Multi-objective job scheduling in green cloud with big data: a hybrid heuristic approach for minimizing energy in datacenters

Abstract

Job scheduling has become one of the most challenging issues in the research field of data centers within the cloud computing sector. Green cloud computing is a paradigm that employs efficient techniques to significantly reduce carbon emissions, CPU frequency scaling, and overall energy consumption. The large volume of data processed necessitates the consideration of data center strategies to mitigate energy consumption, along with other factors such as environmental impact, operational cost, and system reliability. Thus, this research aims to develop a new job scheduling scheme in a big data-assisted green cloud environment by formulating a multi-objective optimization problem that considers makespan, power consumption, latency, and resource utilization. For handling big data, the MapReduce framework is employed. To implement the green cloud, a Hybrid Firefly Water Strider Optimization (HFWSO) algorithm is developed by combining the Firefly Algorithm (FFA) and Water Strider Algorithm (WSA). This algorithm optimizes the number of servers and jobs allocated based on data volume, ensuring efficient job distribution across servers. Task scheduling is performed to solve a multi-objective function under various constraints. The makespan of the developed model is 148.3 s, energy consumption is 6.72 kWh, and resource utilization is 76.50%. Thus, the findings demonstrate that it effectively minimizes the energy consumption in data centers and also appropriately allocates jobs to machines. These comparative findings also reinforce the suitability of HFWSO as a powerful metaheuristic for multi-objective optimization in energy-efficient cloud computing environments, making it a promising candidate for practical applications requiring balanced trade-offs among makespan, energy consumption, latency, and resource utilization.

Keywords

Big data green cloud job scheduling mapReduce framework hybrid firefly water strider optimization multi-objective function datacentres minimizing energy consumption

1 Introduction

The emergence of big data has significantly impacted recommendation models in business applications. Well-known frameworks such as Apache Spark and Apache Hadoop are widely employed to perform data analytics tasks that can handle enormous volumes of data in both industry and academia.¹ Since multiple jobs often require real-time access, efficient job scheduling becomes essential to enhance system performance and to optimize resource utilization.² As the term “big data” implies, it involves complex and large datasets, making it challenging to use conventional data processing or database management systems. Key challenges include data analysis, searching, capturing, transferring, visualization, storage, and data sharing. The advantage of possessing a large data source lies in the ability to gather critical information from numerous sources. Compared to individual datasets, big data supports various applications such as improving research quality, linking legal citations, estimating real-time traffic conditions, preventing diseases, identifying business trends, and solving crimes. To meet these demands, the primary factor in job scheduling is the average response time, defined as the time elapsed between job assignment and completion.³ Recently, job scheduling has played a pivotal role in enhancing the performance of big data applications.⁴ Some scheduling algorithms, like First-In-First-Out (FIFO) and fair scheduling, are limited in their performance improvements because job sizes vary widely.⁵ Since jobs arrive continuously, scheduling must handle datasets of different sizes effectively.⁶ From a system perspective, reducing job response times helps prioritize smaller jobs, minimizing delays caused by longer-running jobs. Similarly, from the user's perspective, lower response times lead to improved outcomes and satisfaction.⁷

Cloud computing is a technology that enables both the delivery and access of cloud services via the Internet, as well as the use of hardware and software tools to provide these services in a data-centric manner, commonly referred to as Software as a Service (SaaS).⁸ Unlike traditional approaches, cloud data centers differ significantly from conventional data centers.⁹ Typically, cloud data centers consist of numerous network servers, along with networking subsystems, air-conditioning equipment, power distribution units, cooling infrastructure, and associated storage. However, these data centers consume substantial amounts of energy and produce significant carbon emissions due to the handling of hundreds or even thousands of network devices.¹⁰ Another critical component is the network system in cloud computing, which contributes notably to overall power consumption. Both data and applications require reliable transfer and access through cloud resources over Internet-based services.¹¹ Therefore, a major challenge for traditional methodologies is maintaining energy efficiency while reducing network congestion and traffic within the cloud environment.¹²

Generally, energy consumption is a major component of the operational costs of data centers.¹³ Data centers consume a significant amount of energy, often measured in kilowatt-hours per year, and this consumption continues to rise.¹⁴ Therefore, data centers must handle large tasks that involve various computational factors, such as communications, Central Processing Unit (CPU) usage, and memory demands. Depending on the nature of these tasks, each server may experience varying levels of energy consumption and different processing times.¹⁵ To address this, data centers are actively working to reduce energy consumption, which has led to the emergence of green cloud computing as a significant research area. Green computing focuses on utilizing energy-efficient resources, including personal devices, to reduce power usage. The primary goals of reducing energy consumption include improving overall performance, extending battery and network lifetime, meeting performance requirements, lowering power consumption, optimizing resource utilization and profitability, and minimizing CO2 emissions.¹⁶ With the increasing number of users, the volume of jobs that need to be scheduled also increases. Cloud servers typically operate one or more data centers to manage this workload. However, underutilized servers tend to waste energy unnecessarily.¹⁷ Consequently, reducing the number of active servers is an effective approach for lowering energy consumption. To this end, various task scheduling methods have been designed and implemented to reduce energy usage by optimizing how tasks are allocated to servers through algorithmic procedures. Therefore, this paper proposes a scheduling technique designed to minimize energy consumption, network traffic, and congestion.

The primary aspects of the research paper are sequenced as follows.

To design a job scheduling model based on a multi-objective function using big data within a green cloud framework aimed at reducing energy consumption in cloud data centers.

To implement the MapReduce model for handling big data analytics in the green cloud environment, with the primary goal of reducing the dimensionality of large-scale datasets.

To develop a hybrid optimization algorithm, named HFWSO, which integrates the conventional WSA with the classical FFA to achieve optimal job allocation to respective servers or machines.

To formulate a multi-objective function optimized by the proposed HFWSO algorithm. This function incorporates various constraints such as energy consumption, queuing time, delay, number of active servers, and task completion time.

To analyze the performance of the proposed model through comprehensive validation, including comparative and statistical analyses against classical heuristic optimization methods under varying numbers of tasks and Virtual Machines (VMs).

The organization of the research paper is explained as follows. Section II reviews the existing job scheduling work with research gaps and challenging problems. Section III explores the big data-assisted job scheduling in the green cloud, along with its system model. The architectural view of the job scheduling process with the MapReduce framework for handling big data is demonstrated in Section IV. The novel HFWSO algorithm is elucidated for optimizing the tasks and deriving the multi-objective function, which is explained in Section V. Section VI discusses the experimental results. Finally, the new framework ends with Section VII.

2. Literature review

2.1 Existing works

Recently, numerous task scheduling approaches in cloud and fog computing models have been developed. However, energy efficiency, service quality, and scalability remain pressing challenges, and some of the models are explained in this section as follows:

Heuristic-based Methods: In 2019, You et al.¹⁸ described the fair scheduling process as Multi-resource Collapsed Hierarchies (MCH) using a hierarchical structure of a weighted flat tree. MCH was carried out within a feasible time frame. Finally, the simulations were performed using a Google cluster, and the results demonstrated that it achieved the least computation time. Though effective in minimizing computation time, MCH lacked flexibility in adapting to real-time workload variations. Therefore, in 2018, Liu et al.¹⁹ developed the Multiple Job Scheduling and Lightpath Provisioning (MJSLP) model for reducing task completion time. It was processed in two phases: first, it reduced the task completion time by applying a heuristic approach. The simulation results revealed that MJSLP outperformed existing methods in terms of slot utilization and completion time. While MJSLP improved slot utilization and reduced task completion time, its two-phase design was still rule-based and limited by the assumptions embedded in its heuristic formulations. To overcome these limitations, in 2018, Mao et al.²⁰ presented two approaches for task scheduling: a time-aware algorithm and an energy-aware algorithm. Since the key benefits of the two algorithms were complementary, they were integrated into a single model known as the Energy-Performance Trade-off Multi-resource Cloud Task Scheduling Algorithm (ETMCTSA). The enhanced model was implemented in MultiRECloudSim and compared against existing scheduling mechanisms. Due to this parameter optimization, the suggested model demonstrated improved overall performance. However, the energy consumption of this model was relatively low, and the processing time remained high. Subsequently, in 2016, Deng et al.²¹ explored the EcoPower mechanism to schedule loads or jobs distributed across different cloud data centers. The simulation was carried out successfully, but the model faced complexity in addressing the Quality of Experience (QoE) holistically, particularly under heterogeneous and unpredictable loads.

Stochastic and Learning-based Methods: In 2020, Chen et al.²² implemented a Quality of Service (QoS) model to estimate data center performance. In addition, a new model, Cross-Entropy-based Stochastic Scheduling (CESS), was deployed to optimally tune QoS parameters for all jobs within the cloud. The investigated work increased both the service rate and job arrival rate, providing assurance of improved scheduling performance. Yet, it raised concerns about convergence time and generalizability. Later, in 2021, Caviglione et al.²³ developed a Deep Reinforcement Learning (DRL) model to acquire the optimal solution for allocating jobs to respective machines. In contrast to other techniques, the optimization method achieved more accurate results and improved scheduling performance. However, it faced challenges with task security and data loss due to the high task density. Similarly, in 2022, Yan et al.²⁴ implemented DRL to tackle scheduling issues. The main objective of this model was to appropriately allocate tasks to the corresponding virtual machines. The findings proved that it achieved significant improvements in energy consumption and response time, making it applicable for real-time implementations. However, such DRL-based models often involve high computational overhead and limited interpretability. In the same year, He et al.²⁵ proposed a workload scheduling algorithm to effectively reduce power costs under various Service Level Agreements (SLAs). Consequently, a new online control algorithm, named Online Delay-Guaranteed Workload Scheduling (ODGWS), was introduced, which accurately identified the worst delay value for delivering delay-tolerant workloads. The simulations showed that the proposed model reduced power consumption by approximately 5% while maintaining effective job scheduling performance.

Recent Models: In 2025, Ergin et al.²⁶ developed a finite element-based model that helped reduce delay by eliminating stress concentration. Subsequently, a Deep Neural Network (DNN) model with an Evolved Artificial Rabbit Optimization (EVARO) algorithm²⁷ was developed in 2024 by Nguyen et al. The Cauchy motion search mechanism was used to further enhance performance. Despite its improved performance, it faced challenges when executing resource-intensive tasks. To address this, the Blood-Sucking Leech Optimizer (BSLO)²⁸ was developed by Bai et al. in 2024 to solve optimization issues by exploiting its strong exploration and exploitation capabilities. However, in several cases, it tended to get trapped in local optima, limiting global search efficiency. Therefore, an enhanced approach named the Improved Artificial Fish Swarm Algorithm (IAFSA)²⁹ was suggested by Qin et al. in 2024. It was proficient in addressing local optima issues by reducing the overall processing time. However, it could still struggle with problems where the optimal solution was widely dispersed. Finally, the Termite Life Cycle Optimizer (TLCO)³⁰ was developed in 2024 by Li et al. The TLCO effectively addressed optimization problems where the optimal solution was not concentrated by simulating the termite colony's life cycle and social structure. This approach enabled a balance between exploration and exploitation, allowing the algorithm to efficiently search the solution space and avoid getting trapped in local optima.

Thus, this work proposes an enhanced HFWSO model to address these gaps through a robust combination of the FFA and WSA, where FFA enhances local search and intensification, and WSA improves diversification and global search.

2.2 Problem statement

Numerous job scheduling tasks are explained in Table 1. MCH¹⁸ algorithm saves computational time, reduces total run time, and guarantees job separation through its hierarchical design. However, it requires more memory. The MJSLP algorithm¹⁹ reduces job completion time and achieves superior efficiency in terms of average frequency slots and completion time. On the other hand, it consumes more power. The DRL method²³ efficiently handles workload fluctuations and provides a trade-off between costs, security requirements, and the performance of the data center. Conversely, it suffers from user mobility issues. ETMCTSA²⁰ offers a superior balance between efficiency management and performance, and reduces the energy consumption of VMs through optimized allocation. On the contrary, it does not focus on solving the workload imbalance. Another DRL approach²⁴ achieves superior efficiency in terms of lower energy consumption and average response time, higher job success rate and can handle scalable tasks. In contrast, it is not applicable for real-time job scheduling tasks. EcoPower algorithm²¹ offers superior balancing between the user QoE, environmental protection, and power cost savings and is more suitable for large-scale data centers. However, it is not suitable for complex internal power structures of cloud datacenters. The CESS algorithm²² has achieved satisfactory QoS at a reasonable sojourn time and helps handle more work with lower waiting times. However, it suffers from data center overloading. The Online Delay-Guaranteed Workload Scheduling algorithm²⁵ balances trade-offs among service quality, cost optimality, and efficiency in the cost–delay relationship. It suffers from handling high workloads effectively.

Table 1.
Features and challenges of existing job scheduling in the green cloud.

Author [citation] Methodology Features Challenges The gaps addressed by this work and its role

You et al.¹⁸ MCH • This model saves computational time and reduces total run time. • It guarantees job separation with MCH. • It takes more memory. • The developed model aids in addressing the memory inefficiency by optimizing the HFWSO, ensuring reduced memory and robust job mapping.

Liu et al.¹⁹ MJSLP algorithm • It reduces job completion time. • It achieves superior efficiency in terms of average frequency slots and completion time. • It consumes more power. • The developed model helps to reduce the consumption of energy as the developed HFWSO, leading to balanced job load and reduced power usage.

Caviglione et al.²³ Deep reinforcement learning • It efficiently handles the workload fluctuations. • It provides a trade-off between the costs, its security requirements, and the performance of the data center. • It suffers from a user mobility issue. • The adaptive and lightweight nature of the HFWSO helps to handle dynamic resource re-allocation, thus enhancing the robustness of the cloud setup.

Mao et al.²⁰ ETMCTSA • It offers a superior balance between efficiency management and performance. • It reduces the energy consumption of VMs through allocation. • It does not focus on solving the workload imbalance. • The developed HFWSO model infuses the balanced resources and task distribution by employing the behavior of swarms, thus ensuring even allocation of workload.

Yan et al.²⁴ DRL • This approach achieves superior efficiency in terms of less energy consumption and average response time, with a higher job success rate. • It has the ability to handle scalable tasks. • It is not applicable for real-time job scheduling tasks. • By using the hybridized objective tuning and rapid convergence, the proposed HFWSO allows for a real-time decision-making process, which is essential for ideal scheduling.

Deng et al.²¹ EcoPower algorithm • It offers superior balancing between the user QoE, environmental protection, and power cost savings. • It is more suitable for large-scale data centers. • It is not suitable for multifaceted interior power structures of cloud datacenters. • By employing the local searching strategy of firefly, the HFWSO aids in designing the internal data center dynamics to enhance the efficiency of energy allocation.

Chen et al.²² CESS algorithm • It has achieved satisfactory QoS at a reasonable sojourn time. • This helps in handling more work with a lower waiting time. • It suffers from severity of cloud datacenter overloading. • The HFWSO helps to enhance the throughput as it optimizes the job clustering and reduces queue delays even under heavy workloads.

He et al.²⁵ Online Delay-Guaranteed Workload Scheduling • It balances the parameter between service quality and cost optimality. • It is efficiency in the cost–delay relationship. • It suffers from handling the higher workload. • The proposed HFWSO improves the scalability by using the exploration ability of the Water Strider and the exploitation feature of the firefly, leading to enhanced performance during peak load conditions.

Author [citation]	Methodology	Features	Challenges	The gaps addressed by this work and its role
You et al.¹⁸	MCH	• This model saves computational time and reduces total run time. • It guarantees job separation with MCH.	• It takes more memory.	• The developed model aids in addressing the memory inefficiency by optimizing the HFWSO, ensuring reduced memory and robust job mapping.
Liu et al.¹⁹	MJSLP algorithm	• It reduces job completion time. • It achieves superior efficiency in terms of average frequency slots and completion time.	• It consumes more power.	• The developed model helps to reduce the consumption of energy as the developed HFWSO, leading to balanced job load and reduced power usage.
Caviglione et al.²³	Deep reinforcement learning	• It efficiently handles the workload fluctuations. • It provides a trade-off between the costs, its security requirements, and the performance of the data center.	• It suffers from a user mobility issue.	• The adaptive and lightweight nature of the HFWSO helps to handle dynamic resource re-allocation, thus enhancing the robustness of the cloud setup.
Mao et al.²⁰	ETMCTSA	• It offers a superior balance between efficiency management and performance. • It reduces the energy consumption of VMs through allocation.	• It does not focus on solving the workload imbalance.	• The developed HFWSO model infuses the balanced resources and task distribution by employing the behavior of swarms, thus ensuring even allocation of workload.
Yan et al.²⁴	DRL	• This approach achieves superior efficiency in terms of less energy consumption and average response time, with a higher job success rate. • It has the ability to handle scalable tasks.	• It is not applicable for real-time job scheduling tasks.	• By using the hybridized objective tuning and rapid convergence, the proposed HFWSO allows for a real-time decision-making process, which is essential for ideal scheduling.
Deng et al.²¹	EcoPower algorithm	• It offers superior balancing between the user QoE, environmental protection, and power cost savings. • It is more suitable for large-scale data centers.	• It is not suitable for multifaceted interior power structures of cloud datacenters.	• By employing the local searching strategy of firefly, the HFWSO aids in designing the internal data center dynamics to enhance the efficiency of energy allocation.
Chen et al.²²	CESS algorithm	• It has achieved satisfactory QoS at a reasonable sojourn time. • This helps in handling more work with a lower waiting time.	• It suffers from severity of cloud datacenter overloading.	• The HFWSO helps to enhance the throughput as it optimizes the job clustering and reduces queue delays even under heavy workloads.
He et al.²⁵	Online Delay-Guaranteed Workload Scheduling	• It balances the parameter between service quality and cost optimality. • It is efficiency in the cost–delay relationship.	• It suffers from handling the higher workload.	• The proposed HFWSO improves the scalability by using the exploration ability of the Water Strider and the exploitation feature of the firefly, leading to enhanced performance during peak load conditions.

While shifting big data from one data center to another, the cloud requires more energy or power. Thus, green cloud computing emerges, which is performed using various algorithms and techniques that consume less power while maximizing performance in terms of power minimization. Therefore, an energy-efficient task scheduling scheme is required for green computing, which considers its primary objective as the minimization of power consumption. Moreover, scheduling is also one of the most complicated activities within the cloud computing framework, as it maximizes the efficiency of cloud workloads. The major aim of task scheduling is to maximize the makespan utilization, throughput, and overall profitability. This motivation brings attention to job scheduling mechanisms suitable for cloud data centers.

In a green cloud environment, many independent VMs are created and deployed on a single server. While using data centers, energy is required to schedule the tasks. Hence, every data center incurs varying energy costs, power consumption, and carbon emissions. Some former models also face persistent challenges, which are addressed in the proposed work. Thus, the primary objective of this study is to allocate jobs using the aid of an appropriate data center that has lower energy cost and consumption, which helps achieve an efficient green cloud system.

3. Big data-assisted job scheduling in green cloud: System model of green cloud

3.1 System model of green cloud

Green cloud computing has emerged with potential services, applications, and computing capabilities that offer sustainable environmental benefits. It is referred to as a paradigm for server integration, manufacturing, and designing, where the setup enables the effective utilization of cloud resources in an environmentally friendly manner. The key role of employing a green cloud is to minimize energy consumption, leading to enhanced energy efficiency. It also aims to reduce hazardous particles that affect the environment. In general, the green cloud is designed to implement environment-aware computing in data centers. Since it is environmentally focused, power consumption and CO2 emissions are key issues. The system model of the green cloud is illustrated in Figure 1.

Some of the significant components are described below.

Green brokers: In the cloud sector, brokers or consumers receive user requests and schedule them to the respective machines or cloud data centers.

Resource allocator: It serves as the interface between consumers and the cloud environment and is responsible for managing energy-efficient resources.

VMs: They are used for processing requests from users. They dynamically start and stop processes once requests are received and requirements are met.

Router: It is used to transfer requests and responses between the user and the internet.

Carbon emission directory: Once a request is sent, this directory retrieves information related to energy efficiency.

Datacenter: Due to the vast scale of cloud systems, numerous data or requests are received from various companies, industries, or organizations. To handle all this data, a datacenter is utilized, where information is stored securely.

Figure 1.

Diagrammatic view of the green cloud system.

3.2 Big data analytics in green cloud

Big data analytics has become a pivotal component for job scheduling in green cloud environments, mainly due to the massive volume of information produced by the rapid growth of modern technologies. As the name suggests, “big data” involves vast and complex datasets. This data is organized into three layers: hardware-based infrastructure, software-based management, and big data applications that deliver diverse services.

Challenges of using big data in green cloud: As big data comprises a large amount of information and datacenters, it consumes more energy and power. Therefore, higher energy consumption increases the environmental impact. Typically, big data is acquired from cameras, sensors, and other data sources. Due to this large-scale data acquisition, the green cloud faces critical issues, particularly related to energy consumption. The data center is another major concern in big data analytics, as it needs to store large volumes of data, and this storage process results in higher power consumption, making it a significant challenge. Furthermore, big data introduces a dimensionality issue, which can be mitigated by employing the Green Hadoop and MapReduce frameworks. The schematic illustration of big data analytics in the green cloud is depicted in Figure 2.

Figure 2.

Pictorial depiction of the big data analytics model over a green cloud environment.

4. Optimal job scheduling in a big data-assisted green cloud environment

4.1 MapReduce framework for handling big data

Other than the Hadoop method, the MapReduce³¹ process is now considered for handling big data analytics. Nowadays, analyzing big data has become a challenging task; in such scenarios, a dimensionality reduction model is employed to lessen the data size and complexity. Thus, this paper utilizes the MapReduce framework to manage big data in the cloud environment. The schematic model of the MapReduce approach is illustrated in Figure 3.

Figure 3.

Diagrammatic Illustration of the MapReduce framework for handling big data.

The basic components of the MapReduce framework are briefly explained below:

Mapper: It is mainly used to generate intermediate key-value pairs as random numbers or values.

Reducer: After the generation of pairs, the Reducer processes all intermediate values corresponding to their respective intermediate keys.

Partitioner: Initially, it divides the data based on the intermediate key space and ensures that the intermediate key-value pairs are properly assigned to the Reducers.

Combiner: It optimizes performance by aggregating mapped data locally before sorting, thereby saving bandwidth, energy, and power.

Conversely, the model comprises both mapping and reducing functions. Initially, big data consists of large datasets containing irrelevant or redundant information in the form of data, tasks, or jobs. Hence, the mapping function is utilized to reduce the dimensionality of big data, as expressed in Equation (1).

\begin{aligned} B D = [M p (\cdot), R d (\cdot)] \end{aligned}

(1)

Here, the term $B D$ is the final output after applying both mapping and reducing operations on the data, $M p (\cdot)$ is the mapping function applied to the data, and $R d (\cdot)$ is the reducing function. These smaller segments are processed in parallel by the mapper functions. Further, these segmented data are combined in terms of key-value pairs for reducing the size of big data by deriving Equation (2).

\begin{aligned} M p (\cdot) = m p i n g (D_{z}) \end{aligned}

(2)

In the above equation, the mapping is accomplished concerning big data that is annotated as $m p i n g (D_{z})$ . The mapping function $M p (\cdot)$ takes input data and converts it into a set of key-value pairs, and the term $D_{z}$ is a data chunk or segment of the original big data D. The subscript z denotes the $z^{t h}$ partition. After the mapping phase produces intermediate key-value pairs from the segmented data, these mapped results are passed to the reducing phase. Therefore, the reduction can take place via the following equation, as Equation (3).

\begin{aligned} R d (\cdot) = r d c e (M p (\cdot)) \end{aligned}

(3)

Here, the term $R d (\cdot)$ is the reducing function, which takes the set of mapped key-value pairs and processes them to produce a summarized output and $r d c e (M p (\cdot))$ is the final reduced data, which contains one aggregated value per unique key. Thus, the MapReduce framework has been successfully implemented to reduce the large-scale dimension of big data.

4.2 Proposed job scheduling framework

Recently, more individuals have started utilizing cloud services, including social networking, content delivery, streaming media, online gaming, web hosting, and so on. Some practical models operate with distinct characteristics for scheduling jobs in the cloud sector. In addition, it mostly relies on the concept of virtualization, implemented through VMs. Hence, job scheduling with VMs becomes critical, as it involves heterogeneous storage and computing resources. Furthermore, user demand significantly impacts the utilization of cloud datacenters, introducing additional challenges. The scheduling process enhances cloud efficiency by optimizing resource allocation across various scenarios. The major challenge is reducing power and energy consumption, as sometimes machines remain idle. Hence, some scheduling algorithms are deployed to improve energy conservation and promote the green computing paradigm. Maintaining energy efficiency and minimizing energy consumption are key motivations, as cloud datacenters handle enormous data and requests. Over the past few years, various heuristic algorithms have significantly addressed the scheduling issue. However, since these algorithms often produce suboptimal results, job scheduling deployment remains difficult. Moreover, the generated solutions frequently fail to enhance performance or reduce operational costs. Due to the inherent characteristics of green cloud computing, job scheduling remains a challenging process, as it requires significant energy consumption. Therefore, a novel job scheduling method is proposed for the green cloud environment to mitigate energy consumption in datacenters, as illustrated in Figure 4.

Figure 4.

Architectural representation of the proposed job scheduling mechanism using big data in the green cloud.

The main objective of the proposed scheduling is to minimize energy consumption in big data-related green clouds using multi-objective optimization functions. In order to handle big data, the MapReduce framework is implemented to address large-scale dimensionality challenges. Subsequently, the HFWSO is developed to render the optimal solution to implement the green cloud. This novel algorithm optimizes server and job allocation while enhancing scheduling performance in big data environments. The optimal scheduling process aims to address multiple objectives, including reducing energy consumption in datacenters, minimizing job queuing time, decreasing task allocation delays, lowering task completion time, minimizing makespan, and improving resource utilization. Finally, the performance is validated under multiple constraints, and the results demonstrate that the proposed method effectively minimizes energy consumption while efficiently allocating jobs to machines.

5 Developing hybrid firefly water strider optimization for optimal job scheduling in green cloud

5.1 Proposed HFWSO-based scheduling

The proposed algorithm is developed by combining two conventional algorithms, namely WSA and FFA, as they achieve a rapid convergence rate and effectively resolve local optimum issues. On the other hand, when using these algorithms separately, they have some drawbacks, such as difficulty in finding the global optimum, challenges in handling large-scale data, lack of adaptability, and sensitivity to high-dimensional search spaces. To combat all the issues, a new hybrid algorithm is proposed, termed HFWSO.

HFWSO introduces a fitness-aware adaptive control mechanism that guides the switching behavior between the two algorithms. Unlike WSA's fixed random number generation within the range [0,1], the proposed HFWSO computes a dynamic switching parameter based on population statistics, specifically the mean fitness, best fitness, and worst fitness, as shown in Equation (4).

\begin{aligned} r n d = \frac{m n F t - b t F t}{w t F t - b t F t} \end{aligned}

(4)

In the above equation, the random value $r n d$ is generated by the mean fitness as $m n F t$ , and best and worst fitness values as $b t F t$ and $w t F t$ , respectively. Thus, the random value is obtained by Equation (4). If the value is greater than 0.5, then the updating process is done by WSA; else, it is done through FFA.

WSA³²: The WSA is inspired by the behavior of water striders, an insect species known for their remarkable ability to move and survive on the water's surface. Due to this fascinating capability, they have attracted significant attention in the field of optimization. The behavior of water striders can be divided into several stages, including territory establishment, mating, foraging behavior, and death.

Mathematical model

Initialization: During the initialization phase, water striders are represented as agents created from eggs distributed across the lake's surface. This initialization process is mathematically expressed using Equation (5).

\begin{aligned} D_{x}^{0} = u b + r n d (u b - l b), w h e r e x = 1, 2, \dots, X \end{aligned}

(5)

Here, the upper and lower bound value is indicated by $u b$ and $l b$ , respectively. The term X refers to the total water striders, and $r n d$ is the random value that lies between [0, 1]. Further, the initial location of $x^{t h}$ strider is noted by $D_{x}^{0}$ . Once the initialization is done, the fitness value is computed for each search agent for subsequent processing.

Establishing territory region: The territory establishment is required to survive, feed and mate. Let us consider the total territories as Y. Initially, with the aid of fitness value, it is divided into distinct groups as $\frac{X}{Y}$ . Hence, the $y^{t h}$ territory is filled with $y^{t h}$ members. Subsequently, the female striders have to identify the best location for feeding that declares the best fitness representing the female, whereas the worst signifying the male strider as keystone, accordingly.

Mating: This is accomplished by ripple signals to produce the new striders. At the beginning stage, the male sends the ripple signals, and on the other side, the female responds with either repulsion or attraction of signals. Thus, both the attraction and repulsion are carried out by the probability value as b. The mating happens between the striders once the female sends the attraction signal, where the new position is obtained. When the female rejects the signal, then the male strider gets away from the place that leads to generating the new location. It is formulated using Equation (6).

\begin{aligned} D_{x}^{n + 1} = {\begin{array}{cc} D_{x}^{n} + E \cdot r n d, & i f m a t i n g i s d o n e \\ D_{x}^{n} + E \cdot (1 + r n d), & e l s e \end{array} \end{aligned}

(6)

Here, the term $D_{x}^{n}$ refers to the position concerning $n^{t h}$ iteration. The random number $r n d$ is calculated newly by using Equation (4). Then, the variable E declares the Euclidean distance measure that is computed by the initial position of the male $D_{x}^{n - 1}$ and the final position of the female strider is $D_{f s}^{n - 1}$ in the same territory area. Thus, the distance can be calculated using Equation (7).

\begin{aligned} E = D_{f s}^{n - 1} - D_{x}^{n - 1} \end{aligned}

(7)

Feeding: After mating, the females consume more energy, which tends to increase the food intake. This behaviour relies on the concept of the objective function. If the value exceeds the preceding state, the strider has identified the food. Otherwise, it moves forward to reach the best value of more fitness strider. Therefore, the new position is generated by Equation (8).

\begin{aligned} D_{x}^{n + 1} = D_{x}^{n} + 2 \cdot r n \cdot (D_{b t}^{n} - D_{x}^{n}) \end{aligned}

(8)

Here, the term $D_{b t}^{n}$ is used to update the position around the best search agents.

Strider death and larva succession: Once the food attainment process is done, the striders are compared with the objective value. If the value of the new fitness becomes lower, the keystone is dead in the designated territory region; otherwise, it remains to survive in the world. Further, the new larva is matured and acts as a keystone. Based on this new generation, the positions are arbitrarily initialized. It is shown in Equation (9).

\begin{aligned} D_{x}^{n + 1} = l b_{y}^{n} + 2 \cdot r n (u b_{y}^{n} - l b_{y}^{n}) \end{aligned}

(9)

Here, the upper and lower boundary values of the strider position in $y^{t h}$ territory are $u b_{y}^{n}$ and $l b_{y}^{n}$ , correspondingly.

FFA²⁸: It considers the firefly population to do the optimization process. The brightness traits of fireflies mainly determine it. The following three rules are listed below.

Due to its bright nature, the firefly can attract other fireflies.

A high level of brightness leads to achieving the high attractiveness feature.

The low-brightness firefly moves towards the brighter firefly.

Based on the criteria, the FFA is accomplished. The search agents are initialized with the position. It is expressed as Eqution (10).

\begin{aligned} F_{x, y}^{0} = F_{x, min} + r d \cdot (F_{x, max} - F_{x, min}), w h e r e x = 1, 2, \dots, X \end{aligned}

(10)

The $x^{t h}$ variable of $y^{t h}$ agent is noted by $F_{x, y}^{0}$ , the minimum and maximum values mentioned as $F_{x, max}$ and $F_{x, min}$ , respectively. Based on the current iteration, when the fitness value of one solution is greater than another, the displacement is calculated and given in Equation (11).

\begin{aligned} D t_{u, v} = \sqrt{{(F_{u} - F_{y})}^{2}} \end{aligned}

(11)

In order to update the new position, the $u^{t h}$ firefly is determined by its corresponding value. Hence, the new solution is obtained through Equation (12).

\begin{aligned} F_{u v}^{n e w} = F_{u} + α \cdot r d \cdot Δ F_{u v} + r d \end{aligned}

(12)

Here, the variable as $r d$ signifies the random value and the term $α$ is computed using Equation (13).

\begin{aligned} α = α_{0} e^{- η D t_{u, v}^{2}} \end{aligned}

(13)

The term $α_{0}$ represents the zero distance attractiveness that is fixed as 1 and the updated value of firefly as $Δ F_{u v}$ , which is formulated by Equation (14).

\begin{aligned} Δ F_{u v} = (F_{u} - F_{y}) \end{aligned}

(14)

Finally, a new solution is generated regarding the global best value determined by Equation (15).

\begin{aligned} F_{u}^{n e w} = {\begin{array}{ccc} F_{u} & i f & F_{u} \in F_{g b s t} \\ F_{u g b s t}^{n e w} & i f & F_{v} \in F_{g b s t} \\ F_{u v}^{n e w} w i t h f t_{b s t} & e l s e \end{array} \end{aligned}

(15)

In the aforementioned equation, the term $F_{u}$ defines the $u^{t h}$ solution in firefly and the new solution is generated when the firefly has the second finest value, which is indicated by $F_{u g b s t}^{n e w}$ . The global best solution is denoted by $F_{v}$ and the best fitness is noted as $f t_{b s t}$ , which is compared with the new solution as $F_{u v}^{n e w}$ . Finally, the global value is annotated as $F_{g b s t}$ . If the first two conditions are not satisfied, the $u^{t h}$ search agent is retained with a low fitness value. The pseudo-code for the proposed algorithm is explained below.

Figure 5.

Flow chart representation of the suggested HFWSO algorithm.

The flow chart representation of the novel hybrid algorithm is shown in Figure 5.

5.2 Derived multi-objective function

The multi-objective function is derived by optimizing the task and machine allocation process. During job scheduling, the developed HFWSO algorithm optimally selects the jobs and VMs. For example, the model considers 50 tasks and 3 machines, and the optimization process is carried out to determine which job should be allocated to which machine. Due to these optimal allocation results, the proposed method can effectively minimize various datacenter performance factors. The key benefit of designing the multi-objective function lies in enhancing system performance by reducing energy consumption, execution time, resource utilization, makespan, queuing delay, and latency, while simultaneously improving QoS parameters. Thus, the objective function for the proposed job scheduling approach is expressed in Equation. (16).

\begin{aligned} O F = \underset{{T_{x}}}{\arg min} [E g y + Q T + D e + T + A S] \end{aligned}

(16)

Here, the $T_{x}$ represents the tasks that varies from 100 to 500. The $E g y$ signifies the energy consumption of data centres, queuing time as $Q T$ , delay as $D e$ , T and $A S$ refers to the task completion time and the number of active servers that is minimized while allocating the jobs. Thus, this scheduling model efficiently enhances the scheduling performance.

5.3 Constraints in multi-objective function

Diverse constraints are employed for formulating the multi-objective function, which is described below.

Energy consumption ( $E g y$ ): It is defined as the evaluation of how much energy is consumed while allocating the jobs to certain machines. It is calculated using Equation (17).

\begin{aligned} E g y = P w r \times \frac{t}{1000} \end{aligned}

(17)

Here, energy is measured through Kilowatt-hours and power $P w r$ is represented by Watts and time t is in hours, correspondingly.

Queueing time ( $Q T$ ): Queuing time measures the waiting time a job spends in the queue before execution. It arises due to resource contention and scheduling delays. Minimizing queuing time enhances system throughput and responsiveness, reducing user-perceived latency.

Delay ( $D e$ ): Delay refers to the average time taken by tasks from the point of job submission to the beginning or end of their execution on VMs. It is expressed using Equation (18).

\begin{aligned} D e = \frac{max \sum_{x = 1}^{X} T_{x}}{T o t a l t a s k s} \end{aligned}

(18)

Here, the term $T_{x}$ represents the execution or waiting time for a task x assigned to a specific VM, X denotes the total number of tasks processed by a given VM, $\sum_{x = 1}^{X} T_{x}$ represents the total processing time or queuing delay for all tasks executed on a particular VM, and the max operator is used to identify the maximum accumulated time among all VMs.

Task completion time ( $T$ ): It is calculated by how much time is utilized for completing the task on the server after allocation.

Active server ( $A S$ ): The active server is used for processing the jobs or tasks over the cloud infrastructure.

6. Results analysis

6.1 Experimental setup

The proposed task scheduling was implemented in MATLAB 2020a, and the experimental analysis was carried out. Here, the performance of the proposed model was compared over the conventional models in terms of convergence analysis and statistical analysis, run time, average completion time, percentage of used resources, QoS and waiting time, makespan, latency and resource consumption, etc. Here, the proposed algorithm has considered the 10 total populations and a total iteration count of 100. The proposed work is compared against SSA,³³ JAYA,³⁴ WSA³² and FFA.³⁵

Generated Synthetic Data: A synthetic dataset was generated in this work to evaluate the robustness and effectiveness of the developed HFWSO algorithm. This dataset was carefully designed to mimic real-time cloud workload scenarios, as it encompasses variations in task arrivals, resource requirements, and execution times, thereby allowing for a systematic and controlled evaluation of the proposed scheduling model. Moreover, it replicates several cloud workload patterns, such as variable task arrival rates, heterogeneous resource demands, and fluctuations in task execution times. Using synthetic data enables a comprehensive experimental evaluation across a wide range of configurations and facilitates benchmarking the performance of the developed algorithm against related heuristic approaches under consistent testing conditions. The dataset includes details such as task limits, datacenter limits, number of datacenters, number of tasks, task weight, task size, task execution time, task waiting time, task energy consumption, datacenter size, datacenter busy time, and datacenter energy usage.

6.2 Validation metrics

Validation metrics used in the developed model are given below.

Average computation time: It refers to the time taken to complete all tasks assigned to the active servers. The average computation time is calculated and used to analyze the performance of the model.

Makespan: It is defined as the total elapsed time from the start of processing to completion. Specifically, it denotes the duration from the moment a job is allocated to a server until the corresponding machine completes the assigned tasks.

Resource consumption: It measures how effectively the scheduling algorithm utilizes computing resources, such as processors, memory, and network bandwidth, to complete jobs and tasks efficiently.

QoS: It represents the overall performance of a service, such as a cloud computing service, from the user's perspective. It includes parameters like availability, reliability, response time, and throughput.

6.3 Setting the configurations

The proposed model designs five different configurations, which comprise a greater number of tasks and more VMs. Table 2 demonstrates the configuration settings for the proposed job scheduling model. To perform the job scheduling, five configuration cases are defined. In the first case, 100 tasks and 10 VMs are used for allocation. In the second case, 200 tasks are allocated to 20 different VMs. In the third case, 30 VMs are used to allocate 300 jobs. In the fourth case, 400 tasks and 40 VMs are considered. Finally, 500 jobs and 50 VMs are used for job scheduling.

Table 2.
Configuration settings for proposed job scheduling.

Configuration case. No. of tasks No. of VMs

1 100 10

2 200 20

3 300 30

4 400 40

5 500 50

Configuration case.	No. of tasks	No. of VMs
1	100	10
2	200	20
3	300	30
4	400	40
5	500	50

Figure 6.

Convergence analysis of the suggested job scheduling process using big data in terms of different cases, as (a) configuration case 1, (b) configuration case 2, (c) configuration case 3, (d) configuration case 4 and (e) configuration case 5.

6.4 Convergence analysis on the recommended job scheduling model

Figure 6 depicts the convergence analysis of the proposed model compared with other existing algorithms across different iterations. Figure 6(b) shows the convergence performance when the cloud executes 200 tasks distributed among 20 virtual machines. The cost function values are 10.1% for SSA, 23.5% for JAYA, 6.7% for WSA, and 37% for FFA, all of which are higher than that of the proposed HFWSO algorithm. This demonstrates the improved convergence rate achieved by the novel heuristic model. As a result, the proposed method effectively allocates jobs in a green cloud with big data while reducing energy consumption in datacenters.

6.5 Active server analysis on the recommended job scheduling model

Figure 7 illustrates the number of active servers utilized for job scheduling. By reducing the number of active servers, the proposed model significantly minimizes energy consumption in datacenters. Consequently, the recommended model enhances overall system performance.

Figure 7.

Active server analysis of the proposed job scheduling process over other algorithms.

6.6 Comparison analysis on the recommended job scheduling system

Figure 8 illustrates the multi-objective analysis of the proposed job scheduling mechanism, while Table 3 presents the numerical evaluation of the developed HFWSO. With respect to different configuration cases, the objective constraints are analyzed and compared against classical heuristic algorithms. Figure 8(d) depicts the makespan analysis of the proposed scheduling mechanism. When the cloud network executes 300 tasks across 30 VMs, the proposed model reduces the makespan by 7.4% compared to SSA, 1.85% compared to JAYA, 11.1% compared to WSA, and 5.5% compared to FFA, respectively.

Figure 8.

Comparison analysis on proposed job scheduling algorithm using big data with green cloud over other existing optimization algorithms by (a) computation time, (b) resource consumption, (c) energy consumption, (d) makespan, (e) QoS and (f) queueing time.

Table 3.

Numerical evaluation of the developed model.

VMs	Makespan (sec)	Energy Consumption (kWh)	Resource Utilization (%)
20	148.30	6.72	76.50
40	189.40	9.84	81.20
60	245.60	13.25	83.40
80	302.10	16.91	85.70

6.7 Statistical analysis on the recommended job scheduling algorithm

Table 4 presents the statistical analysis of the proposed job scheduling method. This evaluation is performed using parameters such as best, worst, median, mean, and standard deviation. The mean represents the average of the best and worst values, while the median refers to the midpoint between them. The standard deviation indicates the degree of variation across multiple executions. Based on the results, the enhanced model demonstrates superior performance in improving scheduling efficiency.

Table 4.
Statistical evaluation of the proposed job scheduling system.

Metrics SSA³³ JAYA³⁴ WSA³² FFA³⁵ HFWSO

Configuration case 1

BEST 1582.6 1570 1582.9 1600.2 1569.5

WORST 1629.7 1629.7 1629.7 1629.7 1629.7

MEAN 1587.5 1581.4 1587.8 1611.7 1577.5

MEDIAN 1582.6 1571.7 1586.3 1606.3 1569.5

STANDARD DEVIATION 10.748 15.841 10.263 13.095 12.759

Configuration case 2

BEST 2995.4 3022.5 2980.4 3072.8 2950.9

WORST 3157.7 3182.8 3182.8 3182.8 3172.2

MEAN 3030.7 3028.4 2992.8 3106.6 3006.6

MEDIAN 3023.4 3022.5 2989.3 3075.6 2978.6

STANDARD DEVIATION 42.909 22.504 29.472 42.347 60.14

Configuration case 3

BEST 4272.5 4242 4338.9 4319.6 4235.7

WORST 4423.5 4427.7 4427.7 4427.7 4416.2

MEAN 4337.2 4269.8 4350.4 4364.9 4267.2

MEDIAN 4324.8 4242 4338.9 4325 4249.9

STANDARD DEVIATION 57.436 61.68 24.509 48.219 37.912

Configuration case 4

BEST 5867.2 5891.9 5844.4 5947.6 5773.2

WORST 5978.5 5978.5 5978.5 5978.5 5978.5

MEAN 5923.3 5900.4 5877.3 5958.1 5863.5

MEDIAN 5956 5901.9 5875.5 5947.6 5826.5

STANDARD DEVIATION 42.421 12.438 31.329 12.971 82.482

Configuration case 5

BEST 7510.7 7415.6 7598 7427.3 7305.6

WORST 7657.6 7631.2 7657.6 7657.6 7621.2

MEAN 7560.9 7526.6 7606.8 7528.4 7377.7

MEDIAN 7548.6 7498.4 7602.4 7499 7357.6

STANDARD DEVIATION 55.806 75.008 12.847 95.749 101.68

Metrics	SSA³³	JAYA³⁴	WSA³²	FFA³⁵	HFWSO
Configuration case 1
BEST	1582.6	1570	1582.9	1600.2	1569.5
WORST	1629.7	1629.7	1629.7	1629.7	1629.7
MEAN	1587.5	1581.4	1587.8	1611.7	1577.5
MEDIAN	1582.6	1571.7	1586.3	1606.3	1569.5
STANDARD DEVIATION	10.748	15.841	10.263	13.095	12.759
Configuration case 2
BEST	2995.4	3022.5	2980.4	3072.8	2950.9
WORST	3157.7	3182.8	3182.8	3182.8	3172.2
MEAN	3030.7	3028.4	2992.8	3106.6	3006.6
MEDIAN	3023.4	3022.5	2989.3	3075.6	2978.6
STANDARD DEVIATION	42.909	22.504	29.472	42.347	60.14
Configuration case 3
BEST	4272.5	4242	4338.9	4319.6	4235.7
WORST	4423.5	4427.7	4427.7	4427.7	4416.2
MEAN	4337.2	4269.8	4350.4	4364.9	4267.2
MEDIAN	4324.8	4242	4338.9	4325	4249.9
STANDARD DEVIATION	57.436	61.68	24.509	48.219	37.912
Configuration case 4
BEST	5867.2	5891.9	5844.4	5947.6	5773.2
WORST	5978.5	5978.5	5978.5	5978.5	5978.5
MEAN	5923.3	5900.4	5877.3	5958.1	5863.5
MEDIAN	5956	5901.9	5875.5	5947.6	5826.5
STANDARD DEVIATION	42.421	12.438	31.329	12.971	82.482
Configuration case 5
BEST	7510.7	7415.6	7598	7427.3	7305.6
WORST	7657.6	7631.2	7657.6	7657.6	7621.2
MEAN	7560.9	7526.6	7606.8	7528.4	7377.7
MEDIAN	7548.6	7498.4	7602.4	7499	7357.6
STANDARD DEVIATION	55.806	75.008	12.847	95.749	101.68

6.8 Comparison of the developed model with recent literature models

Table 5 presents a comprehensive comparison of the proposed HFWSO algorithm against recent state-of-the-art metaheuristic approaches. Here, the energy consumption of the developed HFWSO is 48 kWh. The results prove that the developed model outperforms existing models such as DNN-EVARO, BSLO, IAFSA, and TLCO. It also demonstrates superior capability in providing a faster convergence rate, improved stability, and a lower error rate. This superiority of the model stems from its hybridized nature. When using only the FFA, the model may produce suboptimal results and may be prone to overfitting. Thus, employing the hybridized form enables the algorithm to achieve optimal solutions by leveraging the strengths of WSA and FFA, effectively reducing overfitting. Moreover, HFWSO faces challenges when applied to large-scale problems, particularly in distributed contexts. However, by integrating the MapReduce framework into the developed HFWSO model, this issue is effectively resolved, as MapReduce acts as a programming model for processing large datasets using parallel and distributed algorithms, offering efficient solutions to hybrid optimization challenges. By utilizing the exploration capability of WSA and the exploitation efficiency of FFA, the developed HFWSO achieves rapid and accurate solutions, allowing it to escape local optima. Furthermore, HFWSO demonstrates remarkable scalability and robustness in handling large-scale job scheduling within green cloud environments. These comparative findings reinforce the suitability of HFWSO as a powerful metaheuristic for multi-objective optimization in energy-efficient cloud computing, making it a promising candidate for practical applications that require balanced trade-offs among makespan, energy consumption, latency, and resource utilization.

Table 5.
Comparison with recent metaheuristics.

Configuration DNN-EVARO²⁷ BSLO²⁸ IAFSA²⁹ TLCO³⁰ HFWSO

Energy Consumption(kWh)

1 50.498 49.624 49.082 49.734 48

2 101.8 98.232 101.88 99.217 97

3 149.93 150.46 150.33 149.1 146

4 201.39 201.4 199.85 200.73 198

5 250.62 252.83 250.73 251.16 249

Configuration	DNN-EVARO²⁷	BSLO²⁸	IAFSA²⁹	TLCO³⁰	HFWSO
Energy Consumption(kWh)
1	50.498	49.624	49.082	49.734	48
2	101.8	98.232	101.88	99.217	97
3	149.93	150.46	150.33	149.1	146
4	201.39	201.4	199.85	200.73	198
5	250.62	252.83	250.73	251.16	249

6.9 Discussion

This section provides an in-depth discussion of the limitations, future scope, and implementation details of the developed model.

Implementation Details: The simulation of the developed model was carried out using MATLAB 2020a on a Windows 11 operating system, which ensures stable and modern software execution. The system was equipped with an Intel Core i3 processor, a mid-range CPU capable of handling moderate computational workloads typical of scheduling simulations. Additionally, 8 GB of RAM was used to ensure smooth processing, while a 500 GB hard drive provided ample storage capacity, supporting extensive experimentation without storage constraints. This combination of hardware and software specifications reflects a practical, widely accessible, and cost-effective setup for implementing the proposed model under real-world conditions.

Limitations and Future Scope: Although the developed model demonstrates enhanced performance using a cost-effective implementation platform, it faces limitations when tested with synthetic data, as such data may not fully capture the unpredictability of real-world cloud environments. Furthermore, employing hybrid optimization for effectively distributing workloads across multiple machines remains a complex task, requiring careful consideration of communication overheads and data dependencies.

To address these issues, future research will incorporate benchmark datasets and real-world job traces for experimental evaluation, thereby improving the external validity of the study. This enhancement will also facilitate more comprehensive comparisons with other state-of-the-art scheduling algorithms using standardized devices, thus increasing the practical relevance and impact of the proposed work. Finally, the integration of a deep learning model with the hybrid optimization algorithm is envisioned to provide more accurate and optimal workload distribution in future implementations.

7. Conclusion

This study presented a novel job scheduling mechanism that utilized a hybridized heuristic approach for green cloud computing with big data. Initially, big data was distributed across the green cloud infrastructure, which required efficient management, a task that was accomplished through the MapReduce framework. Within the green cloud network, components such as the green broker, cloud offers, and datacenters played a crucial role in allocating jobs to servers or machines. To achieve this, a new hybrid optimization algorithm, termed HFWSO, was introduced by integrating the existing Water Strider Algorithm (WSA) and Firefly Algorithm (FFA). This hybrid algorithm was implemented to provide optimal scheduling results for varying numbers of tasks. The optimized values demonstrated how tasks were effectively selected and allocated to corresponding machines. Subsequently, the proposed algorithm addressed multiple objectives, including energy and resource consumption, latency, makespan, queueing time, and the number of active and inactive servers. Through optimization, a multi-objective function was formulated, which yielded superior results across all performance parameters. Finally, the proposed model was validated and evaluated under various constraints and was compared against classical models. When the cloud system employed 20 servers and 200 tasks, the energy consumption of the proposed HFWSO model was significantly lower by 10.5% compared to SSA, 21% compared to JAYA, 31.5% compared to WSA, and 47.3% compared to FFA. Thus, the results proved that the developed HFWSO model effectively reduced datacenter energy consumption by optimally allocating jobs to machines within a big data-assisted green cloud environment.

Footnotes

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

References

Yuan

Zhou

. Multiqueue scheduling of heterogeneous tasks with bounded response time in hybrid green IaaS clouds. IEEE Trans Ind Inf 2019; 15: 5404–5412.

Yuan

Zhou

. Profit-Sensitive spatial scheduling of multi-application tasks in distributed green clouds. IEEE Trans Autom Sci Eng 2020; 17: 1097–1106.

Wen

Garg

Aujla

, , et al. Running industrial workflow applications in a software-defined multicloud environment using green energy aware scheduling algorithm. IEEE Trans Ind Inf 2021; 17: 5645–5656.

Żotkiewicz

Guzek

Kliazovich

, et al. Minimum dependencies energy-efficient scheduling in data centers. IEEE Trans Parallel Distrib Syst 2016; 27: 3561–3574.

Kaur

Garg

Aujla

, et al. A multi-objective optimization scheme for job scheduling in sustainable cloud data centers. IEEE Transactions on Cloud Computing 2022; 10: 172–186.

Kumar

Aujla

Garg

, et al. Renewable energy-based multi-indexed job classification and container management scheme for sustainability of cloud data centers. IEEE Trans Ind Inf 2019; 15: 2947–2957.

Lin

Cui

Peng

, et al. A two-stage framework for the multi-user multi-data center job scheduling and resource allocation. IEEE Access 2020; 8: 197863–197874.

Dong

Guo

Zhou

, et al. Task-aware flow scheduling with heterogeneous utility characteristics for data center networks. Tsinghua Sci Technol 2019; 24: 400–411.

Xiong

Huang

, et al. A johnson's-rule-based genetic algorithm for two-stage-task scheduling problem in data-centers of cloud computing. IEEE Trans Cloud Comput 2019; 7: 597–610.

10.

Luo

. Time- and cost- efficient task scheduling across geo-distributed data centers. IEEE Trans Parallel Distrib Syst 2018; 29: 705–718.

11.

C-M

Chang

R-S

Chan

H-Y

. A green energy-efficient scheduling algorithm using the DVFS technique for cloud datacenters. Future Gener Comput Syst 2014; 37: 141–147.

12.

Chen

Liu

, et al. Scheduling jobs across geo-distributed datacenters with Max-min fairness. IEEE Trans Netw Sci Eng 2019; 6: 488–500.

13.

Kang

Pan

Liu

. An online algorithm for scheduling big data analysis jobs in cloud environments. Knowl Based Syst 2022; 245.

14.

Tang

, et al. Load balance based workflow job scheduling algorithm in distributed cloud. J Netw Comput Appl 2020; 152: 102518.

15.

Ajmal

Iqbal

Khan

, et al. Hybrid ant genetic algorithm for efficient task scheduling in cloud data centers. Comput Electr Eng 2021; 95: 107419.

16.

Mahendra

Subhash

. Shinde “Task scheduling and resource allocation in cloud computing using a heuristic approach,”. J Cloud Comput 2018; 7.

17.

Kumar

Kulshrestham

. Energy efficient task scheduling in cloud data center. Int J Distrib Cloud Comput 2018; 6: 12–18.

18.

You

Wang

, et al. Low complexity hierarchical scheduling for diverse datacenter jobs. IEEE Commun Lett 2019; 23: 48–51.

19.

Liu

Zhang

, et al. Joint jobs scheduling and lightpath provisioning in fog computing micro datacenter networks. J Opt Commun Netw 2018; 10: 152–163.

20.

Mao

Peng

, et al. A multi-resource task scheduling algorithm for energy-performance trade-offs in green clouds. Sustain Comput: Inf Syst 2018; 19: 233–241.

21.

Deng

Shen

, et al. Eco-Aware online power management and load scheduling for green cloud datacenters. IEEE Syst J 2016; 10: 78–87.

22.

Chen

Wang

Chen

, et al. Stochastic workload scheduling for uncoordinated datacenter clouds with multiple QoS constraints. IEEE Trans Cloud Comput 2020; 8: 1284–1295.

23.

Caviglione

Gaggero

Paolucci

, et al. Deep reinforcement learning for multi-objective placement of virtual machines in cloud datacenters. Soft COmput 2021; 25: 12569–12588.

24.

Yan

Huang

Gupta

, et al. Energy-aware systems for real-time job scheduling in cloud data centers: a deep reinforcement learning approach. Comput Electr Eng 2022; 99: 107688.

25.

Shen

Hao

, et al. Online delay-guaranteed workload scheduling to minimize power cost in cloud data centers using renewable energy. J Parallel Distrib Comput 2022; 159: 51–64.

26.

Ergin

Kayran

. Investigation of variable taper angle design in highly tapered laminates for delaying delamination. Compos Struct 2025; 357: 118918.

27.

Nguyen-Ngoc

Nguyen-Huu

De Roeck

, et al. Deep neural network and evolved optimization algorithm for damage assessment in a truss bridge. Mathematics 2024; 12: 2300.

28.

Bai

Nguyen-Xuan

Atroshchenko

, et al. Blood-sucking leech optimizer. Adv Eng Softw 2024; 195: 103696.

29.

Qin

Feng

Tang

, et al. Condition assessment of a concrete filled steel tube arch bridge using in-situ vibration measurements and an improved artificial fish swarm algorithm. Comput Struct 2024; 291: 107213.

30.

Minh

H-L

Cao

, et al. An integrated surrogate model-driven and improved termite life cycle optimizer for damage identification in dams. Mech Syst Signal Process 2024; 208: 110986.

31.

Maitrey

Jha

. Mapreduce: simplified data analysis of big data. Procedia Comput Sci 2015; 57: 563–571.

32.

Kaveh

Dadras Eslamlou

. Water strider algorithm: a new meta-heuristic and applications. Structures 2020; 25: 520–541.

33.

SeyedaliMirjalili

AHG

ZahraMirjalili

ShahrzadSaremi

, et al. Salp swarm algorithm: a bio-inspired optimizer for engineering design problems. Adv Eng Softw 2017; 114: 163–191.

34.

Venkata Rao

. Jaya: a simple and new optimization algorithm for solving constrained and unconstrained optimization problems. Int J Ind Eng Comput 2016; 7: 19–34.

35.

Gandomi

Yang

X-S

Talatahari

, et al. Firefly algorithm with chaos. Commun Nonlinear Sci Numer Simul 2013; 18: 89–98.

Multi-objective job scheduling in green cloud with big data: a hybrid heuristic approach for minimizing energy in datacenters

Abstract

Keywords

1 Introduction

2. Literature review

2.1 Existing works

2.2 Problem statement

3.1 System model of green cloud

4.1 MapReduce framework for handling big data

5.1 Proposed HFWSO-based scheduling

Mathematical model

6.1 Experimental setup

6.2 Validation metrics

6.3 Setting the configurations

Table 2. Configuration settings for proposed job scheduling. Configuration case. No. of tasks No. of VMs 1 100 10 2 200 20 3 300 30 4 400 40 5 500 50

6.5 Active server analysis on the recommended job scheduling model

Table 5. Comparison with recent metaheuristics. Configuration DNN-EVARO 27 BSLO 28 IAFSA 29 TLCO 30 HFWSO Energy Consumption(kWh) 1 50.498 49.624 49.082 49.734 48 2 101.8 98.232 101.88 99.217 97 3 149.93 150.46 150.33 149.1 146 4 201.39 201.4 199.85 200.73 198 5 250.62 252.83 250.73 251.16 249

7. Conclusion

Footnotes

Funding

Declaration of conflicting interests

References

Table 2.
Configuration settings for proposed job scheduling.

Configuration case. No. of tasks No. of VMs

1 100 10

2 200 20

3 300 30

4 400 40

5 500 50