Cost-efficient resource scheduling in cloud for big data processing using metaheuristic search black widow optimization (MS-BWO) algorithm

Abstract

In a cloud computing system, resources can be accessed at a minimal cost whenever users raise request needs. The primary goal of cloud computing is to provide cost-efficiency of service scheduling to clients fast while using the least number of resources. Cloud Service Provisioning (CSP) can match consumer needs with minimal use of resources. There are several metaheuristic optimization algorithms have been developed in the field of CSP resource minimization and adequate computing resources are required to ensure client satisfaction. However, it performs poorly under a variety of practical constraints, including a vast amount of user data, smart filtering to boost user search, and slow service delivery. In this regard, propose a Black Widow Optimization (BWO) algorithm that reduces cloud service costs while ensuring that all resources are devoted only to end-user needs. It is a nature-inspired metaheuristic algorithm that involved a multi-criterion correlation that is used to identify the relationship between user requirements and available services and thereby, it is defined as an MS-BWO algorithm. Thus finds the most efficient virtual space allocation in a cloud environment. It uses a service provisioning dataset with metrics like energy usage, bandwidth utilization rate, computational cost, and memory consumption. In terms of data performance, the proposed MS-BWO outperforms exceed than other existing state-of-art-algorithms including Work-load aware Autonomic Resource Management Scheme(WARMS), Fuzzy Clustering Load balancer(FCL), Agent-based Automated Service Composition (A2SC) and Load Balancing Resource Clustering (LBRC), and an autonomic approach for resource provisioning (AARP)

Keywords

Cloud service provisioning Resource utilization Virtual machine Metaheuristic black widow optimization

1 Introduction

Cloud service providers employs a variety of techniques to optimize the cloud resource for improving the cloud user accessibility for requested service. However, they are limited by the higher price and consumption of resources. Therefore, having sufficient computing resources is essential for big data processing, several methods have been required by the cloud service providers to meet the needs of their client. To fulfil incoming client requests, Minimum-Provisioning-Cost (MPC) resource provisioning is used. This method overlooks vital resources, such as VM space capacity, when providing cloud services for big data required by cloud customers [22, 37]. Self-provisioning cloud service resources have been outlined to decrease network latency, improve network sustainability, enhance path connection, and cut costs. The problem is that it doesn’t consider every possible way to enhance the circulation of resource funds [7]. Genetic algorithms [8] (GAs) are used to create cost-aware service request scheduling (CSRS) for the cloud. The emphasis is on catering individually to each user who makes a legitimate service request. On the other hand, as the number of valid user requests grows, the complexity of the system rises because of the increased exhaustive search.To address this issue, an optimization algorithm known as Particle Swarm Optimization(PSO) techniques [26 , 30] for cloud resource selection is introduced. Particles explore the problem space in search of an optimal or near-optimal solution. Makespan, Flowtime, and task execution cost are all minimized by the algorithms. However, it was unable to find a solution for global optimization [36].

An effective prediction system, especially useful in the massive data centers that provide cloud-computing services, makes use of the cloud’s infrastructure, which features adaptive resource consumption. Further, improvement of managing a resource allocation is reached by using Autoregressive Neural Networks (AR-NNs) in the prediction system [4]. Augmented Shuffled Frog Leaping (ASFL)-based technique simplifies cloud resource provisioning and process structure used for validation purposes, it simply considers the cloud-computing energy-efficiency, security, and scheduling cost. Due to limited overheads, easier and faster deployment, and higher portability, containers are widely used in cloud architectures [17]. The term "Automated Service Composition" (A2SC) is used to explain how to keep virtual machine costs low while maintaining service. An improved optimal solution can be reached in a reasonable amount of time, reducing the risk of task failure, and increasing reliability. However, it cannot ensure the availability of the necessary resources to deliver the services in a cloud environment [38].

Revenue maximization is a technique introduced in the later period to facilitate the provision of multiple services in a cloud infrastructure. It runs heterogeneous, realistic, and synthetic workloads in a simulated environment and simulates energy-efficient monolithic and parallel-scheduling models [17]. This algorithm for scheduling uses 20% less power while increasing processing speed by 8%. Nevertheless, it was not able to provide the services at the lowest computational cost by combination method includes PSO [29] and Simulated Annealing (SA). In order to schedule multiple objectives at once, cloud computing makes use of a modified version of the bat algorithm and makes its resources available on-demand[?]. The method is effective because it optimizes resource utilization, service quality, performance, and the user experience. Individually, offered more effective on provision cloud computing resources [34]. The cost and memory space of the various services were not considered. Therefore, optimized cloud resource provisioning (OCRP) helps cloud computing systems achieve the lowest cost of service possible [9]. The storage capacity could not be reduced enough using this approach. Similarly, Load Balancing Resource Clustering (LBRC) is presented to effective utilization of the resource through cluster formation where each cluster center is clustered optimally, the meta-heuristic Bat-algorithm can be used to achieve faster convergence [1]. Additionally, it includes a new dynamic task assignment policy to ensure the shortest possible makespan and execution cost [13].

The lack of resource optimization, high computational cost, VMs memory capacity, and other errors arise during resource optimization.Though there are several algorithms available [12 , 33], a natural-inspired optimization technique provides a reasonable better outperformer that could tackle the above-mentioned problems [10]. Therefore, in this paper, Nature-inspired Metaheuristic Search Black Widow Optimization (MS-BWO) algorithm is presented which reduces the cost of cloud computing services and normalized resource allocation. MS-BWO’s primary goal is to provide cloud customers with a wide range of services while consuming a low number of resources. CSP is a two-step process that maximizes resource use. When a user makes a request, a multi-criterion cloud server matches them up with the services that are available in the cloud. End-user services are identified by the correlation. Multi-criteria black widow optimization is developed to ensure that requested services are delivered to customers quickly.

The main contribution of the paper is summarized as follows:

To identify the relationship between user requirements and accessible services using the multi-criterion correlation principle.

Also on the cloud, a metaheuristic search black widow optimization algorithm determines the most efficient cloud virtual machine.

Securing the user’s requested cloud services for big data processing through a virtual machine with minimal resource and computational cost use.

The outlining of the paper structure as a follow-up: Section 2 examines related efforts to address the problems of cloud computing. Section 3 provides an architectural and flow schematic of the MS-BWO technique. Section 4 provides a brief overview of multi-criteria correlation metrics and nature-inspired metaheuristic optimization algorithms. For the proposed MS-BWO method, the performance evaluation is presented in Section 5. Section 6 sums up the study’s findings and suggests further research avenues.

2 Related Works

In this section, cloud computing consumption models such as Software as a Service (SAAS), Infrastructure as a Service (IAAS), and Platform as a Service (PAAS) are organized hierarchically [20], including resource cost-efficient algorithm, cloud computing task scheduling algorithms and an optimization algorithm of round-robin resource allocation mechanism for a better understanding of the proposed MS-BWO method’s progression.

2.1 Resource cost-efficient algorithms

For every properly governed client service request, server resources will respond to data requests from external clients via storage and server datasets. Cloud computing has potential applications in many different areas, but it also faces unique challenges in each of these settings. Applications that are time- and location-sensitive pose a special threat to IoT devices [19 , 28]. Quality of service (QoS) over target times is typically found to be faster at some point during an E-health credential restore [15, 24]. As a means of reducing the computational complexity of cloud computing and, by extension, the energy and cost required to run it, a new clonal selection strategy is introduced [31]. The simulator for cloud optimization of resources and energy (SCORE) tool [6] can be used to determine how efficiently cloud resources are being used. To make the best use of both public and private computing resources within the constraints of time and money, we discuss an ant colony optimization-based multi-objective scheduling (ACO-MOS) approach [21]. However, other resources such as memory or bandwidth are ignored.

2.2 Task scheduling algorithms

Cloud delivery cost per user service is significantly increases to proliferate to meet user needs. The cloud’s complex infrastructure requires efficient resource management and periodic monitoring. Thus, it can provide hardware and on-demand apps over web networks. Task scheduling improves cloud performance by reducing resource use and allocating tasks efficiently. Multi-cloud SLA-based task scheduling achieves efficient resource allocation consider for both the cloud vendor and user [13 , 37]. It is kind of a hybrid method for initializing PSO particles instead of randomly. This strategy can minimize cloud user task execution time and maximize cloud provider resource usage. Results are quite impressive as compared on standard PSO and other heuristic PSO initialization approaches to reduce makespan, execution time, waiting time, and virtual machine imbalance. Subsequentially, studies on multi-tenancy scheduling on cloud platforms to determine the primary Graphics Processing Units (GPUs) scheduling approaches used and addressing key multi-tenancy scheduling issues such as cloud computing, quality of service, and cloud storage [14, 33]. For an effective computing paradigm needs to be benefits modern IoT frameworks supporting dynamic changes of cloud resource of different tasks require different cloud allocations may cause load and power imbalances, affecting cloud based IoT resource utilization and task scheduling. A hybrid bio-inspired algorithm finds discrete and continuous solutions from multiple sources. Thus, it effectively schedules tasks in 2.18-second average response time and 3.6-second average waiting time, respectively [16, 24]. Cloud computing’s on-demand provisioning and resource availability are ideal for scientific workflows. Based on the demand on apps, an Augmented Shuffled Frog Leaping Algorithm (ASFLA)-based technique for resource provisioning and workflow scheduling in IaaS cloud environments. It uses a custom Java-based simulator for testing ASFLA’s efficiency by varying the sizes. The simulation results improve minimum execution cost and schedule deadlines [17, 21]. Dynamic voltage-enabled energy-efficient workflow task scheduling (DV-EWTS) can be used to solve the problem of critical spare time after integrating numerous servers. The heterogeneous-earliest-finish-time (HEFT) algorithm is used to determine the initial scheduling order for all activities. Tasks can be distributed across lower voltage and frequency slots using the DV-EWTS to allow the makespan to be extended [34]. In comparison to traditional virtualization, the container system was introduced in 2013 for sorting out computer resources quickly. Thus, a general-purpose scheduling framework enables cloud service providers or researchers to organize resources more effectively [35].

2.3 Cloud optimization algorithms

To benefit cloud providers who use the most resources while keeping costs low, a multi-objective cuckoo search optimization (MOCSO) algorithm is presented that reduces the time it takes to make decisions. Because of its low production time and low resource requirements, it is well-suited for use in the cloud [18]. Using the cuckoo search algorithm in tandem with meta-heuristic and optimization strategies yielded positive results in cloud computing. The cuckoo-search algorithm is employed to cut back on the time and energy spent on computation. Due to its success in cloud computing, which relies heavily on optimization and meta-heuristics, the provision of cuckoo-search services requires a larger investment of time and money. Shorter wait times for customer support are achieved through a modified round-robin resource allocation mechanism [25]. There was no link discovered between consumer needs and provided services using this method. Using fitness constraints, cloud tasks can be scheduled using the whale optimization algorithm (WOA). Energy, resource effectiveness, and service quality have all been included. Task implementation time and cost on virtual machines are negligible because the scheduled task is contingent on the three constraints. A higher quality of service guarantees a more efficient system and a more logical scheduling of tasks [30]. It was proposed [16] that a hybrid approach using elite-based differential evolution could help with the difficulties of multi-objective work scheduling in cloud computing systems (EDE). It was implemented to broaden exploration possibilities and prevent getting stuck in local optimum solutions. The presented method has been refined to increase its time commitment. The best solution accuracy, algorithm convergence capabilities, and cloud computing performance can all be increased with the help of multi-objective optimization [3], which is based on enhanced particle swarm. The issue of multi-objective optimization has been addressed by the development of a new algorithm. Stochastic optimization of cloud virtual machine and network resource allocation is now complete [2]. If this option had been taken, time and money could have been spared during the process of setting up cloud services. Hadoop Map Reduce now employs cutting-edge methods for determining the optimal allocation of resources for individual tasks [14]. To provide the required application service to customers, it is necessary to manage the load balancing across multiple VMs. Using fuzzy logic and the K-means clustering scheme, manage load balancing across multiple VMs to provide the customer application service. It simply optimizes the selection of VMs from available cloud resources for scheduled service cost. However, it failed to assign multiple service requests from different customers, reducing performance redundancy [11]. In order to provide the application service requested by the customer while utilizing grey wolf optimization in a cloud environment, it is necessary to manage the load balancing on multiple VMs. It is as simple as optimizing the computational speed and convergence rate of the VMs to reduce the cost of providing scheduled services using the resources that are available in the cloud. On the other hand, it was unable to assign multiple service requests from multiple customers, which decreased the resource efficiency and increased the cost [5].

Using the methodologies discussed above, it is difficult to achieve resource-optimized service provisioning in the cloud. The MS-BWO approach is utilized to overcome these cloud-related issues and obstacles.

2.4 Motivation

Effective decision-making processes over metaheuristic cloud space finding search for user requests are necessitated by the virtual space allocation in a cloud environment, which has a major effect on big data resource allocation. A user can be expected to seek out more storage space in the cloud as their big data population expands. A user’s allotment of cloud storage can be increased as necessary without negatively impacting the service’s efficiency. Trying to keep up with the current level of user demand is incredibly difficult. If the current user requests an increase in the utility space, an efficient pre-task service scheduling is initiated to deal with the problem by finding alternative cloud space for the neighbouring big data of others.

2.5 Research gaps identified

The research gap has been identified to include different perspectives on the cloud resource allocation problem to improve cloud base service request accessibility by cloud users. To effectively allocate resources considering big data factors like volume, value, and veracity over cloud space utilisation, an MS-BWO algorithm should be introduced. It does adequately support maintaining a consistent storage footprint across disparate types of big data.

2.6 Problem formulation

Table 1 displays list of notations and its definition helps to understand variables and terms for future use.

Table 1
List of Notations

Variables Definition

VM _p State vector of all virtual machine

VM _map Mapping state

VM _obs Observing state

S _v Virtual space allocation in a cloud

I _req Interactive service request

R _av Available resource in a cloud server

U _i Service request arises from cloud users

U _Req(i) Service accessibility for a receive service

P_s (U_Req(n)) Power required for nth user service request

T _s Delivery time

γ Correlation coefficient

S _a Cloud-based big-data processing services

E _VM Energy utilization of virtual machine

VM _k Number of Virtual machines ∈ k=1,2,...m

VM _Window Candidate widow matrix size R_var × R_pop

R _pop Rate of spider population

R _var Varying row vector

FC Fitness function

β Arbitrary number of window array

M _bs Male black spider

F _bs Female black spider

S₁ & S₂ Offspring’s

$p_{th}^{FC}$ Threshold point of the fitness function

X _bs Position updating of the black spider

X _p Position of prey

$X_{p}^{'}$ Updating position of prey

ω Coefficient of array window

D^′ Updated distance between black spiders

$\min_{error}^{D^{'}} (X_{p}, X_{bs})$ Minimum error

$\max_{error}^{D^{'}} (X_{p}, X_{bs})$ Maximum error

${Avg}_{error}^{D^{'}} (X_{p}, X_{bs})$ Average error

$\min_{ed}$ Minimum Euclidean distance

ed ₀ Initial Euclidean distance of M_bs and F_bs

B _ur Bandwidth utilization

C _cost Computational cost

MU Memory utilization

Variables	Definition
VM _p	State vector of all virtual machine
VM _map	Mapping state
VM _obs	Observing state
S _v	Virtual space allocation in a cloud
I _req	Interactive service request
R _av	Available resource in a cloud server
U _i	Service request arises from cloud users
U _Req(i)	Service accessibility for a receive service
P_s (U_Req(n))	Power required for nth user service request
T _s	Delivery time
γ	Correlation coefficient
S _a	Cloud-based big-data processing services
E _VM	Energy utilization of virtual machine
VM _k	Number of Virtual machines ∈ k=1,2,...m
VM _Window	Candidate widow matrix size R_var × R_pop
R _pop	Rate of spider population
R _var	Varying row vector
FC	Fitness function
β	Arbitrary number of window array
M _bs	Male black spider
F _bs	Female black spider
S₁ & S₂	Offspring’s
$p_{th}^{FC}$	Threshold point of the fitness function
X _bs	Position updating of the black spider
X _p	Position of prey
$X_{p}^{'}$	Updating position of prey
ω	Coefficient of array window
D^′	Updated distance between black spiders
$\min_{error}^{D^{'}} (X_{p}, X_{bs})$	Minimum error
$\max_{error}^{D^{'}} (X_{p}, X_{bs})$	Maximum error
${Avg}_{error}^{D^{'}} (X_{p}, X_{bs})$	Average error
$\min_{ed}$	Minimum Euclidean distance
ed ₀	Initial Euclidean distance of M_bs and F_bs
B _ur	Bandwidth utilization
C _cost	Computational cost
MU	Memory utilization

Consider the state vector of all participating virtual machine (VM) in the scene present is taken as VM_p, which includes VM mapping state VM_map, and VM observing state VM_obs, respectively. Therefore, it is typically written as VM_p = (VM_map, VM_obs). Finding the optimal distribution of cloud-based virtual space, denoted by S_v, is the mission at hand. The pattern of interactive service allocation is observed, and pre-task scheduling information is provided for each service that has been requested from the acquired database. The allocation pattern for interactive services is determined by a combination of factors, which is given in equation (1).

$\begin{matrix} prob (S_{v} / {VM}_{p}) = prob (({VM}_{m} ap, {VM}_{o} bs) / {VM}_{p}) \\ \approx prob ({VM}_{m} ap / {VM}_{p}) \times prob ({VM}_{o} bs / {VM}_{p}) \end{matrix}$ (1)

If an interactive service request is raised from a user I_req, and an available resource in a cloud server R_av. This is achieved by deciphering the shared distribution resulting from the occurrence of a request and the allocation of a given resource. Therefore, the computational complexity of providing access to a distributed service is reduced. The prediction of the marginal service is as follows if the interaction service is not performed: $prob (S_{v} / {VM}_{p}) \approx prob ((I_{req}) / {VM}_{p}) \times prob (R_{av} / {VM}_{p})$ (2)

In the presence of an interactive service request operation, as shown by equation (2), the condition correction terms exist. Even if there are multiple interactive service requests involved in the scene’s trajectory, the looping terms are included in equation(3). Because of the interplay between user requests and cloud services, the marginal and conditional distributions are computed and identify the relation probability together.

$prob (S_{v} / {VM}_{p})_{M > 2} \approx \sum_{i = 1}^{M} prob (S_{v (i)} / {VM}_{p}, S_{i}^{\inf})$ (3)

Where, M represents the total number of service requests. Considering a set of influencers in a cloud-based virtual space $S_{i}^{\inf}$ , the relation predictor estimated the relation correlation of service request. Where, it is a positive integer.

3 Methodology

This section describes the operational flow of the proposed MS-BWO system architecture, in which the user requesting service is handled appropriately by evaluating multi-criterion correlation metrics and the optimal resource management criterion. Each individual criterion’s mathematical description is discussed in detail below.

3.1 System model

CSPs in distributed cloud computing infrastructures provide a great deal of flexibility to Internet-based services that are tailored to match the needs of the customer. Whatever the cost of resource consumption mandatory service provisioning in the cloud is a huge concern since it involves everything from power usage to memory consumption to network bandwidth. The cloud server is configured to utilize as minimal resources as possible to provide the user with a variety of services. Figure 1 depicts the MS-BWO architecture, a cloud resource and cost optimization technique based on customer service demand requests. In order to provide more services to its clients, CSP can use fewer resources by providing a variety of services via a cloud server (CS) by accepting access requests from many different users denoted as U_i = {U₁, U₂, …, U_n} where, i=1,2,...,n. The MS-BWO method compares the quantity of services requested by consumers with the number of services already offered by a cloud server. As an illustration, consider the case of a service request with a resource criterion U_req(i) = {U_req(1), U_req(2), …, U_req(n)}.

Fig. 1

Architecture diagram of proposed MS-BWO.

When a user’s needs are correlated, they are identified. The best resources for delivering the required services to the user are then determined. The basic task of container-based scheduling is to construct different containers for different tasks. The task scheduler is responsible for capturing specified data and always accepting server resource availability. This helps determine the appropriate server for storing the chosen containers. The task schedule oversees auto-scaling servers based on resource availability and task arrival rate in the cloud data center. The cloud environment includes service providers and users’ infrastructures. When more requests for similar resources reach the cloud, there is a resource deficit, and the issue is contemplating resource allocation to users. Each virtual machine has its own specifications, such as memory, cost, and task size. The MS-BWO method is described in detail in the sections that follow.

3.2 Multi-criterion Correlation Metrics (MCM)

When it comes to determining the availability of cloud services, BM-BWO is a key factor. In a cloud context, it uses a multi-criterion correlation metric to match the user’s requested service to the available resources. Service quality and accessibility are examples of factors that can be correlated using this method. Furthermore, the two variables’ link is quantified using more than one criterion in its analysis. Correlation is a typical approach that assesses the cost of bandwidth and memory. The user first sends a series of queries to the cloud server, each with a unique identifier. As far as the cloud server is concerned, users’ requests for additional bandwidth, memory, and other resources are considered. Energy utilization is the amount of electricity a virtual machine uses to provide services to a certain user at a given time. The formula for calculating energy is given in equation (4) $E_{VM} = P_{s} (U_{req (n)}) * T_{s}$ (4)

When computing the total amount of power required to deliver the services, utilize P_s (U_req(n)) and T_s be the delivery time. In the first step, the bandwidth is calculated based on a communication path’s average request transfer rate. Measurement is carried out in Bps (bits per second) (5). $Bandwidth = \frac{U_{req (n)}}{T_{s}}$ (5)

Second, when a customer requests a cloud service, the virtual machine’s memory consumption is the total amount of storage space it takes to provide the service. The units of measurement are megabytes (MB). Third, the amount of time it takes for the cloud server to respond to a user’s request is an important aspect. An evaluation is done using Eqn. (6) to guarantee that the user is looking for services that use as little of these resources as possible.

$γ = \frac{\sum_{i = 1}^{i = n} U_{r e q (n)} * S_{a}}{\sqrt{{(U_{r e q (n)})}^{2} * S_{a}^{2}}}$ (6)

A correlation coefficient lies in the interval of γ ∈ (-1, + 1) used to represent the ‘n’ number of cloud services, and the set of available cloud services for big data processing denoted as S_a. Thus, the server can identify the relationship between user requests and the available resources to speed up the process of responding to them. As you can see, the MCM evaluation is laid out in the following manner.

Algorithm 1

Multi-criterion correlation metrics

1: Input: Number of user requests U_req(1), U_req(2), …, U_req(n-1), U_req(n)

2: Output: Identify the services that are requested by the user

3: for i = 1 to user request U_req(i) do

4: if γ = +1 then

5: Perfect matches exist between user request and service

6: else if γ = -1 then

7: Irregular matches infer with respect to user request and service

8: else if γ = 0 then

9: The service does not have the capacity to handle user requests.

10: end if

11: end for

12: end

The correlation coefficient (γ) yields three different outcomes when used. In other words, there is a one-to-one correlation between the requested and provided services. This means that if the user’s request and the services are perfectly aligned, the condition is denoted as γ = +1. Thereby, Positive correlation is the term used to describe it. Otherwise, the correlation is negative, with γ = -1. Consequently, response times to user requests are reduced.

3.3 MS-BWO technique for optimal resource utilization

To provide optimal user services, VMs are optimized after determining the user’s needs using the multi-criteria correlation metrics. The proposed MS-BWO technique allows the cloud server to provide efficient services with minimal energy, bandwidth, memory, and computational costs. Thus, optimization of cloud servers is based on humpback black spider hunting behavior. Hence, it replicates characteristics when the spider looks for prey and swirls around it, crossing seven important stages namely recognition, orientation, approach, grasp, envenomation, prey paralysis, and final digestion.

Each stage encrusted with a unique characterize function is taken place which is clearly shown in Fig. 2. Multitask scheduling has a significant impact on the scheduling algorithm’s performance in terms of execution time and cost. As a result of the MS-BWO technique, many user arrival requests are effectively answered by making the virtual machine and task execution optimally useful. Consider a specific number of virtual machines (VM₁, VM₂, . . . , VM_m) in order to explain this function. Users’ requests for services are treated as prey by a CSP, which then provides cloud computing as a service through the usage of a VM. To find the best VM option for supplying user-requested services, we followed the siblings of the user’s most often used path with the least number of resources allocated.

Fig. 2

Nature-inspired Metaheuristic Search Black Widow Optimization (MS-BWO) technique.

Figure 2 depicts the influence of MCM on MS-BWO flow that was inspired by nature. It finds the most efficient virtual machines for users. For the user service request assignation problem, the black spider’s nature-inspiring phenomime is considered as a variable row vector, indicated as 1 × R_var. Variable vector quantity is then used to determine the windows of the available virtual machine’s establishment which is given in equation (7).

${VM}_{window} = ({VM}_{1}, {VM}_{2}, . . ., {VM}_{R_{var}})$ (7)

The number of VMs started has a direct impact on the fitness condition (FC), which is displayed in a window with floating integer values. Then, it is expressed as in equation (8). $\begin{matrix} FC & = f ({VM}_{window}) = f ({VM}_{1}, {VM}_{2}, \dots, {VM}_{R_{var}}) \\ \approx \arg {E_{VM} + Bandwidth + γ} \end{matrix}$ (8)

Initializing the spider population creates a candidate widow matrix of size R_var × R_pop, which is then used to begin the optimization process. The reproductive stage is then carried out via mating, in which the female spider eats the male one after completing the mating process.

Given that the pairings are self-contained, they begin to mate with the aim of spawning a new generation. Although each pairing produces thousands of eggs, the number of muscular spider infants remains constant. In the MS-BWO algorithm, the term (β) refers to the arbitrary numbers of windows array, and then offspring are made by using the equation (9), where M_bs and F_bs represent the parents, and S₁ and S₂ represent the offspring.

$β = \frac{s_{1} - (1 - β) \times F_{bs}}{M_{bs}} or β = \frac{s_{2} - (1 - β) \times M_{bs}}{F_{bs}}$ (9) Based on this, there are four optimization setpoints are defined as follows:

Condition	Set rules
If β < 1	Number of offspring growing chance is kept to ‘low’
If β ≥ 1	Number of offspring growing chance is kept to ‘high’
If $β < p_{th}^{FC}$	Male black spider is killed by female spider after mutation, offspring chance is ‘moderate’
If $β \geq p_{th}^{FC}$	Male black spider is alive after mutation, offspring chance is ‘high’

Where $p_{th}^{FC}$ be the thresholding point of the fitness function lies in the interval of (1,0). Taking the minimum argument functions of the above terms gives the smallest value of all VM resources. FC uses little resources and selects the fittest VM. A bubble net is used to feed and search the prey. The cloud services big data processing are then provided by a resource-optimized VM. The black spider locates and surrounds its prey in this phase. So, the optimization algorithm thinks the existing answer is best. This is done to determine the resource optimal VMs. In this way, the black spider updates its position as per the results of equations (10),(11) and (12), equations (10),(11) and (12). $X_{bs} (j + 1) = X_{p} (j) - A * D$ (10) $A = B (2 R_{m_{r}} - 1); B \in (0, 2) & R_{m_{r}} \approx (0, 1)$ (11) $D = | \bar{ω} X_{p} (j) - X_{bs} (j) |; \bar{ω} = 2 R_{m_{r}}$ (12) Where, D denotes the distance between the position vector of the prey X_p (j) and black spider (X_bs (j)), $\bar{ω}$ is the coefficient of array window, and ^′j′ denotes a current iteration. To produce black spider siblings-net behavior, a spiral updating location and declining encircling method are applied. It hunts for prey with bubble nets, and the whale’s new location is determined somewhere between the initial position and the current solution. The position of the black spider and prey is then spiraled to imitate the humpback black spider’s helix-shaped movements. The global optimal is then calculated by updating it and comparing it to another black spider. The updated findings are expressed as follows in equations (14): $X_{bs} (j + 1) = D^{'} e^{r + s} cos cos (2 π 1) + X_{p}^{'} (j)$ (13) $D^{'} = | X_{p}^{'} (j) - X_{bs} (j) |$ (14) Where, s is a random number between -1 and 1, and r is a constant used to represent the logarithmic curve, D’denotes the updated distance between black spiders. The overall position updating, MS-BWO performance is calculated as in equation (15).

$\begin{matrix} X_{bs} (j + 1) = X_{p}^{'} (j) - A * D; \\ p_{th}^{FC} < 0.5 D^{'} e^{r + s} cos cos (2 π q) + X_{p}^{'} (j); \\ p_{th}^{FC} \geq 0.5 \end{matrix}$ (15) Finally, the ideal approach for providing the services required by the user is an updated VM. The MS-BWO technique’s pseudocode is shown in the following manner.

Algorithm 2

Nature-inspired Metaheuristic Search Black Widow Optimization (MS-BWO)

1: Input: Number of user requests U_req(1), U_req(2), …, U_req(n) (i.e., prey), Virtual machines VM₁, VM₂, …, VM_m (i.e., black spider)

2: Output: Select resource-efficient virtual machines for user requests

3: while k < Max_iteration do

4: Set A, B, R_var, R_{m
_r}, r, s

5: for i=1 to R_var do

6: Compute VM_window

7: for j=1 to R_pop

8: Calculate the current best solution’s fitness condition using ((7))

9: if $p_{th}^{FC} < 0.5$ then

10: β < 1 then

11: Update the position to pick the offspring’s utilizing ((10))

12: else if β ≥ 1 then

13: Allow a black spider at random X_r (j)

14: The current best offspring growth is updated ((14))

15: end if

16: else if $p_{th}^{FC} \geq 0.5$ then

17: M_bs is still alive, updating the current best solution ((12))

18: end if

19: Repeat till max iteration is reached.

20: Obtain the greatest service solution.

21: end for

22: end for

23: Perform a new cannibal operation executed

24: Compute mutation rate

25: Update optimal service solution

26: k = k + 1

27: if k ≠ (k - 1) then

28: Assign R₀ = 0

29: else

30: R = R₀ + 1

31: end if

32: end while

33: end

Fig. 3

The Proposed MS-BWO algorithm flowchart for optimizing resource and computational cost.

Based on user demands, the aforementioned algorithm 2 selects the most efficient virtual machine from a pool of available virtual machines. The black-spider numerals (VM) are first set up. When the fitness level is established, the best ideal virtual machine is selected. Position update behavior results in initializing the most optimal black spider. There’s a better black-spider position now, thus the best solution gets selected. To choose a black spider, the location is changed if the likelihood is larger than 0.5. To find the best solution, this process is performed many times. In the end, a resource-optimized virtual computer is the most ideal alternative. According to the proposed model, 0.5 is maintained in terms of likelihood for sufficient optimization. To maintain this condition, the proposed MS-BWO algorithm defines four optimistic points that direct resource allocation based on the service requested by the user. If it exceeds 0.5, the system’s performance for resource allocation in big data processing is suffered. Afterward, the service provider offers the needed cloud services for big data processing using the virtual machine’s optimized resources. The proposed MS-BWO technique algorithm is summarized as a flowchart 3, which is more effective since it makes better use of cloud resources.

4 Performance Evaluation

In this section, we analyze the performance of the proposed MS-BWO technique by assessing several QoS metrics such as energy consumption, bandwidth utilization rate, computational cost, memory consumption, and resource utilization efficiency. We investigate the proposed MS-BWO further by contrasting it with other state-of-art algorithms already in existence, including WARMS [5], FCL [11], A2SC [38], LBRC [1], and AARP [7], respectively.

4.1 Experimental Setup

To evaluate the efficiency of the proposed MS-BWO algorithm,the simulation parameters are presented in Table 2. 100 IoT devices performing numerous activities of size (100000 at maximum extent) are equivalent to 3000 MIPS and 512 MB of processing and memory capacity. Long-term communication is adequate to separate cloud access and data storage server memory data in the cloud space of the speed range of 2 to 5 Mbps correspondingly, in the cloud environment. Using the Java programming language and the cloudsim simulator, an experimental evaluation of the proposed MS-BWO technique has been implemented. In addition, resource-efficient service delivery through the utilization of personal cloud datasets (http://cloudspaces.eu/results/datasets).

Table 2
Simulation parameters

Parameters Specification

Maximum count of IoT devices 100

Available number of multiple tasks 1000

Size of IoT activities 100000

No. of server 10

No. of cloud center 2 (1-accessibility and 1-storage)

CPU capacity ∼4000 MIPS

IoT device RAM Size 512 MB

Individual server’s CPU capacity ∼45,000 MIPS

Individual server’s RAM size 4096 MB

Server storage space 100 GB

Parameters	Specification
Maximum count of IoT devices	100
Available number of multiple tasks	1000
Size of IoT activities	100000
No. of server	10
No. of cloud center	2 (1-accessibility and 1-storage)
CPU capacity	∼4000 MIPS
IoT device RAM Size	512 MB
Individual server’s CPU capacity	∼45,000 MIPS
Individual server’s RAM size	4096 MB
Server storage space	100 GB

The proposed MS-BWO algorithm is tested on five simulated datasets namely (DS₁, DS₂, DS₃, DS₄, DS₅) of varying number of user requests which can be trace out by Google trace log dataset. It’s a data set containing the log entries for six different Google compute clusters over the course of five days. Table 3 shows the number of user service requests can be analyzed, and the best VMs for carrying out those services’ respective tasks can be assigned and displays the costs of the various VM instances hosted on each server, as well as the varying individual sever CPU capacity and RAM size of each VM instance. The tasks’ sizes, expressed in Millions of Instructions (MI), are generated arbitrarily during runtime for experimental purposes. Ten servers with varying loads and capacities are used in this experiment. Task separation and the allocation of available resources identifying the numerical limits of each cluster and categorizing tasks based on how much of a given resource is used. The five created synthetic datasets generating the energy consumption and computing time for the operating devices (containers and VM instances). The MS-BWO algorithm’s two key parameters are and, with anticipated values of 0.2, 0.4, 0.6, 0.8, and 1.0. In the experiment, the population size is [0, 100], and the population values are chosen at random from 20, 40, 60, 80, and 100. An empirical test is used to fix the parameter values. The suggested technique is performed for 100 iterations with each parameter set to a fixed value, recording the $minimum (\min_{error}^{D^{'}})$ , $maximum (\min_{error}^{D^{'}})$ and average $error ({avg}_{error}^{D^{'}})$ . In this experiment, other MS-BWO parameters are kept fixed while one parameter is optimized.

Table 3

Simulation parameters

Parameters	Specification
Maximum count of IoT devices	100
Available number of multiple tasks	1000
Size of IoT activities	100000
No. of server	10
No. of cloud center	2 (1-accessibility and 1-storage)
CPU capacity	∼4000 MIPS
IoT device RAM Size	512 MB
Individual server’s CPU capacity	∼45,000 MIPS
Individual server’s RAM size	4096 MB
Server storage space	100 GB

$\begin{matrix} \min_{error}^{D^{'}} (X_{p}, X_{bs}) & = \min_{ed} (X_{p}, X_{bs}) \\ = \min (\sqrt{\sum_{j = 1}^{j = n} (X_{p}, X_{bs})}) \end{matrix}$ (16)

$\begin{matrix} \max_{error}^{D^{'}} (X_{p}, X_{bs}) n & = \max_{ed} (X_{p}, X_{bs}) \\ = \max (\sqrt{\sum_{j = 1}^{j = n} (X_{p}, X_{bs})}) \end{matrix}$ (17)

$\begin{matrix} {aveg}_{error}^{D^{'}} (X_{p}, X_{ps}) & = \frac{{ed}_{0} (X_{p}, X_{bs})}{c} \\ = \frac{\sqrt{\sum_{j = 1}^{j = n} (X_{p}, X_{bs})}}{c} \end{matrix}$ (18)

Where Euclidean distance between previous and present best solutions are denoted as $\underset{ed}{mini}$ and $\underset{ed}{maxi}$ as in equations (16), (17) and (18). At the initialize stage of Euclidean distance before mate between M_bs and F_basis represented by ed₀. The available nubmer of VMs is counted during the average evaluation before service allocation is given as c,c ∈ (R_var, R_pop), respectively.

This dataset is utilized in the process of active personal cloud measurement, and it accepts two arguments, which are a provider and test type respectively. The script runs the respective files, which are referred to as load_-and_- transfer and service_-variability, depending on the type of test being conducted. The term "service variability" refers to differences in the quality of services of a similar nature that are provided by various virtual machines. For carrying out resource optimal service delivery, certain measurement traces are obtained through the utilization of personal cloud datasets. To account for service unpredictability, the dataset contains 20 columns of field information. During the process of executing resource-optimized service provisioning, out of a total of 17, only a few columns out of the field’s multiple columns are used.

4.2 Performance metric evaluation

The results of the performance metric evaluation as seen in the following sub-sections, which displays the cloud services big data processing that were requested by users. Consider user request (services) range between 10 and 100 for the simulation purpose.

4.2.1 Effect of energy consumption

In computing, the task’s energy consumption is determined by the amount of energy needed by the forwarding channel and the computing server’s resources.

The level of energy consumption is determined based on the required services of the cloud user using equation (4). Existing approaches such as WARMS [5], FCL [11], AARP [7], A2SC [38], and LBRC [1], respectively, are shown to consume significantly more power than the proposed MS-BWO method, which, as can be shown in Figure 4a, saves a significant amount of energy while delivering a variety of cloud services. There are ten runs, each with a different set of user requirements. After every 10 iterations, the energy consumption of the proposed MS-BWO methodology is significantly reduced when compared to that of other methods. According to the findings, the MS-BWO technique is superior to WARMS [5], FCL [11], AARP [7], A2SC [38], and LBRC [1] in terms of its ability to cut overall energy consumption by 42%, 39%, 37%, 33.5%, and 29.5% percentages respectively.

4.2.2 Effect of bandwidth utilization rate

A virtual machine’s bandwidth utilization rate (BUR) is the percentage of total available bandwidth that the machine is using to deliver requested services to its customers. Measurement of bandwidth utilization (B_ur) is done as per equation (19).: $B_{ur} = \frac{Bandwidth Utilized}{Total Available Bandwidth} * 100$ (19)

The results of utilizing the bandwidth are displayed in Figure 4b. The proposed MS-BWO approach uses less bandwidth than the other methods, such as WARMS [5], FCL [11], AARP [7], A2SC [38], and LBRC [1], respectively, when applied in a cloud environment. This is the case. The user makes a number of queries to the server that is hosted in the cloud. The CSP offers services that require a relatively little amount of bandwidth. Within this context, black spider optimization chooses a virtual machine with optimal bandwidth. During optimization, the bandwidth consumption of a single virtual machine is broken down to determine the optimal setting. After that, it looks for the most effective response to the user’s problem. When there are 100 user requests being processed, the virtual machine consumes 36 percent of the available bandwidth. According to WARMS [5], FCL [11], AARP [7], A2SC [38], and LBRC [1], the percentage of bandwidth that is being utilized is, respectively, 56%, 47%, and 45%. When compared to AARP [7], A2SC [38], and LBRC [1], the suggested MS-BWO method has been shown to reduce bandwidth utilization rate by 38.2%, 35.4%, 33.5%, 27.5%, and 25% percentages respectively, after every iteration.

4.2.3 Effect of computational cost

In computing cost, the length of time it takes for a virtual machine to provide a service to the user is quantified [8]. The following equation (20) is used to calculate the computing cost.

$\begin{matrix} C_{cost} & = No . of user request \times \\ allocation of per user request \end{matrix}$ (20)

Figure 4c illustrates the relationship between the number of user requests and the cost of the calculations (i.e., services). Assume that the consumer requested anything between 10 and 100 different services. The total number of individuals who make service requests to a cloud server. The CSP delivers services on-demand that are determined by the available resources. The necessary data can be determined by using the correlation metric. The MS-BWO technique that has been proposed finds the connection between user requests and the services that are offered. As a result, it shortens the amount of time it takes for the cloud server to respond. After determining which services were requested by the user, the MS-BWO chooses a VM that can provide a quicker response. The chosen virtual machine (VM) makes the services readily available to end-users. As a result, the MS-BWO technique requires far less computational effort than other methods. The computational cost of responding to one hundred user requests takes ten milliseconds while using WARMS [5], FCL [11], AARP [7], A2SC [38], and LBRC [1], respectively. The remaining 9 cycles are used to obtain the average results of the comparison. When compared to WARMS [5], FCL [11], AAARP [7], A2SC [38], and LBRC [1], the suggested MS-BWO technique has a computational cost that is reduced by 48.2%, 45.7%, 43%, 37.6%, and 30.4% respectively.

4.2.4 Effect of memory utilization

It is defined as storage space a virtual machine use in order to provide services for its users. Megabytes (MB) is the unit of measurement, and it is expressed as in equation (21).

$\begin{matrix} MU & = No . of user requests \times \\ service provisioning space used \end{matrix}$ (21)

Memory utilization is compared to the services that users have requested in Figure 4d. When compared to WARMS [5], FCL [11], AARP [7], A2SC [38], and LBRC [1], it is abundantly obvious that the suggested MS-BWO approach has a lower impact on memory consumption performance. Due to the fact that MS-BWO chooses the virtual machine (VM) in the population that uses the least amount of memory. Therefore, the virtual machine (VM) that requires the least amount of storage space is selected to provide the services that the user has requested. Take into consideration 100 user requests, 12 MB of storage for the MS-BWO technique, 43MB, 41MB, 39 MB, 32 MB, and 28 MB respectively for the WARMS [5], FCL [11], AARP [7], A2SC [38], and LBRC [1] techniques respectively. When deploying cloud services, this should be kept in mind. When compared to WARMS [5], FCL [11], AARP [7], A2SC [38], and LBRC [1], the MS-BWO approach requires 35% fewer bytes of storage space, 32% fewer bytes of storage space, and 27% fewer bytes of storage space, respectively.

Fig. 4

Comparative analysis of Cloud service optimization factors.

Fig. 5

Execute resource-optimized service provisioning using five different personal cloud datasets.

Fig. 6

Performance evaluation of task allocation with respect to number of service affordable in the stack queues using proposed MS-BWO set points.

4.2.5 Computational speed

The number of VMs is operated for given number of users requested services because ongoing performance of the running application over the VMs makes a request that requires internal memory CPU capacity of ∼4000 MIPS to the virtual machine. All of which are allowed by the VM because the Individual server’s CPU capacity of the VM can execute up to 12,500 user requested services. The storage bandwidth limits for virtual machines with premium storage and premium storage caching enabled are distinct. The computational speed of responding to one hundred user requests takes ten milliseconds while using WARMS [5], FCL [11], AARP [7], A2SC [38], and LBRC [1], respectively. The remaining 9 cycles are used to obtain the average results of the comparison as in Figure 4e. When compared to WARMS [5], FCL [11], AARP [7], A2SC [38], and LBRC [1], the suggested MS-BWO technique has a computational speed that is increased by 46.5%, 43.4%, 41%, 38.6%, and 35.4%, respectively.

4.2.6 Convergence rate

The goals of both pipelining and retiming are to speed up computations by decreasing and/or levelling out delays between user requested services. When there is no more than one combinational operation between any two user-requested services, performance is maximized for a given granularity. When there are 100 user requests being processed, the virtual machine consumes 36 percent of the available bandwidth. According to WARMS [5], FCL [11], AARP [7], A2SC [38], and LBRC [1], the percentage of convergence rate that is being utilized is, respectively, 56%, 47%, 43%, 39% and 38%. When compared to WARMS [5], FCL [11], AARP [7], A2SC [38], and LBRC [1], as in Figure 4f, the suggested MS-BWO method has been shown to reduce bandwidth utilization rate by 35.2%, 32.4%, 29.5%, 26.5%, and 24.2% respectively, after every iteration.

4.3 Dataset computation

As shown in Tables 4 –7, the suggested MS-BWO method was statistically compared to currently available state-of-the-art methods in a cloud environment. Using data from the WARMS [5], FCL [11], AARP [7], A2SC [38], and LBRC [1] datasets and their corresponding energy consumption, bandwidth use rates, computational costs, and memory utilization, the performance metrics of minimum value, maximum value, and average are stated and calculated. When it comes to scheduling and executing a maximum number of user requests in CDC, existing algorithms all performed much worse than the MS-BWO technique.

Table 4
Statistical analysis of energy consumption across five datasets

Dataset Index Proposed MS-BWO WARMS [5] FCL [11] AARP [7] A2SC [38] LBRC [1]

Minimum 9.36 11.816 14.494 18.68 21.32 25.76

DS-1 Maximum 48.25 50.706 53.384 62.43 69.78 76.92

Average 28.805 31.261 33.939 40.555 45.55 51.34

Minimum 10.7 13.156 15.834 20.02 22.66 27.1

DS-2 Maximum 49.59 52.046 54.724 63.77 71.12 78.26

Average 30.145 32.601 35.279 41.895 46.89 52.68

Minimum 11.03 13.486 16.164 20.35 22.99 27.43

DS-3 Maximum 49.92 52.376 55.054 64.1 71.45 78.59

Average 30.745 32.931 35.609 42.225 47.22 53.01

Minimum 11.34 13.796 16.474 20.66 23.3 27.74

DS-4 Maximum 50.23 52.686 55.364 64.41 71.76 78.9

Average 30.785 33.241 35.919 42.535 47.53 53.32

Minimum 13.68 16.136 18.814 23 25.64 30.08

DS-5 Maximum 52.57 55.026 57.704 66.75 74.1 81.24

Average 33.125 35.581 38.259 44.875 49.87 55.66

Dataset	Index	Proposed MS-BWO	WARMS [5]	FCL [11]	AARP [7]	A2SC [38]	LBRC [1]
	Minimum	9.36	11.816	14.494	18.68	21.32	25.76
DS-1	Maximum	48.25	50.706	53.384	62.43	69.78	76.92
	Average	28.805	31.261	33.939	40.555	45.55	51.34
	Minimum	10.7	13.156	15.834	20.02	22.66	27.1
DS-2	Maximum	49.59	52.046	54.724	63.77	71.12	78.26
	Average	30.145	32.601	35.279	41.895	46.89	52.68
	Minimum	11.03	13.486	16.164	20.35	22.99	27.43
DS-3	Maximum	49.92	52.376	55.054	64.1	71.45	78.59
	Average	30.745	32.931	35.609	42.225	47.22	53.01
	Minimum	11.34	13.796	16.474	20.66	23.3	27.74
DS-4	Maximum	50.23	52.686	55.364	64.41	71.76	78.9
	Average	30.785	33.241	35.919	42.535	47.53	53.32
	Minimum	13.68	16.136	18.814	23	25.64	30.08
DS-5	Maximum	52.57	55.026	57.704	66.75	74.1	81.24
	Average	33.125	35.581	38.259	44.875	49.87	55.66

Table 5

Statistical analysis of bandwidth utilization rate across five datasets

Dataset	Index	Proposed MS-BWO	WARMS [5]	FCL [11]	AARP [7]	A2SC [38]	LBRC [1]
	Minimum	13.67	16.126	18.804	21.45	23.67	28.34
DS-1	Maximum	52.45	54.906	57.584	67.89	72.34	82.34
	Average	33.06	35.516	38.194	44.67	48.005	55.34
	Minimum	15.01	17.466	20.144	22.79	25.01	29.68
DS-2	Maximum	53.79	56.246	58.924	69.23	73.68	83.68
	Average	34.4	36.856	39.534	46.01	49.345	56.68
	Minimum	15.34	17.796	20.474	23.12	25.34	30.01
DS-3	Maximum	54.12	56.576	59.254	69.56	74.01	84.01
	Average	34.73	37.186	39.864	46.34	49.675	57.01
	Minimum	15.65	18.106	20.784	23.43	25.65	30.32
DS-4	Maximum	54.43	56.886	59.564	69.87	74.32	84.32
	Average	35.04	37.496	40.174	46.65	49.985	57.32
	Minimum	11.66	20.446	23.124	19.99	22.66	26.68
DS-5	Maximum	38.88	59.226	61.904	46.99	57.78	63.96
	Average	25.27	39.836	42.514	33.49	40.22	45.42

Table 6

Statistical analysis of computational cost across five datasets

Dataset	Index	Proposed MS-BWO	WARMS [5]	FCL [11]	AARP [7]	A2SC [38]	LBRC [1]
	Minimum	7.34	9.796	12.474	15.67	18.34	22.56
DS-1	Maximum	34.56	37.016	39.694	42.67	53.46	59.64
	Average	20.95	23.406	26.084	29.17	35.9	41.1
	Minimum	8.68	11.136	13.814	17.01	19.68	23.9
DS-2	Maximum	35.9	38.356	41.034	44.01	54.8	60.98
	Average	22.29	24.746	27.424	30.51	37.24	42.44
	Minimum	9.01	11.466	14.144	17.34	20.01	24.23
DS-3	Maximum	36.23	38.686	41.364	44.34	55.13	61.31
	Average	22.62	25.076	27.754	30.84	37.57	42.77
	Minimum	9.32	11.776	14.454	17.65	20.32	24.54
DS-4	Maximum	36.54	38.996	41.674	44.65	55.44	61.62
	Average	22.93	25.386	28.064	31.15	37.88	43.08
	Minimum	11.66	14.116	16.794	19.99	22.66	26.68
DS-5	Maximum	38.88	41.336	44.014	46.99	57.78	63.96
	Average	25.27	27.726	30.404	33.49	40.22	45.42

Table 7

Statistical analysis of memory utilization across five datasets

Dataset	Index	Proposed MS-BWO	WARMS [5]	FCL [11]	AARP [7]	A2SC [38]	LBRC [1]
	Minimum	6.78	9.236	11.914	12.45	15.34	20.12
DS-1	Maximum	33.56	36.016	38.694	38.34	47.65	54.23
	Average	20.17	22.626	25.304	25.395	31.495	37.175
	Minimum	8.12	10.576	13.254	13.79	16.68	21.46
DS-2	Maximum	34.9	37.356	40.034	39.68	48.99	55.57
	Average	21.51	23.966	26.644	26.735	32.835	38.515
	Minimum	8.45	10.906	13.584	14.12	17.01	21.79
DS-3	Maximum	35.23	37.686	40.364	40.01	49.32	55.9
	Average	21.84	24.296	26.974	27.065	33.165	38.845
	Minimum	8.76	11.216	13.894	14.43	17.32	22.1
DS-4	Maximum	35.54	37.996	40.674	40.32	49.63	56.21
	Average	22.15	24.606	27.284	27.375	33.475	39.155
	Minimum	11.1	13.556	16.234	16.77	19.66	24.44
DS-5	Maximum	37.88	40.336	43.014	42.66	51.97	58.55
	Average	24.49	26.946	29.624	29.715	35.815	41.495

The simulation is carried out for 100 user requests for IoT device cloud service, and it is computed for every 10-iteration made in the five different datasets. After 10 runs, the energy usage and bandwidth utilization rate of the proposed MS-BWO technique are comparatively lower than other methods. Its evidently shown in Figs. 5a and 5b. The results demonstrate that the MS-BWO strategy reduces energy usage and bandwidth utilization rate by around (38-30%) and (34-25%) as compared to WARMS [5], FCL [11], AARP [7], A2SC [38], and LBRC [1], respectively. Tables 5 clearly show that the proposed MS-BWO is reasonably better for the progressive nature of the datasets. Further, it is evidently operated for non-IoT device efficiency analysis to accommodate the cloud service on the demand of the user request.

Fig. 7

Comparative chart of Resource efficiency with other state-of-the-art-methods.

The proposed MS-BWO approach links user queries to available services. So, it speeds up cloud server response time. Then it selects the fastest VM to provide the requested services. The chosen VM serves consumers promptly. Thus, MS-BWO is less computationally expensive than other approaches. Because MS-BWO selects the memory-efficient VM from the population. Thus, the user’s request is served by the VM with the least storage space. The results demonstrate as in 5c, 5d that the MS-BWO strategy reduces the computational cost and memory utilization around by (44-28%) and (35-24%) as compared to WARMS [5], FCL [11], AARP [7], A2SC [38], and LBRC [1], respectively. Tables 7 clearly demonstrate that the suggested MS-BWO is reasonably superior for datasets with progressive nature.

Fig. 8

Comparative chart of Accuracy with other state-of-the-art-methods.

4.4 Discussion

In this subsection, a description of the task allocation of the cloud customer requested services is provided to ensure that the services are properly accessed in terms of making the most efficient use of the resources that are available. The performance evaluation of the task allocation is depicted in Figure 6a, with the optimistic point set at β = 1. At this point, the number of offspring that can grow is restricted, and as a result, the maximum number of tasks that can be allocated within the average amount of resources is also restricted. However, when compared to the proposed MS-BWO, other already-existing methods have a significantly lower cost. The reason for this is that the number of offspring chances are restricted to stop any further mutations from occurring, which also implies that other cloud customer services are limited while the task at hand is being carried out as required. On the other hand, if there are more offspring than expected, the point will say β > 1, indicating that there are more than 1. After that, the task allocation is tweaked ever-so-slightly, but the proposed MS-BWO technique maintains performance just slightly above the anticipated threshold. To accomplish this, the location of the spider is rapidly changed for a brief period while the storage space allocated to the task that is currently being processed remains unchanged. The proposed MS-BWO is significantly superior to other methods, even under the critical offspring condition. This is the case even though other methods have been used. Figure 6b demonstrates this point very clearly. In a similar manner, an alternative action is being carried out while the male spider has become mutated, and the results may result in a low or high chance of offspring. It is represented as two distinct conditions, which are 0.5< β<1 and 0< β<0.5. As a result, the number of tasks that are allocated is changed without disrupting the cloud’s existing customer service. It can be seen in Figure 6c, as well as in Figure 6d, respectively.

4.5 Impact of resource efficiency and accuracy

The ratio of the number of ideal cloud resources consumed by the virtual machine for producing cloud services big data processing from the overall number of cloud resources is known as resource efficiency. It is expressed as a percentage, and the corresponding formula (22) to this measurement is as follows:

$% Resource efficiency = \frac{optimal resources utilized by VM}{Total available cloud resources} * 100$ (22)

The overall performance of the five different datasets in terms of their resource efficiency and accuracy are estimated. Other techniques, such as WARMS [5], FCL [11], AARP [7], A2SC [38], and LBRC [1], do not compare well to the suggested MS-BWO method’s superior resource use efficiency. Because while delivering cloud services, MS-BWO considers a wide variety of elements, including energy consumption, bandwidth, memory, and the cost of computation. The originally built-up population of virtual machines (VMs) can be optimized by the usage of these characteristics. Then, each VM is checked to ensure that it satisfies the fitness criteria, and that its position is compared to the best solution that is currently available. If the new virtual machine is found to be more effective than the best option already available, then it will be selected. And so on, until a virtual machine is chosen that is both resource and cost-efficient. A VM is used by the CSP to deliver the services that the user has requested. This results in the more effective use of resources. The results of a comparison between the proposed MS-BWO approach and current WARMS [5], FCL [11], AARP [7], A2SC [38], and LBRC [1] are presented in Figure 8. According to these results, MS-BWO improves the use of resource efficiency by 36%, 34%, 33%, 17%, and 12%, respectively.

$Accuracy (%) = \frac{(TP + TN)}{(TP + TN + FP + FN)} * 100$ (23)

\The analysis then extrapolates the VMs’ utilization across a range of user-requested service counts. The health of virtual machines can be determined by comparing them to one of three predefined threshold levels: underfitting, normal, or overfitting. To get the most precise reading of users’ service demand, we put five datasets through their paces by analyzing underfitting and overfitting thresholds. Furthermore, we used Equation 23 to evaluate the proposed MS-BWO models’ precision in identifying VMs’ states with minimal resource and computational cost use. From what can be seen in Figure 8, the optimal underfitting and overfitting thresholds are 75% and 95%, respectively.

5 Conclusion and Future Work

In this paper, an efficient technique called a nature inspired black window optimization with multi-criterion correlation metrics is developed for obtaining resource and cost minimized cloud services big data processing through metaheuristic search phenomena. Therefore, it has achieved an accurate relationship between user requirements and accessible services. Additionally, it has determined the most efficient cloud virtual machine and user’s requested cloud services big data processing availed by minimal resource and computational cost. The MS-BWO approach is tested for energy consumption, bandwidth utilization, computational cost, memory consumption, and resource utilization efficiency. Optimization allows the cloud service provider to provide desired services while maximizing cloud resource usage. The simulation has executed for 100 customer requests for IoT device cloud service, and the results are generated for every 10-iteration in the five different datasets. After ten runs, the proposed MS-BWO methodology uses less energy and has a lower bandwidth use rate than existing methods. The results show that the MS-BWO method reduces energy usage and bandwidth utilization rate by (38-30%) and (34-55%), respectively, when compared to WARMS [5], FCL [11], AARP [7], A2SC [38], and LBRC [1]. Similarly, on the other hand, the MS-BWO technique connects user queries to available services. Thus, it reduces the time it takes for cloud servers to respond and the chosen VM provides prompt service to customers. As a result, MS-BWO is less computationally expensive than other methods. MS-BWO chooses the most memory-efficient VM from the population. Thus, the VM with the least amount of storage space serves the user’s request. The results show that the MS-BWO technique reduces the computational cost and memory use by (44-28%) and (35-24%), respectively, when compared to WARMS [5], FCL [11], AARP [7], A2SC [38], and LBRC [1]. In future work, balance the storage space continuity and reduce user data conservation by enhancing intelligent filtering for the user discovery and faster service delivery over heterogeneous big data. Consequently, it derives conceptual information from machine data to support the business commodity.

References

Adhikari

, Nandy

and Amgoth

, Meta heuristic-based taskdeployment mechanism for load balancing in IaaS cloud, Journalof Network and Computer Applications 128 (2019), 64–77.

Apinaya Prethi

K.N.

and Sangeetha

, A multi-objective optimization of resource management and minimum batch VM migration for prioritized task allocation in fog-edge-cloud computing, Journal of Intelligent & Fuzzy Systems Preprint (2022), 1–11.

Bezdan

, Zivkovic

, Bacanin

, Strumberger

, Tuba

and Tuba

, Multi-objective task scheduling in cloud computing environment by hybridized bat algorithm, Journal of Intelligent & Fuzzy Systems 42(1) (2022), 411–423.

Chalack

V.A.

, Razavi

S.N.

and Gudakahriz

S.J.

, Resource allocationin cloud environment using approaches based particle swarm optimization, International Journal of Computer ApplicationsTechnology and Research 6(2) (2017), 87–90.

Dewangan

B.K.

, Agarwal

, Choudhury

and Pasricha

, Work load aware autonomic resource management scheme using grey wolf optimization in cloud environment, IET Communications 15(14) (2021), 1869–1882.

Fernández-Cerero

, Fernández-Montes

, Jakóbik

, Kołodziej

and Toro

, SCORE: Simulator for cloud optimization of resources and energy consumption, Simulation Modelling Practice and Theory 82 (2018), 160–173.

Ghobaei-Arani

, Jabbehdari

and Pourmina

M.A.

, An autonomic approach for resource provisioning of cloud services, Cluster Computing 19(3) (2016), 1017–1036.

Guerrero

, Lera

and Juiz

, Genetic algorithm formulti-objective optimization of container allocation in cloud architecture, Journal of Grid Computing 16(1) (2018), 113–135.

Gupta

, Bhagat

and Rawat

, Fault aware hybrid harmony search technique for optimal resource allocation in cloud, Journal of Intelligent & Fuzzy Systems 42(4) (2022), 3677–3689.

10.

Gupta

, Rawat

, Tripathi

R.P.

, Mundrad

, Mundra

, Goyal

M.K.

, Kaur

and Agarwal

, Nature inspired fault tolerant task allocation in cloud computing using neural network model, Journal of Intelligent & Fuzzy Systems 43(2) (2022), 1959–1968.

11.

Hamdani

and Aklouf

, Enhanced active VM load balancing algorithm using fuzzy logic and K-means clustering, Multiagent and Grid Systems 17(1) (2021), 59–82.

12.

, He

, Yang

, Nie

and Ran

, Emergency material scheduling optimization model and algorithms: A review, Journal of Traffic and Transportation Engineering (English Edition) 6(5) (2019), 441–454.

13.

Jawade

P.B.

and Ramachandram

, Multi-objective secure task scheduling based on SLA in multi-cloud environment, Multi agent and Grid Systems 18(1) (2022), 65–85.

14.

Jia

, Yang

, Grundy

, Keung

and Hao

, A systematic review of scheduling approaches on multi-tenancy cloud platforms, Information and Software Technology 132 (2021), 106478.

15.

Kaladevi

, Janakiraman

, Ramalingam

and Muthusankar

, An improved ensemble classification based secure two stage bagging pruning technique for guaranteeing privacy preservation of DNA sequences in electronic health records, (2022), 1–18. DOI: https://doi.org/10.3233/JIFS221615

16.

Kalimuthu

R.K.

and Thomas

, An effective multiobjective task scheduling and resource optimization in cloud environment using hybridized metaheuristic algorithm, Journal of Intelligent & Fuzzy Systems (2022), 1–13.

17.

Kaur

and Mehta

, Resource provisioning and work flow scheduling in clouds using augmented Shuffled Frog Leaping Algorithm, Journal of Parallel and Distributed Computing 101 (2017), 41–50.

18.

Kaur

and Kadam

, A novel multi-objective bacteria for aging optimization algorithm (MOBFOA) for multi-objective scheduling, Applied Soft Computing 66 (2018), 183–195.

19.

Kim

N.Y.

, Ryu

J.H.

, Kwon

B.W.

, Pan

and Park

J.H.

, CF-Cloud Orch: Container fog node-based cloud orchestration for IoT networks, The Journal of Super computing 74(12) (2018), 7024–7045.

20.

Kotteswari

and Bharathi

, Performance evaluation of IaaS cloudusing stochastic neural network, Journal of Intelligent &Fuzzy Systems 43(4) (2022), 4613–4628.

21.

Kumar

and Sharma

S.K.

, A Comprehensive Investigation of Workflow Scheduling in Cloud Computing Environment. Ambient Communications and Computer Systems, Springer, Singapore, 2022, pp. 141–152.

22.

, Xia

, Tao

, Yan

, Gao

and Zhan

, Reliability-Aware and Energy-Efficient Work-flow Scheduling in IaaS Clouds, IEEE Transactions on Automation Science and Engineering, 2022. doi: 10.1109/TASE.2022.3195958.

23.

Luo

, Yin

, Hu

, Wang

, Liu

, Fan

and Luo

, Container-based fog computing architecture and energy-balancing scheduling algorithm for energy IoT, Future Generation Computer Systems 97 (2019), 50–60.

24.

Pandi

, Perumal

, Balusamy

and Karuppiah

, A novel performance enhancing task scheduling algorithm for cloud-based E-health environment, International Journal of E-Health and Medical Communications (IJEHMC) 10(2) (2019), 102–117.

25.

Madni

S.H.H.

, Latiff

M.S.A.

, Ali

and Abdulhamid

S.I.M.

, Multi-objective-oriented cuckoo search optimization-based resource scheduling algorithm for clouds, Arabian Journal for Science and Engineering 44(4) (2019), 3585–3602.

26.

Menaka

and Sendhil Kumar

K.S.

, Workflow scheduling in cloud environment–Challenges, tools, limitations & methodologies: A review, Measurement: Sensors (2022), 100436.

27.

Moorthy

R.S.

and Pabitha

, Accelerating analytics using improved binary particle swarm optimization for discrete feature selection, The Computer Journal (2021), bxab089.

28.

Moorthy

R.S.

and Pabitha

, Design of Wireless Sensor Networks Using Fog Computing for the Optimal Provisioning of Analytics as a Service. In Machine Learning and Deep Learning Techniques in Wireless and Mobile Networking Systems, CRC Press, 2021, pp. 153–173.

29.

Ramasamy

, Srirangan

and Ramalingam

, Fuzzy and position particle swarm optimized routing in VANET, InternationalJournal of Electrical and Computer Engineering Systems 12(4) (2021), 199–206.

30.

Ramya

and Padmapriya

, An implementation of energy efficient fuzzy-optimized routing in wireless sensor networks using Particle Swarm Optimization (PSO) and Whale Optimization Algorithm (WOA), Journal of Intelligent & Fuzzy Systems (2022), 1–16.

31.

Saif

M.A.N.

, Niranjan

S.K.

and Al-Ariki

H.D.E.

, Efficient autonomic and elastic resource management techniques in cloud environment: Taxonomy and analysis, Wireless Networks 27(4) (2021), 2829–2866.

32.

Shenbaga Moorthy

and Pabitha

, A study on meta heuristic algorithms for feature selection, In International Conference on Intelligent Data Communication Technologies and Internet of Things, 2018, 1291–1298.

33.

Shenbaga Moorthy

and Pabitha

, Optimal provisioning andscheduling of analytics as a service in cloud computing, Transactions on Emerging Telecommunications Technologies 30(9) (2019), e3609.

34.

Singh

, Juneja

and Malhotra

, A novel agent based autonomous and service composition framework for cost optimization of resource provisioning in cloud computing, Journal of King Saud University-Computer and Information Sciences 29(1) (2017), 19–28.

35.

Tang

, Qi

, Cheng

, Li

, Khan

S.U.

and Li

, Anenergy-efficient task scheduling algorithm in DVFS-enabled cloudenvironment, Journal of Grid Computing 14(1) (2016), 55–74.

36.

Venu

and Vijayanand

K.S.

, Task scheduling in cloud computing: Asurvey, International Journal for Research in Applied Science & Engineering Technology (IJRASET) 8(5) (2020), 2258–2266.

37.

Wang

, Hayat

M.M.

, Ghani

and Shaban

K.B.

, Optimizing cloud-service performance: Efficient resource provisioning via optimal workload allocation, IEEE Transactions on parallel and Distributed Systems 28(6) (2016), 1689–1702.

38.

Zia Ullah

, Hassan

and Khan

G.M.

, Adaptive resource utilization prediction system for infrastructure as a service cloud, Computational Intelligence and Neuroscience (2017).