stages of presentation of the multiobjective load balancing are provided below; Step 1 Initialization A population of dragonflies is arbitrary.A population comprises a set of results.The result is developed according to the number of user tasks as well as VM.Initially, tasks are assigned to VM randomly.Then, according to the Fitness Function (FF) the results are maximized.A first result is provided in Eq. (1).A length of solution implies the supremacy of a task and dragonfly refers to available nodes.There are 10 operations and The previous function is showcased as 1 which is allocated to VM 4 , task 2 is declared to VM 3 and task 10 is declared to VM 4 .According to the solution encoding, population matrix (SM) has been implemented with 0 .or 1.The framework depends upon the asso- ciation between task and VM .The task is related through VM i denotes PM(i, j) = 1 , or PM(i, j) = 0 .However, the column contains a single component for 1, otherwise 0.
Firstly, the initial population is depending upon dragonflies.

Step 2 Fitness calculation
Once the solution is generated, the fitness of all solutions is calculated.The FF is provided in Eq. (9).
Step 3 Update using dragonfly algorithm  In order to enhance the solution, 5 major factors have been applied such as separation, alignment, cohesion, attraction to food and distraction from opponent.Separation is estimated with the help of Eq. (10).
where X implies the location of present individual, X j shows the location of j th neigh- boring individual and N represents the count of neighboring individuals.An alignment is measured utilizing Eq. (11).
where V j indicates the velocity of j th neighbouring individual.Hence cohesion is evaluated by Eq. ( 12).
where X denotes the location of recent individual, N implies count of neighborhoods and X j indicates the location of the jth neighbouring individual.Attraction to food source is determined with the given function: where X depicts the place of present individual and X + refers the place of food source.A direction visible an enemy is measured by: where X signifies a place of recent individual, and X − illustrates the location of the enemy.
Once the position is calculated, then velocity vector is determined by Eq. (15).
where s represents the separation weight, S i is a separation of ith individual, a defines align- ment weight, A i implies alignment of ith individual, c refers the cohesion weight, C i signifies cohesion of ith individual, f illustrates food factor, F i showcases food source of ith individual, e denotes enemy factor, E i refers position of enemy of the ith individual, w dem- onstrates inertia weight, and k depicts the iteration counter.Once the step vector is meas- ured, the position vectors are estimated in the following: where k is the present iteration.

Step 4 Select best solution
When the DA and QODA are compared, DA has optimal fitness value (DA best ) which is lower than DA fitness value (FA best ) , the best place of DA is interchanged by QODA.( 10) Otherwise, AODA fitness value is minimum than DA fitness value, the position is changed by DA.

Step 5 Termination criteria
The iteration is terminated if a better solution is attained.The consequent solution is submitted to CC platform.

Quasi Oppositional Based Learning (QOBL)
It accomplishes maximum attention from developers in Computation Intelligence (CI).The main aim of QOBL in evolutionary processing is to maximize the solution accuracy and simulate the convergence rate to reach global solution.It contains higher probability in optimization method for generating suboptimal solution.Here, recent populations as well as inverse values are produced at the same time and results in optimal candidate solution.Hence, inverse value is produced correctly at mirror position of recent population.From [24], the opposite population contains optimal chance to attain global optimal solution when compared to randomly provided population.The QOBL is obtained by describing 2 of significant mathematical features.

Opposite Number
It is referred as a mirror point of candidate solution from center of search space.When X is a number in real plane with search interval [a, b] , the parallel opposite number (OX) in 1D search space is described by (17).
where X implies the randomly invoked candidate solution, a and b are lower and higher limits of search space.The previous definition is improved to d-dimensional search space and formalized by (18).

Quasi Opposite Number
Furthermore, the mechanism of OBL based learning is improved to quasi-oppositional relied learning that showcases that quasiopposite number is nearby global optimal solution when compared to opposite number.The quasiopposite number is meant to be among the center of search space ( a j +b j ∕ 2 ) while opposite number a j + b j − X j , is depicted by (19).
The pseudo code to accomplish quasiopposite value is given in the following: where r 1 denotes a randomly produced value from (0, 1).

Jumping Rate
It assists the DA to move from recent solution to fresh candidate solution where it has optimal fitness value than present one.According to the jumping rate, as said in (20), novel population is developed and then quasiopposite population has been determined.The selection of jumping value guides DA to eliminate suboptimal solution and stimulate the DA to attain globally optimized solution.Basically, the jumping rate is decided from [0, 0.6].
where J r,mae and J r, min are lower and higher value of jumping rate, NFC max refers higher value in generation and NFC denotes the value of function call at present iteration.

Performance Validation
The performance validation of the QODA-LB algorithm takes place under several aspects and the proposed model is simulated using CloudSim tool.Since the goal of the QODA-LB algorithm is to allocate the tasks to VMs depending upon the capacity (load) of VM.The task is allocated on VM depending upon execution time (ET), execution cost (ET) and load.For the validation of the experimental results of the QODA-LB algorithm, a series of simulations were carried out under diverse configurations namely (i) PM = 5, VM = 15 and 50 tasks, (ii) PM = 10 and VM = 30 and 75 tasks, (iii) PM = 20 and VM = 50 and 100 tasks.Figure 5 displays the execution time analysis of the QODA-LB algorithm under 10 PMs, 30 VMs and 75 operations.The experimental results have illustrated that FA algorithm has showcases inferior results by showing higher execution time.Meantime, the DA and ADA has attempted to show moderate results by accomplishing better execution time.Hence, the QODA-LB model has presented has showcases optimal function by reaching lower execution time.For sample, under the iteration of 10, the QODA-LB scheme has exhibited to a lower execution time of 0.65 ms and higher execution time of 0.93 s, 0.86 s and 0.73 s are obtained by the FA, DA and ADA techniques correspondingly.In line with this, under the iteration of 20, the QODA-LB framework has shown a lower execution time of 0.67 ms while higher execution time of 0.95 s, 0.87 s and 0.75 s are obtained by the FA, DA and ADA methods respectively.In addition, under the iteration of 30, the QODA-LB approach has shown a least execution time of 0.68 ms and higher execution time of 0.95 s, 0.87 s and .76s are accomplished by the FA, DA and ADA methodologies respectively.Followed by, under the iteration of 40, the QODA-LB approach has depicted to a lower execution time of 0.68 ms and higher execution time of 0.95 s, 0.88 s and 0.77 s are achieved by the FA, DA and ADA techniques correspondingly.Moreover, under the iteration of 50, the QODA-LB approach has depicted a lower execution time of 0.69 ms whereas maximum execution time of 0.95 s, 0.89 s and 0.78 s are incurred by the FA, DA and ADA approaches respectively.

Execution Time Analysis
Figure 6 displays the execution time analysis of the QODA-LB method under 20 PMs, 50 VMs and 100 tasks.The experimental results stated that FA model has shown worst results by illustrating higher execution time.Concurrently, the DA and ADA has attempted to showcase better results by reaching minimum execution time.Therefore, the QODA-LB approach has implied moderate performance by reaching lower execution time.For sample, under the iteration of 10, the QODA-LB technique has showcased at least execution time of 0.67 ms while maximum execution time of 0.95 s, 0.90 s and 0.78 s are obtained by the FA, DA and ADA models correspondingly.Likewise, under the iteration of 20, the QODA-LB approach has provided to a lower execution time of 0.68 ms and higher execution time of 0.98 s, 0.90 s and 0.79 s are attained by the FA, DA and ADA methodologies correspondingly.In addition, under the iteration of 30, the QODA-LB approach has offered to a lower execution time of 0.71 ms while maximum execution time of 0.99 s, 0.93 s and 0.80 s are achieved by the FA, DA and ADA methodologies respectively.On the other side, under the iteration of 40, the QODA-LB method has generated to a lower execution time of 0.72 ms whereas greater execution time of 0.99 s, 0.95 s and 0.80 s are obtained by the FA, DA and ADA methods respectively.Furthermore, under the iteration of 50, the QODA-LB framework has showcased to a lower execution time of 0.73 ms whereas high execution time of 0.99 s, 0.96 s and 0.81 s are incurred by the FA, DA and ADA techniques respectively.

Execution Cost Analysis
Table 2 and Figs. 7, 8 and 9 demonstrates the execution cost analysis of the QODA-LB approach under varying number of PMs, VMs and tasks.Figure 7 depicts the execution cost analysis of the QODA-LB algorithm under 5 PMs, 15 VMs and 50 tasks.The experimental results have pointed FA algorithm has shown worst results by showing higher execution cost.Simultaneously, the DA and ADA has attempted to depict considerable results by achieving lower execution cost.Hence, the QODA-LB approach has implied moderate function by accomplishing least execution cost.For sample, under the iteration of 10, the QODA-LB technology has provided to a lower execution cost of 0.69 while maximum execution cost of 0.80, 0.76 and 0.75 are obtained by the FA, DA and ADA methodologies correspondingly.Likewise, under the iteration of 20, the QODA-LB scheme has shown least execution cost of 0.51 while high execution cost of 0.70, 0.67 and 0.60 are accomplished by the FA, DA and ADA techniques correspondingly.Furthermore, under the iteration of 30, the QODA-LB algorithm has shown to a low execution cost of 0.67 while high execution cost of 0.79, 0.78 and 0.74 are incurred by the FA, DA and ADA methodologies correspondingly.Followed by, under the iteration of 40, the QODA-LB model has showcased a minimal execution cost of 0.68 whereas maximum execution cost of 0.80, 0.77 and 0.74 are obtained by the FA, DA and ADA techniques respectively.Furthermore, under the iteration of 50, the QODA-LB approach has provided to a low execution cost of 0.68 whereas higher execution cost of 0.80, 0.77 and 0.76 are incurred by the FA, DA and ADA techniques.For sample, under the iteration of 10, the QODA-LB approach has depicted to a lower execution cost of 0.07 while maximum execution cost of 0.19, 0.12 and 0.08 are acquired   ADA has tried to depict moderate results by accomplishing better execution cost.Thus, the QODA-LB model has showcased considerable performance by reaching lower execution cost.For instance, under the iteration of 10, the QODA-LB method has illustrated to lower execution cost of 0.09 while maximum execution cost of 0.25, 0.15 and 0.11 are achieved by the FA, DA and ADA techniques respectively.Along with that, under the iteration of 20, the QODA-LB algorithm has provided to lower execution cost of 0.10 whereas maximum execution cost of 0.28, 0.16 and 0.12 are obtained by the FA, DA and ADA methodologies correspondingly.In addition, under the iteration of 30, the QODA-LB scheme has resulted to a lower execution cost of 0.11 while high execution cost of 0.29, 0.17 and 0.13 are obtained by the FA, DA and ADA models respectively.
Followed by, under the iteration of 40, the QODA-LB approach has provided to lower execution cost of 0.12 and high execution cost of 0.32, 0.18 and 0.13 are attained by the FA, DA and ADA techniques.Moreover, under the iteration of 50, the QODA-LB algorithm has offered to a lower execution cost of 0.12 whereas higher execution cost of 0.33, 0.19 and 0.14 are incurred by the FA, DA and ADA schemes correspondingly.

Conclusion
This paper has presented an effective QODA-LB algorithm in CC environment to achieve optimal resource scheduling.The main theme of this model is to limit the execution time and expense of the task and to accomplish a balanced load over all VMs in CC system.The proposed QODA-LB algorithm derives an objective function using three variables namely execution time, execution cost, and load.According to the derived objective function, the QODA-LB algorithm allocates tasks to VM with respect to its capacity.The simulation outcome has depicted optimal load balancing performance and demonstrated better results compared to state of art methods.A set of simulations were carried out to examine the execution cost and execution time analysis of the QODA-LB algorithm under varying number of PMs, VMs and tasks.In future, the performance of the QODA-LB algorithm is further enhanced by the use of deep learning models.
) × number of task Number of task ∑ j=1 (ET of corresponding VM × Size of the task) (Total size of VM − Free space of VM) + Size of task) Size of VM (7) MOF = min 1 (ET) + 2 (EC) + 3 (1 − Load)5 knots, and the population length is 10 and the evaluation of all dragonflies could be 1, 2, 3, 4 and 5.One sample of coded solution was offered in Eq.(8)

Fig. 3
Fig. 3 Flowchart of Dragonfly algorithm min NFC max − NFC NFC max of 0.33 ms while maximum execution time of 0.65 s, 0.43 s and 0.39 s are obtained by the FA, DA and ADA methods.Also, under the iteration of 30, the QODA-LB algorithm has provided least execution time of 0.33 ms and higher execution time of 0.67 s, 0.45 s and 0.4 s are attained by the FA, DA and ADA approach correspondingly.On the other side, under the iteration of 40, the QODA-LB model has shown lower execution time of 0.34 ms while high execution time of 0.68 s, 0.47 s and 0.41 s are obtained by the FA, DA and ADA techniques methods.Moreover, under the iteration of 50, the QODA-LB algorithm

Figure 8
Figure 8 implies the execution cost analysis of the QODA-LB technology under 10 PMs, 30 VMs and 75 tasks.The experimental results have shown that FA model has depicted inferior results by showing higher execution cost.Meantime, the DA and ADA has attempted to represent better results by showing minimal execution cost.Hence, the QODA-LB model has implied optimal performance by accomplishing low execution cost.For sample, under the iteration of 10, the QODA-LB approach has depicted to a lower execution cost of 0.07 while maximum execution cost of 0.19, 0.12 and 0.08 are acquired

Fig. 8 Fig. 9
Fig. 8 Execution cost analysis of QODA-LB model under 10 PMs, 30 VMs and 75 tasks Dr. T. P. Latchoumi got her Ph.D. in Computer Science and Engineering from Sathyabama Institute of Science and Technology in 2019.She used to work at Christ college of Engineering and Technology (2010-2015).From 2016 to 2021 she has been working at VFSTR (Deemed to be University) in Andhra Pradesh.Currently she is working in SRM Institute of Science and Technology, Tamilnadu, India.Her research interests include Artificial Intelligence, Machine Learning and sensor networks Dr. Latha Parthiban got her Ph.D. in Computer Science and Engineering from Pondicherry university.Currently she is working in Pondicherry university, India.Her research interests include Artificial Intelligence, Machine Learning and sensor networks.

Table 1 and
4igs.4, 5 and 6 shows the execution time analysis of the QODA-LB algorithm under varying number of PMs, VMs and tasks compared with existing systems FA, DA and ADA.Figure4shows the execution time analysis of the QODA-LB algorithm under 5 PMs, 15 VMs and 50 tasks.The experimental results stated that FA algorithm has demonstrated poor results by exhibiting maximum execution time.At the same time, the DA and ADA has tried to show certainly acceptable results by attaining slightly lower execution time.However, the QODA-LB model has exhibited better performance by attaining minimum execution time.For instance, under the iteration of 10, the QODA-LB algorithm has resulted to a minimum execution time of 0.32 ms whereas higher execution time of 0.63 s, 0.42 s and 0.38 s are incurred by the FA, DA and ADA techniques respectively.Likewise, under the iteration of 20, the QODA-LB model has lead to generate lower execution time

Table 2
Execution cost analysis