IJCNC 01

A Multi-Objective Optimization Approach to Load Balancing and Task Scheduling in IoT-Fog-Cloud Networks

M.Parveen Taj1 and N.Muthumani2
1Research scholar, Department of Computer Science, PPG College of Arts and Science,
Coimbatore- 641035, Tamil Nadu, India
2Principal, PPG College of Arts and Science, Coimbatore- 641035, Tamil Nadu, India

ABSTRACT
Although distributed systems can exist across multiple data centers, they require fog and cloud computing paradigms for data management in an Internet of Things (IoT) era. The integrated IoT, fog, and cloud (IoT-fog-cloud) method enables the processing of large amounts of IoT data in real-time. Alternatively, a lack of Load Balancing (LB) and improper handling of network resources can reduce the Quality of Service (QoS) under such circumstances. In real-time applications, increasing traffic to fog nodes causes delays and increases energy consumption. This problem was resolved by an effective LB algorithm. However, good resource utilization can be achieved whenever an effective LB is incorporated with Task Scheduling (TS). Hence, this article proposes a Multi-Objective Weight Optimized Task Scheduling and Load Balancing (MOWOTSLB) for IoT-fog-cloud systems. This research aims to effectively schedule workloads in a balanced manner, which conserves energy, enhances QoS, and reduces task execution time. This scheme employs a Horse Herd Optimization Algorithm (HOA) for TS. This HOA optimizes the scheduling of users’ task requests to suitable computing resources according to the fitness function calculated using makespan, execution cost, and energy utilization. Though it improves resource utilization, some Physical Machines (PMs) are overloaded, and others are underloaded during uncertain fluctuations in workloads. This causes high energy and resource wastage in the data center. To solve this problem, the HOA is also adopted for Virtual Machine (VM) migration in this article. By assessing the fitness function of PMs such as load, migration cost, energy, and bandwidth use, the HOA chooses the finest VMs to drift to the appropriate PMs. This successfully strikes a balance between load distribution among PMs and energy usage. The simulation findings show that the MOWOTSLB outperforms current LBTS schemes in 0.95 throughput, 30 ms delay,100 ms response time, 175 kj energy consumption, 12.5 mb memory usage, and 900s lifetime.

KEYWORDS
IoT-fog-cloud systems, Load balancing, Task scheduling, Horse herd optimization & VM migration

1. INTRODUCTION
   

Since the emergence of IoT technology, several applications, like power grids, healthcare, automation, etc., have produced enormous amounts of data. This data from various systems needs to be effectively handled to satisfy QoS criteria such as low latency, high reliability, real-time responsiveness, and scalability [1-2]. Preserving the consistency of performance throughout the IoT infrastructure necessitates optimizing essential attributes such as energy usage, execution time, latency, execution cost, throughput, and resource usage [3]. Because it offers substantial memory and processing power to handle the enormous amount of data generated by IoT components, cloud computing has emerged as a key technology [4-5]. It is the perfect option for business applications and services due to its adaptability, scalability, affordability, and userfriendliness. Users can use any computing capacity, storage, or services they require over the Internet at no additional cost while leveraging all available resources. However, latency problems with the conventional centralized cloud computing architecture might make it difficult to carry out time-sensitive tasks effectively [6-7].

Fog computing can help resolve some of the challenges faced by IoT applications functioning over the cloud. It links cloud servers with edge devices through a virtualized platform to enhance management and reduce communication lag. Distance reduction improves analytical speed for data processing locally and transmitting it to the cloud [8-9]. It minimizes network latency compared to cloud usage; however, it has limited storage, network, and computation resources [10]. A sustainable model for IoT applications may be developed to utilize fog and cloud resources by combining fog and cloud computing models [11]. Efficient allocation of workloads in IoT-fog-cloud environments has gained increased importance due to the rising number of IoT applications [12]. The fog or cloud layer assigns IoT programs to available computing resources using a well-designed workload scheduling strategy [13]. Many factors concerning QoS, like latency, power usage, processing time, resource demand, and throughput, need to be considered when assigning tasks to resources in a heterogeneous system [14]. Balancing the workload distributed among resources is critical to the effectiveness of IoT applications. Such management can be accomplished by closely monitoring each resource.

Nonetheless, inefficient allocation of workloads and system resources can lead to higher energy consumption and a lowered QoS. Due to the traffic congestion on fog gateways, there are high delays in delivering real-time services that are difficult to tolerate [15]. A system needs to be able to effectively manage the challenges that result from increased load, power consumption, and latencies produced by heavy data transfer and handling workloads as the proliferation of IoT devices continues. Many researchers studied about various effective LB algorithms. However, good resource utilization can be achieved whenever an effective LB is combined with TS in IoTfog-cloud systems. Therefore, this study combines LB with the TS algorithm to effectively schedule workloads in a balanced manner, which saves energy, increases efficiency, and decreases the time needed to complete all tasks.

This manuscript proposes the MOWOTSLB scheme utilizing the HOA in IoT-fog-cloud computing. It includes an enhanced TS algorithm for allocating the task of the user to different computing resources. It can also reduce the cost and the execution time, as well as improve the resource utilization of the TS problem using the HOA. This HOA assists in introducing a multiobjective optimization for efficiently handling TS issues. The HOA optimizes the users’ TS requests to suitable computing resources according to the fitness function calculated using makespan, execution cost, and energy utilization. Though it improves resource utilization, some PMs are overloaded and others are underloaded during uncertain fluctuations in workloads. This causes high energy and resource depletion in the information center. As a result, the VM migration is applied with the TS to effectively balance energy utilization and load fairness among PMs in each fog node. The HOA can also be applied to choose the VMs to drift to the appropriate PMs. Optimizing the migration of VMs involves taking into account the fitness function of every possible PMs; this finds the optimum mapping relationship between the chosen VMs and the suitable PM. Accordingly, the contribution of this study is as follows:

 To adopt and implement a novel optimization scheme called HOA for TS to assign the user’s tasks to multiple computing resources.
 To select the optimal VMs to drift to the appropriate PMs using the HOA, resulting in reducing both energy and resource wastage. International Journal of Computer Networks & Communications (IJCNC) Vol 18, No 3, May 2026

 To reduce the cost and execution period while enhancing the resource utilization in IoT-fogcloud setups efficiently.
 To demonstrate the efficiency of MOWOTSLB against conventional algorithms in terms of different metrics such as make span, energy utilization, resource usage, etc.

The below portions are arranged as follows: Section 2 discusses relevant works. Section 3 describes the MOWOTSLB algorithm. Section 4 assesses the simulation results. Section 5 settles the research and recommends further upgrades.

2. LITERATURE SURVEY

2.1. Task Scheduling

Safi et al. [16] presented a Multi-objective Grey Wolf Optimizer (MGWO) model to minimize the delay and energy utilization in the fog broker during TS. However, the main disadvantage was that it had to account for resource heterogeneity, which might affect resource use and load imbalance. Ghafari & Mansouri [17] designed Enhanced African Vultures Optimization Algorithm-based TS (E-AVOA-TS) for fog-cloud systems. The tasks were initially prioritized to manage the sensitivity of task latencies by the best-worst technique according to the number of tasks, deadline, and file volume. However, load imbalance can impact the resource usage. Liu et al. [18] suggested an optimal scheduling of IoT requests in an IoT-fog-cloud setting utilizing a mixture of Aquila Optimizer and AVOA (AO-AVOA). However, system throughput was low due to the uneven use of resources among nodes.

A Dynamic Multi-Criteria Scheduling (DMCS) technique was created by Bhakhar and Chhillar [19] to improve TS in fog-cloud computing. It ensured that time-sensitive jobs were completed quickly on fog nodes and resource-based tasks were controlled by cloud information centres by dynamically allocating tasks based on parameters including computational difficulty, urgency, and task dimension. Nevertheless, it has problems with scalability and processing overhead in bigger networks. By considering the location of the data storage, Khezri et al. [20] created a Data-Locality-aware Job Scheduling for IoT Fog-cloud (DLJSF) systems. Only the makespan criterion was taken for scheduling, whereas other factors like energy, cost, etc., were required to schedule the activities effectively.

Salehnia et al. [21] developed the Multi-Objective Moth-Flame Optimization (MOMFO) algorithm for TS, which minimizes task request completion time and increases throughput in fogcloud-based IoT services. However, energy consumption has remained high. For the TS in heterogeneous cloud systems, Behera & Sobhanayak [22] created a hybrid Genetic Algorithm and GWO (GA-GWO). The crossover and mutation were used to improve local search and preserve wolf variety. By taking into account several goals at once, such as lowering costs, energy use, and making span, a new fitness function was established. However, the best solution was not always found, particularly when working with large and intricate networks.

To categorize requests and identify processing targeted levels in fog-cloud structures, Srichandan et al. [23] created the Adaptive Neuro-Fuzzy Inference System (ANFIS). At the target layer, the Chaotic Honey Badger (CHB) technique was used to schedule such requests. A chaotic mapping function was integrated with an Opposition-based Learning (OBL) strategy to enhance HBA’s convergence. Though it was difficult to predict the loads of each compute node to schedule requests on available nodes.

Mahapatra et al. [24] presented a Dynamic Energy-and-Latency-Aware Task (DELTa) scheduling for fog-cloud networks. A multi-level queue technique was applied to prioritize tasks and find the suitable node for offloading. Then, the DELTAmethod was used to effectively schedule tasks onto the chosen nodes for execution. However, it may struggle in large-scale, dynamic, and heterogeneous settings. Rateb et al. [25] investigated the TS issue of IoT devices in a cloud-fog setting. First, Aquila and Salp Swarm Algorithms (ASSA) was utilized to pick ideal VMs for executing workflows. Then, Reducing MakeSpan Time (RMST) method minimized the MST of task on chosen VMs. Moreover, utilizing VM inclusion and the Dynamic Voltage Frequency Scaling (DVFS) approach on result from RMST, the static and dynamic energy utilization were minimized. However, bandwidth and memory utilization were impacted due to imbalanced load conditions.

2.2. VM Migration

Kaur et al. [26] used the Smart Elastic Scheduling mechanism (SESA) to create a low energy VM distribution and relocation mechanism. They used cosine similarity and bandwidth usage to improve QoS performance. However, energy usage has remained high. It is also difficult to implement on large-scale networks. Singh & Singh [27] presented a metaheuristic VM relocation utilizing Re-initialization and Decomposition-based Whale Optimization Algorithm (RD-WOA), which integrates the WOA with NSGA-II to find Pareto-optimality solutions. They enhanced work distribution to the VM, reducing VM migrations, costs, and energy usage. However, they did not consider optimal scheduling of tasks, which may impact the makespan and resource usage.

Swarnakar et al. [28] developed a multi-agent-based VM migration for dynamic LB in cloud computing. However, it has high response time and processing time. Yao et al. [29] proposed a low-power LB technique using VM merging in cloud networks. To prevent needless VM migrations, a load state categorization method was first used for PMs with load abnormalities that took into account present and prospective loads. Then, a resource-weighted assortment technique for driftable VMs was created, which picks proper VMs to move using multidimensional resource use while reducing resource fragmentation. Furthermore, to arrange VMs in the best PMs, a resource fitness and load correlation approach was applied. However, the amount of VM migrations and energy usage were high.

Radi et al. [30] created a Modified Genetic-based VM Consolidation (MGVMC) technique for replacing VMs online while accounting for energy usage, SLA breaches, and the amount of VM migrations. They used the GA to transfer VMs to the appropriate PM in such a manner that the number of overutilized and underutilized PMs was kept to a minimum. However, they did not consider the influence on dependability and scalability while expanding the number of jobs. Wu et al. [31] developed Predicted Mixed Integer Linear Programming (MILP) Robust Solver (PMRS) using Γ-robustness theory for VM migration. Initially, VMs’ future actions were predicted, and the problem was formulated as a Γ-robust knapsack problem (Γ-RKP), which is solved using a new MILP technique. However, it is computationally demanding and lacks resilience.

Alsadie and Alsulami [32] introduced a Modified Feeding Birds Algorithm (ModAFBA) for VM association to increase resource distribution and effectiveness in cloud data centers. It implemented adaptive position updating rules and tactics to reduce VM migrations. However, it requires extra data, such as application requirements and workload patterns, to improve the decision-making process. Liu et al. [33] proposed a unique technique for LB and VM migration that uses modeled agents and IoT devices to maximize resource usage and task allocation. They presented a unique Optimized VM Migration Scheme (OVMMS) that migrates VMs using the Squirrel Search Algorithm (SSA) during migration and search. However, energy consumption was not taken into consideration, affecting network efficiency.

Siruvoru and Aparna [34] created a Harmonic Migration Algorithm (HMA) by combining the migration algorithm with harmonic analysis to migrate a VM from an overloaded to an underloaded PM and enable or disable the VM using switching techniques in cloud computing. The tasks were given to the relevant VM in a round-robin fashion, and the VM’s load was forecasted using the gated recurrent unit. However, it relocated the VM when the expected load exceeded the threshold, which is not suitable in real-time cloud systems.

Archana and Kumar [35] successfully introduced an altered BAT method for VM migration in a cloud setting. Spider monkey optimization’s fitness function calculation approach was incorporated into the regular bat algorithm to optimize every bat’s fitness computation and give further possibilities to select the best option rapidly. It allows VMs to join with the neighbouring PMs and finish the migration quickly, rather than being stalled. However, the network’s energy usage throughout the migration method was not taken into account, which influenced network efficiency.

From this literature, it is observed that most of the earlier TS and LB schemes consider single objectives, such as delay or energy, to enhance resource use. They have scalability issues for large-scale networks due to the lack of additional objectives like makespan, throughput, execution cost, etc. Moreover, earlier VM migration strategies often fail to integrate TS, leading to high energy consumption and unwanted VM migrations. As a result, this study proposes the MOWOTSLB scheme using the optimization technique for enhancing resource usage in IoT-fogcloud networks.

3. PROPOSED METHODOLOGY

In this section, the proposed MOWOTSLB scheme is described in detail. First, the system model, followed by the task and resource model. Then, problem formulation and HOA for TS and VM migration optimization problems are described. A graphic version of this manuscript is presented in Figure 1.

Figure 2. Three-Layer IoT-Fog-Cloud Infrastructure

3.1. System Model

Figure 2 shows the edge, fog, and cloud nodes of the IoT-Fog-Cloud network. All edges request several tasks. The traffic patterns include M/M/1, M/M/c, and M/M/∞ delays at each level. Wireless channels connect edge and fog nodes. The fog nodes can handle inquiries if the highest approval rate exceeds the inquiry rate. Additionally, fog nodes can send communications to the cloud for computation

3.2. Task and Resource Model

Assume a list of independent tasks T = {T1,T2, … , T𝑛 } created by the edge devices to be implemented. These are transmitted to the relevant resources to be handled in the fog-cloud computing scenario. Each task T𝑖 contains parameters such as task length (in million instructions), deadline, memory requirements, and amount of input and output files. Consider a group of computation nodes M = M𝑓 ∪ M𝑐 in fog-cloud network, in which M𝑓 = {M1,… , M𝑚𝑓 } denotes the group of fog nodes and M𝑐 = {M1, … , M𝑚𝑐 } denotes the set of cloud nodes. So, the sum amount of computing nodes is signified as 𝑚 = 𝑚𝑓 + 𝑚𝑐 . Every node M𝑗 includes features such as CPU processing rate (measured in millions of instructions per second (MIPS)), memory size, bandwidth, energy utilization, communication delay, resource usage cost.

3.3. Problem Formulation

HOA-based TS aims to distribute tasks among compute nodes as efficiently as possible to diminish makespan, execution cost, and energy utilization although meeting task deadline and resource constraints. The binary allocation matrix Q of dimension 𝑛 × 𝑚 is defined as follows:

Constraints

 Task allocation constraint: Every task should be allocated to precisely to one computation node as defined in Eq. (2).

3.3.1. Fitness Function for Task Scheduling

Makespan: For 𝑛 tasks and 𝑚 computing nodes, the Execution Time (ET) matrix with dimension 𝑛 × 𝑚 is used to determine the ET for task requests on nodes. The task scheduler uses the ET matrix to make TS choices on the fog-cloud platform. The ET of 𝑖 𝑡ℎ task on the 𝑗 𝑡ℎ computing node (𝐸𝑇𝑖𝑗) is determined as:

In Eq. (5), Task_lengthi denotes the length of i th task and CPUj is the handling ability of j th node in MIPS. The primary goal of the TS issue is to determine the optimal fog-cloud system schedule that minimizes the makespan, or task execution time. In this situation, no task will certainly take a considerable amount of time to perform and finish. So, the aim is to reduce the makespan (MK), which is represented in Eq. (6).

Total Execution Cost: The total execution cost (𝐶𝑡𝑜𝑡 ) comprises computational cost (𝐶𝑐𝑜𝑚𝑝), communication cost (𝐶𝑐𝑜𝑚𝑚), and deadline violation cost (𝐶𝑑𝑣), which is defined in Eq. (7).

In Eqns. (8)-(10), Ccpu,j , Cmem,j , and Cbw,j are cost per unit CPU, memory, and bandwidth utilizations on Mj , respectively, DataSizei is the Ti size in MB, and Penaltyi denotes penalty cost for delay.
Total Energy Utilization: The goal is to reduce the energy utilization 𝐸𝑡𝑜𝑡, which is represented in Eq. (11).

In Eq. (11), 𝛼𝑗 and 𝛽𝑗 are the energy utilization rates when node M𝑗 is active and idle, respectively.
The HOA is a metaheuristic algorithm that schedules tasks based on a fitness function to discover the most appropriate computer resources. In this study, the fitness function selects the solution depending upon the minimum 𝑀𝐾, 𝐶𝑡𝑜𝑡, and 𝐸𝑡𝑜𝑡. Thus, the fitness function for the above multiobjective TS issue (𝑓𝑇𝑆), which uses the weighted sum technique to balance the objectives, is defined as follows:

3.3.2. Fitness Function for VM Migration

In fog-cloud network, consider q number of PMs Q = {Q1, Q2, … , Qq} and v number of VMs are available in each PM, denoted as V = {V1, V2,… , Vv }. The edge user can schedule each task request to each VM using HOA, represented by Tv = {Tv1 , Tv2 , … , Tvn }, where vn denotes number of tasks scheduled to each VM. Besides, each VM has various parameters such as CPU usage, memory, bandwidth, MIPS, delay, and the number of processing elements. So, the v th VM in q th PM is defined as: