1 Introduction

Due to cutting-edge advancements in the internet and its services, cloud computing has been developed quickly and is now inseparable from modern web and internet application design. Various kinds of Service-Level Agreements (SLAs) ensure the availability of good infrastructure and services. The cloud platform suppliers allow resources on lease to support the utilization of infrastructure as per client requirements. The end users use services from service providers, which in turn get services from the infrastructure providers. As such, IT sector giants like Google, Amazon, and Microsoft have come up with cloud platforms to provide enabling services for their customers. The elementary approach of Cloud Computing is to provide necessary resources in accordance with the required services. Many requests are generated instantaneously in cloud systems from various geographically distant stations. An unmanaged scenario results in arbitrarily allocated requests to various cloud service providers, leading to overloading and underloading. This imbalanced load distribution will eventually lead to degraded overall system performance. Developing a dynamic and efficient Load Balancing (LB)scheme is very important.

To develop an efficient Virtual Machine (VM)environment, load-balancing procedures must be implemented into the system. The load balancing scheme will suggest which particular Virtual Machine will be allotted to a particular user’s task, on demand, provided multiple requests may be in the queue simultaneously. To handle the system effectively, some requests must be retained in the queue, allowing the more demanding requests to be forwarded to the corresponding service providers. Therefore, the load balancing mechanism is important to decide which demand/task will continue in the queue and which will proceed for services by the providers. Mehta et al. [1] described load balancing in cloud and fog environments using a decentralized heuristic algorithm. It also takes care of Quality of Services (QoS) along with latency. Another recent case study was defined based on QoS scheduling in cloud and IoT environments [2]. It emphasizes optimized task allocation in cloud-IoT applications using VM. Jawade and Ramachandram developed a hybrid algorithm called Dragon Aided Grey Wolf optimization (DAGWO) [3] to optimize secure task scheduling in a multi-cloud environment. This work also took care of multi-objective optimization.

In [4], the authors presented an inclusive study of various load-balancing schemes and revealed all the possible advantages and drawbacks of various schemes that will direct the researchers towards subsequent improvement and developing new schemes for balancing the load. The author in [5] presented a survey of various efficient algorithms, such as 1. heuristic-centered LB algorithms (Round Robin-based algorithm, Balanced Reduce algorithm, Enhanced Max–Min based algorithm, 2. meta-heuristic-based LB algorithms (Firefly algorithm, Artificial Bee colony, Honey Bee algorithm, and more), and hybrid-centered LB algorithms.

The study [6] targeted all the recent LB schemes developed mainly to fit in the cloud systems. The classification is presented for various LB algorithms. The performance has been analyzed to evaluate the performance of all schemes on the CloudSim simulator. This scheme may be critically important for futuristic research in this domain. The suggested scheme has been implemented and executed in a real-time scenario and a systematic comparison of evaluation criteria is also presented [7].

In the same way, the article [8] concluded multiple LB schemes for cloud systems in an efficient manner as per the required taxonomy. The author concluded that the bandwidth required for the network in the cloud environment is significant and that the high traffic in the network may lead to poor use of networking resources. This may lead to some serious issues such as network failure, data loss, delay, etc. Therefore, several effective LB procedures have been developed for efficient network bandwidth usage[9]. Moreover, in [10], a hybrid LB scheme entitled Throttled and Equally Spread Current Execution has been proposed to reduce the response time, improve the provided services, reduce machine cost, and propose avoiding bottleneck issues. Similarly, the study [11] proposed a better task scheduling scheme, considering Particle Swarm Optimization (PSO) as a root algorithm. It offers the capability to achieve effective outcomes, allowing a huge number of tasks to be scheduled simultaneously without troubling the performance and throughput of the system. Talha and Malki also introduced a new hybrid algorithm, PPTS-PSO [12], which solves the workflow scheduling among independent tasks. Mishra and Majhi gave a hybrid metaheuristic algorithm GA-JAYA [13] that was applied in healthcare data scheduling using VM for a cloud environment. However, this scheme did not effectively reflect the behavior of some important parameters, such as cost or energy consumption. Furthermore, the proposed algorithm must be improved to attain a synchronized workflow while scheduling tasks [14, 15]. Therefore, our proposed work aims to optimize the VM using Lateral Hyena and implement PSO for load balance among VMs, thereby accomplishing concurrent task scheduling.

Several authors propose several techniques [15] that employ PSO and other nature-inspired techniques for load balancing through proper scheduling in the cloud environment. However, the load balancing as a typical scenario of modern cloud applications, which supports both push and pull-based operations, has a complex implementation. This research is primarily focused on the scenario of web-based pull requests. The IoT and sensor applications of the cloud have a different approach, as the devices attached may receive push notifications when certain events are triggered.

  • The main contributions of the proposed work are presented below.

  • Optimization of scheduling of VMs using the proposed scheme, which associates the Lateral Hyenaapproach with Particle Swarm Optimization (LH-PSO).

  • Targeting the proposed scheduling scheme for real-time load distribution.

  • Evaluate performance using parameters like average response time, average load, average turnaround time, memory utilization, runtime, and CPU utilization to measure its effectiveness.

This paper is organized as follows: Sect. 1 –presents an understanding of all the fundamental concepts of Virtual Machine, Cloud Computing, Load Balancing, and Task scheduling in the cloud scenarios. Section 2 presents an extensive review and discusses the analysis of existing schemes for balancing the load in cloud systems. Section 3 outlines the proposed methodology and presents the hybrid scheme to achieve concurrent Load Balancing. Section 4 discusses the outcome obtained and compares the existing schemes by implementing the proposed scheme. The complete study and work have been summarized in Sect. 5.

2 Related work

Researchers have worked on numerous approaches and schemes for balancing the load among nodes in cloud environments. The author in [16] proposed a scheme to distribute load dynamically using a hybrid approach that regulates the velocity of MPSO (Modified Particle Swarm Optimization) combined with a Q-learning algorithm based on gbest and pbest.The scheme managed to balance the load for independent tasks, improving throughput, system performance, makespan, and reducing waiting time. Further, this study also suggested dynamic load balance for the dependent task in cloud environments. A PSO-based scheme in [17] offered a static method for scheduling the independent and non-pre-emptive tasks based on the PSO algorithm. Here, the Aiding LB factor optimizes the performance of basic PSO, and resource utilization is improved by 22%. A resource scheduling scheme in [18] integrated load balancing and task scheduling by implementing a fuzzy-based resource scheduling to obtain efficiency and load optimization in the cloud computing area. The scheme intends to be effective, providing a 7% rise in resource scheduling and 33% in response time. It applies response time and success rate factors to schedule privacy-aware resources.

Furthermore, the authors in [19] suggested a resource-aware scheme in the domain of green computing. As we all are aware, green computing areas are facing issues related to low performance and high energy costs. A resource Aware LB clonal algorithm has been proposed to resolve the above-mentioned issues that perform effectively, thereby enhancing the balance as well. The simulation results demonstrate an effective reduction of energy consumption in green computing. Further, it also proposes the application of energy efficiency schemes for big data centers. The scheme in [20] used genetic algorithms for optimization, combined a greedy approach with a modified genetic algorithm, and used a smaller number of iterations for task scheduling. The scheme performed well and suggested a stochastic-based approach for improving cloud resources.

A capacity-based scheme termed CPDALB in [21] offered load balancing by capacity-focused deadline parameters that mainly emphasize a small cost to satisfy the customer. It ignored workflow-based applications and Quality of Service (QoS) parameters. The results indicated that the proposed scheme outperforms all the competitive algorithms. In [22], the authors suggested a genetic algorithm that uses queuing theory for optimized service selection and a heuristic approach for load balancing. The computational outcomes represent the scheme's effectiveness, but the model fails to represent continuous orders and service failure while servicing and assumes all the tasks to be non-preemptive. The study in [23] pointed to task scheduling as a primary challenge for QoS and cloud optimization, proposing two new hybrid heuristic algorithms. One is fuzzy-based PSO, and the other is a combination of simulated annealing with PSO. The experimentation shows clear and effective results for fuzzy-based PSO in terms of makespan, queue length, waiting time and resource utilization. Further, this study suggested improving the work for the robustness of the proposed scheme by focusing on several QoS parameters. Authors in [24] proposed a combination of the firefly algorithm, which can efficiently perform the minimization of search space, with IMPSO, that improves response time, resulting in the FIMPSO algorithm. The hybrid version effectively improved memory utilization, CPU utilization, throughput, and reliability of the full system.

In [25], the paper presented a systematic review of all the algorithms used for context-aware scheduling in cloud environments. Due to its distributed and dynamic nature, different authors use hybrid or modified PSO to obtain optimum results. The comparison table clearly represents PSO as an efficient approach over the genetic algorithm due to its simple concept, better results, fast convergence, and easy programmability. The authors in [26] suggested a task scheduling algorithm offers stochastic development by recognizing two essential factors, virtual machine selection and task scheduling, as the foremost concern for optimizing different processes. It uses gravitation search for solving optimization and NSGA for exploration. The results show low cost, low power consumption and lesser response time via combination of GSA and NSGA. The study in [27] proposed clustering on the basis of resource provisioning and allocation on the basis of the OCRP optimal cloud resource provisioning algorithm. The scheme resulted in improved cost, execution time, and low memory utilization compared to existing schemes. However, gaps are present that need to be proved and clarified to confirm efficiency. The authors proposed the PSO-ALBA algorithm [28], an adaptive load-balancing approach applied with PSO to enhance the performance in both overloaded and underloaded conditions. The proposed algorithm performed better than the existing heuristic scheduling algorithms.

In [29], Lawanyashri et al. proposed a multi-objective hybrid fruitfly optimization technique based on SA for load balancing in cloud computing environments. The authors carried out two separate experiments. The original FOA and Hybrid FOA were compared in the first, while the second FOA-SA-LBwas compared with PSO, HBB-LB, and EFOA-LB. The proposed method achieves greater performance than existing ones. For load balancing, the authors in [30] have combined evolutionary algorithms and fuzzy scheduling practices to enhance the resource allocation to maintain the LB in cloud providers. The proposed technique selects the most suitable host based on the number of VMs. Further, the tasks are assigned among VMs while maintaining load balance. The authors suggested the use of bounding procedures to further enhance load balancing in the future. The article [31] suggested a wolf-based optimization for cloud applications to efficient task scheduling and load balancing thereby achieving lesser runtime and response time. The author of article [36] considered a novel scheme for task scheduling in cloud environments termed Whale Optimization Algorithm (WOA) that aims to improve the cloud system performance with the available computing resources, resulting in a new proposed scheme termed Improved WOA. The results indicate better accuracy, convergence speed, and resource utilization for small- and large-scale tasks.

The original PSO technique is presented in [32], which gives a baseline for the application of PSO in real-time scenarios involving load balancing / other issues in the cloud and similar environments. Nowadays, IT industries are experiencing a rising trend for cloud computing, and the requirement for storage, resources, and infrastructure is very high. Along with all these challenges encountered by Cloud applications, other key challenges are task scheduling and load balancing. Load balancing ensures even workload distribution among all the nodes to improve resource utilization and system performance. So, the present work tries to optimize the system by distributing a uniform load among all the VMs using Lateral Hyena and PSO, implementing concurrent tasks, and minimizing the overloading and under-loading of VMs.

3 The proposed methodology

The paper suggests a hybrid approach that tries to identify the optimal Virtual machine with an average, assigned load to avoid load imbalance among nodes. The main objectives are loading balance and effective task scheduling by implementing the proposed novel scheme. It can be achieved by following the steps mentioned in Fig. 1.

Fig. 1
figure 1

Flowchart of the proposed algorithm

Full size image

At the initial stage, the generated request or task is submitted to the queue assigned for a particular virtual machine and to the VM manager. Subsequently, VMs will be created, followed by uploading the tasks for a particular VM. To start optimization for the PSO algorithm, all the basic parameters need to be initialized along with the initialization of all the particles by their corresponding position value and velocity value for solving a target problem. LH operation needs to be applied to update the fitness value, where the population will be initialized to evaluate the fitness value. A lateral method will be applied to update the fitness value of each searching agent, and the best fitness value will be recorded. The particle with the best value for velocity and position will serve as a reference for the optimal solution. So, the suggested method, Lateral Hyena-based optimization, optimizes extremely loaded VMs by transmitting the tasks to reduced-load VMs. To examine the effectiveness, the scheme is evaluated for several performance factors, such as memory utilization, CPU utilization, average turnaround time, average load, completion/ runtime, and average response time.

PSO is a meta-heuristic-based optimization methodology that aims to improve a problem using repeated iterations. It aims to obtain a contender solution to the identified issue and maintain a reference quality measure. It resolves the identified problem by retaining and recognizing a candidate solution for the object population (particles) and flow is represented in Fig. 2. Further, there is a need to move the object particles/population around the identified Search Space (SS) based on a defined analytical procedure. The velocity and position are major constraints for the planned analytical method. The change in value for a single particle represents its best-reported position and leads to the best position in SS. The best value shows the way to update for an optimal position, as an outcome, will be accountable for moving the complete swarm near the optimal solutions. This algorithm doesn’t exploit the gradient, which is focused on optimization. So, there is no need to diversify the optimization problem, which is mandatory in some standard optimization schemes such as gradient descent and quasi-newton methods. This work resolves major cloud issues such as TS and LB and obtains an optimal solution.

Fig. 2
figure 2

Flowchart of the PSO algorithm

Full size image

Firstly, Eqs. (1) and (2) decide the initialization of velocity and position of the particles of a swarm [27].

$$\text{Position }: {\text{Y}}_{\text{i}}=({\text{y}}_{\text{i},1}, {\text{y}}_{\text{i},2},\dots \dots \dots ..,{\text{y}}_{\text{i},\text{n}})\upvarepsilon {\text{R}}^{\text{n}}$$
(1)
$$\text{Velocity }: {\text{V}}_{\text{i}}=( {\text{v}}_{\text{i},1}, {\text{v}}_{\text{i},2},\dots \dots \dots .., {\text{v}}_{\text{i},\text{n}})\upvarepsilon {\text{R}}^{\text{n}}$$
(2)
$${\text{V}}_{\text{i}}(\text{j }+ 1) =\text{ Inertia }+\text{ cognitive }+\text{ social}$$
(3)

The component Inertia refers to the “force” component, which is applicable to all particles in the PSO environment. The term “social” refers to the social hierarchy of spotted hyenas and the term cognitive refers to the rate at which the parameters' values are adjusted to get optimum values. This is in conjunction with the term “learning rate” used by most deep learning algorithms based on neural networks.

$$\text{Vi}(\text{j }+ 1) =\upomega \times \text{Vi}(\text{j}) +\text{ k}1 \times \text{ random}1 \times \_\text{PBesti }-\text{yi}(\text{j})\_ +\text{ k}2 \times \text{ random}2 \times \_\text{GBesti }-\text{yi}(\text{j})$$
(4)

Here, ω, k1, k2 represent the constants and rand_1, rand_2indicate the random variable. For further updating, the velocity and position of the particles, Eqs. 5 and 6 can be used.

$$\text{Yi }(\text{j }+ 1)=\text{ Yi }(\text{j }) +\text{ Vi }(\text{j }+ 1)$$
(5)
$$\text{Yi }(\text{j }+ 1)=\text{ Inertia }+\text{ cognitive }+\text{ social}$$
(6)

Algorithm 1 Explains basic PSO.

figure g

The algorithm 1 explains the basic PSO process and helps compute optimal particle velocity and position. In Fig. 2, the flowchart for PSO represents the process more clearly. Initially, the population is initialized by assigning values to all the particles. The next step is to calculate the Fitness Value (FV) for all the particles and determine an improved response compared to the optimal value previously. So that the newly improved value can be set as the next reference for FV. The particles possessing the best fitness value will be selected, and the velocity for selected particles needs to be calculated and updated. All the above-mentioned steps will be repeated till the mentioned criteria are met.

3.1 Lateral hyena

The LH approach follows an iterative way for optimization problems. It is inspired by the hunting strategy and management hierarchy used by spotted Hyena, which attacks big prey in packs. It discovers optimal solutions by exploring search space in a similar way to a pack of Hyenas. The algorithm represents four classes of hyenas: spotted, striped, brown, and ardwolf. Their classification process is implemented to pretend the leadership/management hierarchy. Three important steps for hunting are mainly applied to derive optimization. The steps are mentioned as searching for the prey initially, then encircling the finding, and then attacking the prey. The proposed scheme LH evolved to be an efficient meta-heuristic optimization algorithm. It shows the capability to resolve all the unimodal problems, proves effective results for even complex functions, and is easily able to avoid local minima as well. It claims a higher exploration ability for all the identified multi-modal issues.

In addition, the analytical model for the encircling process for hunting can be represented by Eqs. 7 and 8.

$$\text{x}=\left|{\text{ky}}_\text{p}\left(\text{t}\right)-\text{y}\left(\text{t}\right)\right|$$
(7)
$$\text{y}(\text{t}+1)={\text{y}}_{\text{p}}(\text{t})-\text{ax}$$
(8)

Here, t indicates iteration number, a and k are coefficient vectors, ypis the position vector of the prey and y is LH’s position vector.

$$\text{A}=2*\text{f}*{\text{r}}_{1}-\text{a}$$
(9)
$$\text{C}=2*{\text{r}}_{2}$$
(10)

Here, r1, r2, are random vectors that have values from (0,1), and f is decreased linearly from value (2,0).

$$d_\alpha=\left|k_1y_\alpha\left(t\right)-y\right|,d_\beta=\left|k_2y_\beta\left(t\right)-y\right|,d_\delta=\left|k_3y_\beta\left(t\right)-y\right|$$
(11)
$${y}_{1}={y}_{\alpha }-{a}_{1}\left({d}_{\alpha }\right), {y}_{2}={y}_{\beta }-{a}_{2}\left({d}_{\beta }\right), {y}_{3}={y}_{\delta }-{a}_{3}\left({d}_{\delta }\right)$$
(12)
$$\text{y}(\text{t}+1)= ({\text{y}}_{1}+ {\text{y}}_{2} + {\text{y}}_{3}) / 3$$
(13)

Figure 3 illustrates that spotted Hynas is the most dominant class of Hyena in the entire clan. These are followed by striped, who Brown and Aardwolf Hyenas in turn follow. It presents the leadership and management hierarchy that is followed while catching the prey and securing the clan and its movement. This hierarchy is mapped to the optimization problem's characteristics and dealt with how the hyenas deal with the scenario.

Fig. 3
figure 3

Social dominance hierarchy in Hyenas

Full size image

Algorithm 2 Explains the process of Lateral Hyena.

figure h

Algorithm 2, evidently evidentlysignifies the Lateral Hyena process. The input parameters considered are Problem_Size and Population_Size for the algorithm. Firstly, the population initialization is completed by following the spotted Hyena. The next step evaluates the fitness value FV for all search and grading agents. It selects the first optimal solution, second best, and third best solution for the search agents. Afterward, each search agent's corresponding position needs to be revised until it reports the timing parameter, t1 is less than Max_Iteration. After updating the position value for all the search agents, the value of Pg1_best will be revised by the best FV.

3.2 Lateral hyena based-PSO

This section represents the hybrid approach of Lateral Hyena with the PSO algorithm to solve load-balancing issues in the cloud domain. All the advantages of the PSO approach, such as a convenient model, simple implementation, robustness for parameter control, and efficient computations, mark it better than earlier methodologies. So, this work suggests PSOs find optimal VMs for scheduling tasks to balance load distribution into the system. There is a strong requirement to dodge the probability of local minimum problems due to PSO. The issue of LM can be solved or minimized by applying the Lateral Hyena scheme along with PSO. LH performs very effectively when combined with PSO for task scheduling and load balancing in cloud areas.

LH employs the scheme by moving some particles to certain positions by following adaptive mode instead of random positions. This enhancement is implemented by LH to overcome issues of PSO. The complete process is well defined in steps in Algorithm 3.

Moreover, the LH algorithm may apply task scheduling in parallel with LB, by consumption of the least possible memory, simple implementation procedures, and more rapid convergence ability that illustrates its efficiency for execution in cloud environments.

To calculate the fitness value, LH-PSO uses the following computations for position and velocity update:

$$\begin{array}{c}\begin{array}{c}\mathrm{Updating}\;\mathrm{Position}:\;\;{\mathrm Y}_{\mathrm i}\left(\mathrm j+1\right)={\mathrm X}_{\mathrm i}\left(\mathrm j\right)+{\mathrm V}_{\mathrm i}\left(\mathrm j+1\right)+{\mathrm d}_{\mathrm\alpha}=\left|{\mathrm k}_{\mathrm{ny}}{{\mathrm y}_{\mathrm h}}_{\mathrm i}\left(\mathrm t\right)-\mathrm y\right|\end{array}\\\mathrm{Updating}\;\mathrm{Velocity}:\;{\mathrm y}_{\mathrm i}\left(\mathrm j+1\right)=\;\mathrm{Inertia}\;+\;\mathrm{cognitive}\;+\mathrm{social}\;\left(\mathrm t+1\right)+\left({\mathrm y}_1+{\mathrm y}_2+{\mathrm y}_3\dots\;\dots+{\mathrm y}_{\mathrm n}\right)/\mathrm n\end{array}$$

where, n–- > number, h–- > hyena.

Algorithm 3 represents hybrid Lateral Hyena-PSO process.

figure i

Where,

Pop_Size:

Population Size.

Max_Iter:

Maximum Iteration.

Poss:

Minimum Rate of Possibility.

max_terations_1/2:

Threshold value for stabilization using try and error.

4 Results and discussion

The offered scheme LH-PSO offers a hybrid model for load balancing to achieve efficiency. This segment represents all the results obtained by implementing Lateral Hyena-PSO. Cloudsim simulator has been used to analyze the system's effectiveness. The proposed scheme is compared with the already existing scheme considering average runtime, response time, average load, memory utilization, CPU utilization, and average turnaround time. The initial parameters used for the proposed scheme are presented in Table 1.

Table 1 The initial parameters used in the proposed scheme
Full size table

The efficiency of the proposed method is evaluated in terms of runtime by comparing it with existing methods, such as HBB-LB and EBCA-LB. The results are represented in Table 2.

Table 2 Analysis of runtime of the proposed scheme vs. the existing systems
Full size table

The task length and speed of VMs are important constraints for assigning various tasks to the resources. Here, multiple task assignments are started. When 100 tasks are assigned, EBCA-LB spends 78.1 s, HBB-LB spends 29.5 s, and the fuzzy-based scheme uses 24.5 s. However, the proposed scheme runtime is calculated at 21.25 s for the same count of tasks. Correspondingly, for different numbers of task assignments, such as 100, 150, 200, 250, 300, and 400 tasks, the proposed scheme manages lower runtime than the existing schemes [33]. The graph representation for runtime is in Fig. 4.

Fig. 4
figure 4

The runtime of the proposed scheme vs. the existing systems

Full size image

It points out that the proposed system achieves a lower runtime than the existing one in terms of increasing the number of assigned tasks. Additionally, the suggested scheme is investigated similarly by comparing it with existing approaches, such as HJSA and TSLBACO, concerning runtime evaluation to test efficiency. The results are presented in Table 3.

Table 3 The runtime of the LH-PSO vs. the existing systems
Full size table

The number of tasks assigned is increased further to all the VMs. The results indicate that LH-PSO outperforms HJSA and TSLBACO and is a more efficient algorithm for finding VM optimization. For 200 tasks, the existing method takes 0.51 s, HJSA takes 0.62 s, and TSLBACO takes 0.64 s. Conversely, our proposed scheme completes in 0.46 s only. Similarly, our proposed scheme confirms the lesser amount of runtime compared to 400, 600, 800, and 1000 tasks. Typically, increasing the number of tasks results in higher running time. However, our proposed scheme manages the lowest time against the increase in task assignment. The graphical representation is in Fig. 5.

Fig. 5
figure 5

Runtime of the proposed scheme vs. the existing systems

Full size image

The results plotted for comparative analysis clearly show the proposed scheme reduces runtime for task execution over already existing methods. The computational effectiveness of PSO and parallelism for load balancing by LH method enable the suggested scheme to be efficient enough to minimize overall runtime for various tasks.

The other parameters considered for assessing the proposed scheme are response time, average load, CPU utilization, average turnaround time, and memory utilization. The results are presented in Table 4. It can be observed that the RR technique confirms 0.44 s as the average load, 41.09 ms for the average turnaround time, and the average response time of 30.4 s. In the same way, the existing methods SJF, RR, FCFS, IPSO, Firefly, FF-IPSO and FIMPSO confirm higher values for all the considered parameters and are represented graphically in the under-mentioned figures.

Table 4 Average response time performance of LH-PSO vs. the existing methods
Full size table

Figure 6 clearly displays that the suggested scheme LH-PSO confirms minimum average load compared to the present schemes such as FCFS, RR, SJF, IPSO, Firefly, FF-IPSO and FIMPSO. Correspondingly, the average turnaround time is plotted against all the existing schemes in Fig. 7.

Fig. 6
figure 6

Average load of LH-PSO vs. the existing systems

Full size image
Fig. 7
figure 7

Average turnaround time of LH-PSO vs. the existing methods

Full size image

In Fig. 7, it can be observed that the suggested scheme confirms a lower turnaround time than the existing methods. In addition, the calculated average response time of the proposed scheme is compared using the mentioned schemes, which proves its efficacy for the proposed scheme only. The response time is a prominent aspect of cloud applications. The corresponding results are represented in Fig. 8.

Fig. 8
figure 8

Average response time of LH-PSO vs. the existing schemes

Full size image

The suggested system responds very fast to all the user’s requests, thus demonstrating the system's proficiency in response time. From Fig. 8, it can be easily stated that the response time is minimized compared to the traditional approaches. The markable performance of the proposed system specifies its effectiveness. Normally, hybrid systems result in slow response. However, the proposed system has proved to have higher computational efficiency. This makes them work faster. The proposed scheme LH-PSO is evaluated for memory and CPU utilization for different tasks. The evaluation is represented in Table 5.

Table 5 CPU utilization of the proposed scheme vs. the existing methods
Full size table

Different categories of tasks, such as extra-large, large, medium and small, are considered. The existing RD method used CPU up to a slight extent for all categories of tasks. Similarly, the other methods considered, i.e., WRR, DLB, LB-BC, LB-RC, IPSO-Firefly and FIMPSO, result in poor CPU utilization compared to the proposed scheme. It is examined by experimental analysis that the proposed scheme utilized CPU very efficiently at a level of 75% corresponding to small tasks, 83% for medium-level tasks, 99% for large-sized tasks, and 100% for extra-large-sized tasks. The graphical representation is shown below in Fig. 9.

Fig. 9
figure 9

CPU utilization of the LH-PSO vs. the traditional methods

Full size image

By analyzing Fig. 8, it can be easily identified that the proposed system outperforms all the existing methods in terms of CPU utilization. Increasing the number of tasks typically affects the CPU utilization in terms of efficiency. But, the proposed LH-PSO scheme proves its ability to effectively perform CPU utilization, making it suitable in cloud systems for load balancing. The proposed scheme is further analyzed for memory usage and represented in Table 6.

Table 6 Memory utilization of the proposed scheme vs. the existing systems
Full size table

Also, the proposed scheme finds its effectiveness for memory utilization in comparison to the traditional algorithms. Different categories of tasks are considered for comparison, such as extra-large, large, medium, and small. The methodical results reveal the effectiveness of the proposed scheme by showing small task utilization as 62%, medium task utilization as 76%, large task utilization as85%, and 91% extra-large tasks. We can determine the high efficiency of the proposed scheme compared to the existing methods represented in Table 6. This means in Fig. 10 that the proposed scheme utilizes memory very efficiently in terms of memory for all the task categories.

Fig. 10
figure 10

Memory Utilization of LH-PSO and traditional methods

Full size image

The proposed scheme represents efficiency in terms of memory utilization compared to the existing systems. As the number of tasks increases, the effectiveness of the memory utilization remains balanced for the offered scheme over the existing schemes. It validates that the proposed scheme is more reliable than the existing scheme, even for increasing the number of tasks assigned to the VMs. It expresses the capability to identify the optimized VM, perform efficient LB, and provide a fast response to user requests. The work in [34] implements a hybrid scheme termed ant colony and lion optimization designed to resolve task scheduling problems in cloud applications. However, the work represents improved results by considering fewer parameters such as makespan time, imbalance degree, and response time. The limited number of parameters results in unsatisfactory system performance efficiency. The present work reflected several major constraints for an investigation, like average load, runtime, memory utilization, turnaround time, response time, and CPU utilization. The methodical outcomes represent better performance and are more efficient for our proposed scheme concerning all the parameters. It is analyzed from the experimental results that Lateral Hyena-based-PSO is reliable and suitable for resolving task scheduling issues in cloud applications.

Several benchmark papers that used such metaheuristic approaches discuss the overhead induced by such algorithms. A number of such papers are also discussed in the literature review section. In most cases, after the algorithm has been initialized for parameters when the new request arrives, which prompts the cloud system for algorithm execution, the algorithm usually converges in a small number of iterations. This results in minimal time complexity overhead once the schedular starts execution.

The proposed algorithms can be used efficiently to process complex analytics in bulky datasets, and speed can be accelerated with the help of task scheduling in various computing frameworks. Secondly, it can be used in Scientific calculations/simulations where intensive computation is required. With the help of the proposed task scheduling and load balancing scheme, these tasks can effectively decrease the execution time drastically. In the healthcare industry, it may be used to optimize various tasks such as patients’ treatment plans, analysis of imaging data, and medical diagnoses. In the e-commerce industry, the solution may be used effectively to optimize all possible pricing schemes, improve customer retention, and improve personal recommendations.

5 Conclusion

This work proposes a Lateral Hyena-based PSO algorithm, which recommends improvements to address the basic limitations of the PSO algorithm for scheduling various tasks in cloud scenarios. For the successful working of lateral strategy, the optimization between the exploration phase and exploitation phase plays a very important role. The Lateral schemes can efficiently regulate all the control parameters as per the complexity by instantly calculating the fitness value on demand. The scheme calculates fitness convergence standards, which provide efficient optimization by announcing a threshold for stopping the optimization process. The results outline that the proposed scheme is effective compared to the existing methods by reducing the system's average load, turnaround time, and response time. The offered scheme efficiently uses CPU and memory as compared to the traditional ones.

As a future perspective of the current work, research will be directed towards cloud resources based on the user requirement patterns, which can be deduced using a time-series approach based on the hourly, weekly, and monthly usage patterns across data of a considerable number of years of usage. Also, an investigation will be conducted on the suitability of the variants of PSO in accordance with the variance in the size of incoming requests, as suggested in [35]. This is because most of the variants of PSO techniques, in addition to the proposed algorithm, work best under low variance.