An effective multi-objective task scheduling and resource optimization in cloud environment using hybridized metaheuristic algorithm

Abstract

In today’s world, cloud computing plays a significant role in the development of an effective computing paradigm that adds more benefits to the modern Internet of Things (IoT) frameworks. However, cloud resources are considered to be dynamic and the demands necessitated for resource allocation for a certain task are different. These diverse factors may cause load and power imbalance which also affect the resource utilization and task scheduling in the cloud-based IoT environment. Recently, a bio-inspired algorithm can work effectually to solve task scheduling problems in the cloud-based IoT system. Therefore, this work focuses on efficient task scheduling and resource allocation through a novel Hybrid Bio-Inspired algorithm with the hybridized of Improvised Particle Swarm Optimization and Ant Colony Optimization. The vital objective of hybridizing these two approaches is to determine the nearest multiple sources to attain discrete and continuous solutions. Here, the task has been allocated to the virtual machine through a particle swarm and continuous resource management can be carried out by an ant colony. The performance of the proposed approach has been evaluated using the CloudSim simulator. The simulation results manifest that the proposed Hybridized algorithm efficiently scheduling the task in the cloud-based IoT environment with a lesser average response time of 2.18 sec and average waiting time of 3.6 sec as compared with existing state-of-the-art algorithms.

Keywords

Metaheuristic algorithm Internet of Things cloud computing resource optimization scheduling algorithms

1 Introduction

Cloud computing is an essential research area and has been extensively used in various manufacturing, telecommunications, scientific, and educational researches [1]. For instance, storage offers secured storage, recording, and data backup services that offer the finest convenience for various users [2]. Clouds used for educational purposes will virtualize diverse hardware resource types and broadcast them to the internet systems offering a proficient data platform for students, teachers, and departments [3]. In the cloud, resources like software and hardware are offered as services with a ‘pay-as-you-go’ basis. The individual users or the organization using it has to pay for the resources and services; whenever they need it without holding or purchasing hardware infrastructures.

The research that has been carried out recently is focused on resource management, virtualization, green computing, cloud security, task scheduling, and cloud-based IoT [4 –6]. As the services offered by the cloud shows rapid growth; how task scheduling towards computational resources (virtual machines) has risen to be the greatest question and it has to be solved. The ultimate objective of task scheduling significantly includes reduction of energy consumption, task completion, and enhancing load balancing and resource utilization ability in the cloud-based IoT environment.

With a tremendous rise in the total amount of users, diminishing task completion time is more effective for enhancing users’ experience. Effectual load balancing competency may provide complete utilization of Virtual Machines (VMs) and reduces execution efficiency that occurs due to resource overload and wastage due to huge idle resources [7]. Moreover, the two objectives are mutually constrained, i.e. to diminish task completion time and it is more proficient to allocate tasks more centrally over the resources with effectual computational power to load imbalance problems [8, 9]. Henceforth, it is extremely confronting to model and enhance the task scheduling process to fulfill two objectives like load balancing ability and reduction of completion time.

Based on this crisis, task scheduling is measured as an NP-hard crisis, and the optimal solution is not determined in reduced time. Consider a problem, if a solution has to be validated with polynomial time and time to resolve that problem is measured to grow higher when the problem size increases, the present computation model cannot be utilized for determining an appropriate solution in resourceful time. Here, the problem is determined to be NP-complete. Also, there are certain restricted cases like task scheduling with various processors and task scheduling unit time with arbitrary processors [10].

In recent times, meta-heuristic approaches and algorithms are used for resolving various mapping and scheduling crises in the cloud-based IoT environment. Particle Swarm Optimization (PSO) is a swarm-based intelligent approach validated to be more superior in resolving various optimization crises. This probability model determines solution distribution for search space and some novel solutions are produced by sampling it [11].

PSO shows faster convergence speed and provides the finest solution in a lesser time duration. Moreover, PSO can easily rely on a local optimum. Likewise, Ant Colony Optimization (ACO) is derived from the naturally inspired characteristics of an ant for searching food globally and this will provide the finest framework for resolving complex system-based optimization crises. The Pheromone and the searching nature and providing optimal solutions are expanded to certain searches for generating the solutions. This ACO is characterized by the finest global searching ability that may effectually deal with the deficiency encountered by PSO. However, its convergence speed is lesser [12].

In computer systems, there are two scheduling approaches like deterministic and exhaustive algorithms [13, 14]. Generally, deterministic algorithms are faster for scheduling issues in the cloud; however, it possesses some drawbacks. Thus, it is considered practical for both traditional and heuristic algorithms. Unlike exhaustive and deterministic algorithms, the meta-heuristic algorithm provides an optimal solution with reasonable time. It gives superior scheduling results than conventional methods [15]. It works effectually for various computing environments like cluster and grid computing. Motivated by this misbelief, this study introduces a systematic description of scheduling approaches in cloud-based IoT from the meta-heuristic perspective and bridges the link between the traditional and heuristic scheduling approaches. The target is to address the issue in conventional and heuristic algorithms with better task scheduling.

With this naturally inspired and resourceful application of PSO and ACO and by validating its advantages and disadvantages, a Hybridization of Improvised PSO with Improvised ACO (IPSO-IACO) has been proposed in this paper to offer effectual techniques for multi-objective task scheduling in the cloud-based IoT environment. The efficiency of the proposed Hybrid algorithm has been evaluated against other algorithms to prove its effectiveness in solving the workflow scheduling problem in the cloud environment. The results evident that the proposed algorithm reduces the total makespan execution time and balances the load over the VMs with a minimum total monetary cost.

The novel contributions are outlined as below:

A novel Hybrid IPSO-IACO model has been proposed to determine the requirements and task demands for VMs in the cloud-based IoT environment. This model considers execution time and scheduling performance as a constraint of scheduling crisis and attains multiple objectives for enhancing load balancing and diminishing task completion time.

An intelligent task scheduling mechanism can be introduced in the proposed model with the aid of a new IPSO algorithm. This intelligent scheduling facilitates to minimizing the makespan time, average response time, and average waiting time.

A new IACO algorithm has been established in the proposed model to efficiently allocate and manage the cloud resource. By using the IACO algorithm, the demanded resources and pooled resources are concurrently mapped to meet the optimal resource management.

The features of two meta-heuristic algorithms (IPSO and IACO) are associated to formulate a new Hybrid IPSO-IACO model. These features resolve a multi-objective task scheduling crisis appropriately in the cloud-based IoT environment.

The remainder of the work is formulated as trails. The background study of this work is explained more elaborately in Section 2. The mathematical and system model of the proposed Hybrid IPSO-IACO algorithm is demonstrated in Section 3. Section 4 illustrates the experimental outcomes and analysis. Section 5 provides the conclusion and directions for a future extension by addressing the limitations.

2 Literature review

Generally in a cloud-IoT environment, the primary role is task scheduling to determine the optimal mapping of relationships among virtual machines and tasks based on cloud systems and users’ goals. The significant methodology to determine this problem may include multi-objective optimization algorithms. Various investigators have concentrated on the modeling of effectual algorithms for resource management, scheduling, and load balancing; however, further enhancement is needed for handling NP-complete and NP-hard problems. The comparison of various existing task scheduling methods is depicted in Table 1.

Table 1
Comparison of various existing task scheduling methods

Reference Algorithm Target Benefits Drawbacks

[16] Hybrid Ant and GA Response time Better than conventional Ant and GA Resource allocation

[17] AEO Cost and computation time Reduces execution cost Non-optimal solution

[18] CR-PSO Cost, energy, and makespan 1–9 %energy improvement Higher waiting time

[19] Min-Min algorithm Execution time 13%better than GA Time consuming

[20] GA and min-min and max-mix approaches Execution time Faster than GA Stuck with local optimal solution

[21] PSO Transfer and execution time Faster than conventional L-PSO and M-PSO Stuck with local optimal solution

[22] PSO and GA Load balance, make span, communication cost, transfer, and execution time Faster convergence towards the solution Supports data center devoid of dynamic workflow

[23] Bin-packing heuristic and GA Execution cost and time Reduces execution cost compared to traditional algorithms Does not consider load over resources

[24] GA Makespan and load balance Superior than FIFO Time consuming to acquire optimal solution

[25] PSO-based heuristic model Cost and time Three times superior than existing approaches, superior load distribution over resources Time consuming process

Reference	Algorithm	Target	Benefits	Drawbacks
[16]	Hybrid Ant and GA	Response time	Better than conventional Ant and GA	Resource allocation
[17]	AEO	Cost and computation time	Reduces execution cost	Non-optimal solution
[18]	CR-PSO	Cost, energy, and makespan	1–9 %energy improvement	Higher waiting time
[19]	Min-Min algorithm	Execution time	13%better than GA	Time consuming
[20]	GA and min-min and max-mix approaches	Execution time	Faster than GA	Stuck with local optimal solution
[21]	PSO	Transfer and execution time	Faster than conventional L-PSO and M-PSO	Stuck with local optimal solution
[22]	PSO and GA	Load balance, make span, communication cost, transfer, and execution time	Faster convergence towards the solution	Supports data center devoid of dynamic workflow
[23]	Bin-packing heuristic and GA	Execution cost and time	Reduces execution cost compared to traditional algorithms	Does not consider load over resources
[24]	GA	Makespan and load balance	Superior than FIFO	Time consuming to acquire optimal solution
[25]	PSO-based heuristic model	Cost and time	Three times superior than existing approaches, superior load distribution over resources	Time consuming process

In [16], a new Hybrid ant genetic method has been introduced to address the task scheduling issues in the cloud-based IoT environment. The anticipated method embraces the benefits of both genetic and ant colony mechanisms and distributes tasks and VMs into smaller clusters. It efficiently decreases solution space by distributing tasks into clusters and by identifying loaded VMs. Owing to the minimum solution region of the anticipated method, computation and response time is suggestively reduced. However, it failed to allocate the resource to the VMs.

In [17], a modified artificial ecosystem-based optimization (AEO) has been presented for attaining an effective task scheduling mechanism. The anticipated multi-objective scheduling process deals with the minimization of the overall service cost and overall computational time. Nevertheless, this model does not concentrate on convergence rate and optimal solution.

Similar to the [16], a Hybrid Chemical Reaction Partial Swarm Optimization (CR-PSO) has been suggested to allocate multiple independent tasks in VMs [18]. Here, the tasks can be managed according to both the demand and deadline that prolongs the quality of service in terms of cost, throughput, and makespan. The higher waiting time is the major shortcomings of this hybrid method due to the involvement of a large number of parameters in the evaluation phase.

An effectual algorithm has been proposed for allocating a user’s task to VMs or nodes in a cloud environment [19]. This model attempts to improve overall cloud performance. The resource allocation crisis is described to reduce total energy cost by fulfilling certain client specification levels at the probabilistic level. It used the reverse model to cast penalties on clients who do not ensure SLA agreements. Therefore, the heuristic model is utilized to resolve the above-mentioned resource allocation crisis in the cloud-based IoT environment.

In [20], a load balancing model has been presented to fulfill the complete scheduling approach and diminishes the administrator’s role. The Genetic Algorithm (GA) is utilized in the anticipated model to achieve effective task scheduling. Nonetheless, it does not determine the nodes’ ability and configuration of the entire system to hold a backup; therefore that outcomes in a single failure point. A new ACO algorithm is introduced for task scheduling with effectual incoming tasks with the use of loaded cloud nodes [21]. After a certain iteration, the ACO algorithm fails to update node status for scheduling tasks. The author concentrates on determining to trace and seeking models and analyzed the factors that lead to delay in task execution.

A linear programming model has been proposed to address the scheduling crisis and used an adaptive scheduler where the container holds energy conservation and deals with workload transition costs from users to container hosts [22]. A multi-objective function using gaming theory is presented to diminish makespan and energy consumption by considering factors like CPU, memory, and cost-efficiency [23]. A novel cooperative gaming model has been introduced to anticipate the storage aware and communication aware algorithms that optimize user objectives [24]. It is determined by two factors known as storage and network bandwidth.

A multi-objective scheduling algorithm termed as a multi-opt has been established to handles the factors like CPU utilization, memory usage, load balancing, network overhead, and time consumption [25]. For all factors, they determined these metrics and provided a scoring functional objective to choose a more appropriate node for container deployment. Similarly, ACO gas certain characteristics like positive information feedback and heuristic search. It is needed for the global optimization process which has been extensively used in the cloud environment.

An Ant Lion optimization is modeled for scheduling tasks that consider the user’s budget and makespan as optimization objectives [26]. Two essential objective functions are used for computing the solutions and then the quality is determined by linearly adjusting the feedback. Likewise, a task scheduling algorithm is proposed to address the task issues in a cloud computing system [27]. This can be implemented using ABC, bat, and firefly algorithms. It attempts to reduce total task cost and makespan to make the system to balance the load. These are achieved by enhancing the pheromone’s initialization, pheromone update, and heuristic function.

In [28], An ACO-based resource balancing mechanism has been presented that shows better application functionality. QoS and service clustering are the strategies that are constructed by parameterization and selection constraints. The fitness function was provided and global or local updates are done accordingly. This method implements Parento optimal multi-objective set for the searching problem.

A new resource scheduling method based on QoS fulfillment is determined using an ACO and Firefly algorithms [29]. Here, the ACO algorithm is initially considered for evaluating convergence factor and functional quality to validate pheromone update and efficiency with feedback factors to enhance selection probability. Next, the efficiency of local search is enhanced by mutation and cross-over factor. At last, global search and local search are initiated by updating with every ACO algorithm.

A novel VM placement time scheduling is proposed through a discrete event-based model [30]. As well, there is a lack of determining the requirements of various applications during VM placement. The black-box model can be applied for recognizing multi-tier application ability in virtualized platforms. They used various techniques for analyzing the scheduler’s decision, deployment environment, and workload type to recognize various factors like VM co-location. Subsequently, when they consider the relationship between various workload categories based on a diverse virtual network based on CPU, several VM, memory size, and CPU usage by VM [31]. This may show an impact in determining the statistical requirements for providing an appropriate conclusion. Some issues have been identified in single-objective optimization procedures like Min-Min, sufferage, and Max-Min. The Min-Min algorithm is anticipated for performing tasks in lesser time. This model works finely when task length is considered to be drastically higher than applying the conventional Min-Min algorithm [32].

2.1 IPSO for task scheduling

The conventional PSO model is developed by Kennedy and Eberhart and it is extensively utilized optimization based on birds and animal characteristics [33]. While in IPSO, the particles over here to specify the velocity and position. The particles recognize the global best (gbest) and local best (lbest) values where these two parameters are determined by global and local best values. The flow diagram of the IPSO algorithm is represented in Fig. 1.

Fig. 1

Flow diagram of IPSO algorithm.

The IPSO can be exploited for VM scheduling against incoming tasks. Cloud receives a huge amount of tasks from various locations; assigning and executing of tasks is considered to be more challenging. Consider {x₁, x₂, … , x_n } tasks or request that has to be scheduled in VM {vm₁, vm₂, . . , vm_n } . Here, IPSO is appropriately deployed for task scheduling. The IPSO plays an essential role in allocating tasks to VM more effectually. This experimentation is performed over a private cloud environment that receives tasks and those tasks are allocated to VMs.

Algorithm 1 explains the detailed task scheduling strategy. In all iteration, the cluster has to identify the lesser loaded VM as lbest and smallest among them as gbest. The successive tasks are allocated to VM which is related to gbest. If gbest is the same in all successive iterations, the gbest is updated with subsequent lbest from the cluster. This is sustained till every task is being implemented. Time complexity is determined as O (nc) . As ^′c′ is constant and therefore determined as O (n).

Table 7

Algorithm 1: IPSO based Task Scheduling
1. Initialize {vm₁, . . , vm_n} , lbest = 0 ;
gbest = 0 ; clusters ={ c₁, . . , c_n };
2. for all incoming tasks {x₁, … , x_n }
3. Determine lesser loaded VM;
4. Allocate lbest;
5. end for
6. Allocate gbest with lesser lbest values;
7. Allocate successive tasks to VM based on
gbest values;
8. if (successiveallocation = gbestvalues))
9. Move to step 2;
10. else
11. Move to step 6;

2.2 Resource allocation with IACO

The traditional ACO optimization approach is utilized to model a heuristic for solving the optimization problem. ACO determines optimal path selection based on searching food characteristics over real ants. If it found the source, it moves back to the colony by leaving pheromone to identify the path of the food source. When another ant seeks its marked path, it becomes stronger and selects the path having stronger pheromone. Therefore, the shortest path has been determined.

ACO characteristics are determined by the previous solutions attained with random steps. Random probability is depicted by merging heuristic information of prior solutions and pheromone information. This improved model varies from the conventional ACO model based on the three factors: a) constructing state transition rule; b) global updation rule, and c) local pheromone updation. Here, the transition rule is to provide a solution, global updation is for determining the best solution and local pheromone updation is for constructing a complete solution.

Table 8

Algorithm 2: IACO-Based Resource Allocation

1. Begin

2. Initialize VM, resource pool;

3. for all incoming tasks {vm₁, . . , vm_n};

4. Resource demand → sum of the resource request;

5. Essential resource → demand based on request

6. for all VM

7. Total resources → sum of resource+

resource demand;

8. end for

9. if resource → pool –total resources

10. end if

11. else

12. Excess resources → rest of gbest;

13. end for

14. for all VM {vm₁, … , vm_n }

15. while (excess resources)

16. if (Excess = needed)

17. VM executes excess resource;

18. else

19. Resource pool → pool- an essential resource;

20. end if

21. end while

22. Res-pool → res-pool + essential resources;

23. end for

24. end

Algorithm 2: IACO-Based Resource Allocation
1. Begin
2. Initialize VM, resource pool;
3. for all incoming tasks {vm₁, . . , vm_n};
4. Resource demand → sum of the resource request;
5. Essential resource → demand based on request
6. for all VM
7. Total resources → sum of resource+
resource demand;
8. end for
9. if resource → pool –total resources
10. end if
11. else
12. Excess resources → rest of gbest;
13. end for
14. for all VM {vm₁, … , vm_n }
15. while (excess resources)
16. if (Excess = needed)
17. VM executes excess resource;
18. else
19. Resource pool → pool- an essential resource;
20. end if
21. end while
22. Res-pool → res-pool + essential resources;
23. end for
24. end

The drawbacks encountered in conventional ACO are addressed with IACO. Ant moves forward for searching for food and to reach the food source. When the source has been identified (resources match identified) is observed, then pheromone is placed with indication and this pheromone is sensed by other ants as a tracing factor. This tracing process is similar to that of IPSO, but successive food source prediction is performed here to make stronger pheromone.

The resource allocation process of IACO is illustrated in Algorithm 2. This IACO concentrates on transition rule, global and local solution updation. This stronger pheromone matches with available excessive resources values; however, it remains to be excessive and is not allocated. The stronger pheromone determined is based on various factors.

According to the aforesaid analysis, the conventional scheduling process depicts lesser extensibility and adaptability for cloud-based IoT environment. There will be an effectual reduction of task completion time and leads to loads imbalance. This overloaded computation paves a way to higher computational efficiency that does not meet out the QoS requirements. Moreover, the existing methods failed to concentrate on optimal task scheduling in cloud-based IoT environment. The inappropriate scheduling cause delay in task execution and increases the computational complexity. To overcome this, a new Multi-Objective Task Scheduling Mechanism has been introduced in this work. A detailed description of the proposed work will be demonstrated in the following sections.

3 Proposed hybrid bio-inspired algorithm

A Hybridized bio-inspired algorithm (Hybrid IPSO-IACO) has been proposed to attain the optimal task scheduling process for the cloud-based IoT system. Here, IPSO can be considered for task scheduling whereas IACO is applied for resource allocation and management. The flow diagram of the proposed Hybrid model is depicted in Fig. 2.

Fig. 2

Flow diagram of the proposed model for cloud-based IoT.

3.1 Task scheduling and resource allocation with hybrid IPSO-IACO

To overcome the limitations encountered in PSO and ACO, the hybridization of IPSO-IACO is performed. This new hybridization shows better efficiency based on resource allocation and total execution time. Further, the communication overhead encountered in VM and available resources is reduced. The comparison can be carried out with excess resources and from further resource requirements. In the worst-case scenario, it does not work well and humiliates resource allocation performance and it may respond after a huge delay. To get rid of this excessive resources are identified and considered. Algorithm 3 shows the performance of the hybridized model.

In scheduling, particles are specified as VMs; local best are under-loaded virtual machines from every group and global best specifies minimal values of every local best value. The algorithm works continuously to attain gbest and lbest values. In this modified approach, the gbest values may change continuously in all iterations in contrary to PSO. The particle velocity and position are updated based on Equations (1) & (2): $a (t + 1) = a (t) + v (t)$ (1)

$\begin{matrix} a (t + 1) = & v (t) + x_{1} r_{1} (lbest - a (t)) \\ + x_{2} r_{2} (gbest - a (t)) \end{matrix}$ (2) where, a (t) is current VM load (current particle position); x₁x₂ is acceleration coefficients; r₁r₂ is random numbers among (0,1); lbest is under-loaded VM and gbest is least values of all lbest values.

3.1.1 IACO-transition rule

For resource allocation, an ant determines the virtual services’ as the next service for allocating resources from the physical machine based on pseudo-random proportional rules as in Equation (3): $s = {\begin{matrix} \underset{u \in N^{temp}}{argmax} {{τ_{uj}]^{α} * {[η_{uj}]}^{β}) & q \leq q^{0} \\ p_{ij} & else \end{matrix}$ (3) where α, β ⩾ 0; this specifies the heuristic and pheromone information; ^′q′ is a random number that is distributed randomly with [0, 1] and q⁰ is a parameter that shows the association among exploration and exploitation. When q > q⁰; then it is termed as exploration; else it is termed as exploitation. N^temp is a set of eligible virtual services, N^temp is chosen for allocating resources from the available physical machine p_m.

Table 9

Algorithm 3: Hybrid IPSO-IACO for Task
Scheduling and Resource Allocation
1. Begin
2. Initialize VM, resource availability;
3. for all incoming tasks {vm₁, … , vm_n };
4. Resource demand → sum of request;
5. Essential resources → demand based on request;
6. for all VM
7. Sum of resources → sum of request+
demand-based o request;
8. end for
9. if Resource pool → pool –the sum of resource
10. end if
11. else
12. Excess resource 1 → first gbest;
13. Excess resource 2 → second gbest;
14. Excess resource 3 → best of the first
gbest and second gbest;
15. end for
16. for all VM {vm₁, . . , vm_n};
17. if (excess resource > demand resource)
18. VM considers step 13;
19. else
20. Resource pool → pool –demand resource;
21. while (Excess resources 3)
22. if (excess resource 3 > demand resource)
23. VM considers step 14;
24. else
25. Resource pool → pool+demand resource;
26. end while
27. end if
28. Resource pool → pool + remaining resources;
29. end for
30. end

This is determined by a set of tasks that are not allocated to resources and have not been eliminated due to resource constraints with all kinds of resources and physical machines. This is provided in Equation (4): ${Load}_{s} = \sum_{i \in j} (r_{ik}^{a} + q_{ij} * f_{ik}^{a}); \forall j, k$ (4)

The probability of selecting virtual tasks as the next available task for available resources in the physical machine is given as in Equation (5): $p_{ij} = \frac{{[τ_{ij}]}^{α} * {[η_{ij}]}^{β}}{\sum_{u ɛ N^{temp}} {[τ_{ij}]}^{α} * {[η_{ij}]}^{β}}; \forall i ɛ N^{temp}$ (5) where τ_ij is the virtual task that is marked as a chosen path for the next task by allocating resources from the virtual machine. With the ultimate objective of balancing the load in available resources, the heuristic information is determined as in Equation (6): $η_{ij}^{α} = \frac{1}{\sqrt{\frac{1}{M} \sum_{j = 1}^{M} {(R_{j} - \bar{R})}^{2}}}$ (6) where $η_{ij}^{α}$ is heuristic information for balancing loads.

3.1.2 IACO-pheromone rule updation

With ACO, the pheromone rule updation is performed with the global and local pheromone updation rule. When ant intents to provide a solution; the rule has to be updated as in Equation (7): $τ_{ij}^{t} = (1 - ρ^{local}) * τ_{ij}^{t - 1} + ρ^{local} * τ_{ij}^{0}$ (7)

where ρ^local ɛ [0, 1]; it specifies the pheromone evaporation rate and $τ_{ij}^{t}$ is pheromone at ^′t′ time and $τ_{ij}^{t - 1}$ is pheromone during previous time. For constructing a solution, a global updation rule has to be used. In the case of a multi-objective crisis, Pareto optimization is used. Hence, global pheromone updation is used as a solution for the present parent. This is determined in Equation (8): $τ_{ij}^{t} = (1 - ρ^{global}) * τ_{ij}^{t - 1} + \frac{ρ^{global} . {Num}_{ant}}{E (S^{best}) + σ (S^{best})}$ (8) where E (S^best) and σ (S^best) is an energy solution and SD based on the global optimal solution. ρ^global specifies evaporation time and Num_ant is the total amount of ant and $τ_{ij}^{t}$ is pheromone at ^′t′ time and $τ_{ij}^{t - 1}$ is pheromone at (t - 1) time.

The available excess resources are stored with an indication for successive searches. The successive ants have to wait for their match with resource demands from tasks. This considered is marked as stronger and it will be updated. Then, VM initiates task execution. The food source location is changed for every updation. This food source tracing is the same for all successive iteration. Some resources are considered to be excessive and considered for future resource demands. Therefore, the resource pool reduces the time for lending the resources. This enhances management and the dynamic allocation of resources.

The demanded resources and pooled resources are concurrently mapped to meet the resource constraints. While the heuristic information is used to guide local search, other strategies should be taken to keep the diversity of the firework swarm. Here, the hybridization is done by modifying the upper bound of excessive resources. Therefore, the physical machine may identify the available resources, and unfitted resources are moved to the resource pool. This hybridization method can outperform the individual optimization model. This hybridization is done with combined feature evaluation and works parallel. The matching resources are considered for future requirements and system performance is improved than individual models.

Figure 3 demonstrates the flow diagram of the proposed Hybrid IPSO-IACO algorithm. The proposed model meets the multi-objective problem by performing efficient task scheduling, VM placement, and QoS fulfillment. The Hybrid IPSO-IACO algorithm starts with generating a random population and defines a specific number of iterations as a parameter to the algorithm. The population symbolizes various solutions to the workflow tasks issue and every solution is a scattering of the whole workflow tasks over the available VMs.

Fig. 3

Flow diagram of proposed Hybrid IPSO-IACO algorithm.

The initialized population is allowed by the IACO algorithm with the first half of the mentioned iterations; that is, if the number of the iterations is (n), then the IACO algorithm will be repeated (n/2) times. The rationale behind using (n/2) iteration is to alleviate the complexity of the proposed hybrid method, as the performance of the IACO depends on the number of iterations. These parameter values can be adjusted after evaluating the algorithm’s performance on a few trial runs. Through experiments, the Hybrid IPSO-IACO algorithm’s performance was the best, when the defined number of iterations is divided equally between the IACO and IPSO algorithms.

The ACO-based algorithms provide better results than conventional algorithms when the number of iterations is large. However, increasing the number of iterations means that the ACO algorithm will consume more time to reach the optimal solution. Subsequently, PSO-based algorithms provide better results than the other algorithms and in less time. At the same time, the results may not be accurate due to the fast convergence of the PSO-based algorithms to the solution, which may cause being stuck in the local optimal solution. Therefore, the proposed Hybrid IPSO-IACO algorithm is distinguished by the characteristics of the ACO and the PSO algorithms.

The Hybrid IPSO-IACO algorithm is expected to work faster with different sizes of workflow applications compared to other algorithms with the same objectives. Moreover, the Hybrid ACO-PSO algorithm may not get trapped in the local optimal solution. Thus, this proposed model is considered to be more effective for resource allocation and load balancing in the cloud-based IoT environment.

4 Simulation results and discussions

The proposed Hybrid IPSO-IACO model has been assessed through the CloudSim simulator. The experimental setup is shown in Table 2 and it comprises 8 VMs, 512 MB RAM with 500 to 1000 MIPS, and 1000 Hz bandwidth, with one datacenter. Here, no dataset is used in this work. Instead, the simulation environment is set up with a minimum of 80 tasks, 8 VM, VM is Xen, and datacenter 1. Based on this setup, the task scheduling process is performed and examines the available resources. Some configuration setups can be customized and it has various task requests and load balance. Here, the task specifies executing a task that needs various resources in the cloud-based IoT environment. For allocating resources, tasks that arise from the client side with arrival time. Resource allocation is managed by a load balancer.

Table 2
Parameter setup

S.no Parameters Value

1. Total Number of Tasks 10–80

2. Total Number of VMs 8

3. Size of VM 10000

4. RAM 512(MB)

5. MIPS 500 to 1000

6. Bandwidth 1000

7. Per number 1

8. VM Xen

9. Datacenter 1

10. M 37 ants

11. Q 1 (multiplier for pheromone update)

12. VM 10 to 80

13. α 2 (pheromone exponent)

14. β 1 (exponent of computational capacity)

15. γ 0.05

S.no	Parameters	Value
1.	Total Number of Tasks	10–80
2.	Total Number of VMs	8
3.	Size of VM	10000
4.	RAM	512(MB)
5.	MIPS	500 to 1000
6.	Bandwidth	1000
7.	Per number	1
8.	VM	Xen
9.	Datacenter	1
10.	M	37 ants
11.	Q	1 (multiplier for pheromone update)
12.	VM	10 to 80
13.	α	2 (pheromone exponent)
14.	β	1 (exponent of computational capacity)
15.	γ	0.05

In this research work, parameter metrics like average response time, makespan, average resource utilization, throughput, and average waiting time are considered for computation. The performance of the proposed Hybrid IPSO-IACO model is compared with existing approaches like PSO and ACO respectively.

This section explains the proposed Hybrid IPSO-IACO for task scheduling to VMs. This model provides superior outcomes by validating the VM state and clusters during task allocation to VM. This may avoid queuing over the server. IACO has m = 37 (total ants used), Q = 1 (multiplier for pheromone update), VMs ranges from 10 to 80, ∝=2 (pheromones exponent), β = 1 (exponent of computational capacity); δ = 4 (exponent of load balancing factor); rho = 0.05 evaporation rate for stronger pheromone. The task can be allocated to VM with a lesser pheromone constant. The response time specifies the active state of the task until the completion of execution. The proposed Hybrid IPSO-IACO shows improved response time when compared to individual PSO and ACO.

The average response time is measured and this step is repeated by varying the number of tasks. However, IACO lacks in balancing the load more effectually over the VM; but it attains superior response time, the experimentation is repeated with various VM, tasks, where the results attained, are more consistent. VM utilization is done with various combinations of tasks like 10, 20, 30, 40, 50, 60, 70, and 80. The IACO works more efficiently in VM utilization and average response time. This algorithm uses two updations global and local updation for scheduling. Moreover, this model is not considered for scheduling. The hybrid model combines both IACO and IPSO for scheduling purposes.

Table 3 depicts the comparison of average response time for different models. The average response time of the proposed Hybrid IPSO-IACO model is 6.94 seconds which is 1.41 seconds and 1.97 seconds lesser than traditional PSO and ACO models. From these statistical results, the proposed Hybrid IPSO-IACO model evident that it consumes lesser time for allocating resources from the task in the queue. The traditional PSO and ACO consume more time for responding to the incoming tasks and make them wait for a long time over the queue. When there is no response from the host automatically the request declines and leads to end-to-end delay and latency. Thus, the proposed Hybrid IPSO-IACO model provides better outcomes as compared to the conventional standard approaches.

Table 3

Comparison of average response time (in seconds)

Number of Tasks	ACO	PSO	Hybrid IPSO-IACO
10	3.46	3.12	2.81
20	4.65	4.12	3.46
30	5.64	4.95	3.91
40	5.94	5.45	4.42
50	7.01	6.81	5.01
60	7.94	7.21	5.94
70	8.20	7.95	6.21
80	8.91	8.35	6.94

The makespan computation of the various models is illustrated in Fig. 4. Here, the x-axis plots the number of cloudlets and the y-axis plots the makespan in seconds. The makespan results manifest that the proposed Hybrid IPSO-IACO model outperforms well when compared with individual PSO and ACO models. VM facilitates resource demands by tasks that are managed by the proposed model. Incoming tasks may show interval time and demand for various cloud resources. This model has experimented with various tasks and the number of virtual machine. W_max=0.9;W_min = 0.1 and total iterations are 100 in IPSO. In the initial iteration, VM considers resources from resources and for all successive iterations. The anticipated IPSO-IACO model allocates resources from the resource pool or available resources (unused). In particular, the makespan of HPSO-ACO is 125 whereas the PSO and ACO attain 145 and 154 respectively.

Fig. 4

Makespan computation for various models.

The comparison of average resource utilization is shown in Table 4. The average resource utilization of the proposed Hybrid IPSO-IACO model is 0.496 which is nominal when compared to 0.401 of PSO and 0.392 of ACO. In IPSO, the best case is determined when the values are matched with future resource demands. In the average case, values may change when compared to individual PSO. In the best case, excess resources are considered from VMs buffer for fulfilling further resource demands. Thus, resources are matched with a hybridized model while compared to PSO and ACO. Further, the hybridized models consume lesser time for execution.

Table 4

Comparison of average resource utilization for different models

Number of Tasks	ACO	PSO	Hybrid IPSO-IACO
10	0.102	0.105	0.146
20	0.154	0.165	0.198
30	0.190	0.198	0.204
40	0.240	0.254	0.289
50	0.299	0.301	0.344
60	0.299	0.344	0.398
70	0.376	0.391	0.421
80	0.392	0.401	0.496

The resources are utilized either from unused resources from VM or resource pools. Based on this, the algorithm has been examined for resource utilization. IACO consumes an equivalent amount of resources from the cloud and idle resources. The anticipated IPSO-IACO model shows superior resource utilization when compared to PSO. Also, it is more effective in using unused resource pools indeed of taking resources from the cloud. The efficiency of the proposed model is higher in terms of resource utilization when compared to state-of-the-art models.

The throughput computation of the different models is depicted in Fig. 5. Here, a total of 80 tasks are considered with different iterations. The x-axis shows the number of cloudlets and the y-axis plots the throughput. The throughput of the proposed Hybrid IPSO-IACO model is 15.5%and 8.5%higher than PSO and ACO respectively. This results due to the implementation of an efficient task scheduling mechanism in the proposed model. When the task is managed by VMs, then excessive resources are stored over VMs individual buffer. Thus, these resources are used in future task demands in the proposed model. The VM then starts task execution by providing resources from the resource pool. In contrast, the existing models failed to schedule the task in an appropriate manner. It paves a way to attain a lower throughput than the proposed model.

Fig. 5

Throughput computation for different models.

Table 5 demonstrates the average waiting time of tasks for resource utilization. The waiting time is reduced as the resources are used from the resource pool or unused resources are allocated to balance the load. The average waiting time of the proposed Hybrid IPSO-IACO model is 10.95 seconds which is 7.97 seconds and 8.02 seconds lesser than PSO and ACO.

Table 5

Comparison of average waiting time (in seconds)

Number of Tasks	ACO	PSO	Hybrid IPSO-IACO
10	4.562	4.512	3.654
20	6.924	6.124	4.612
30	7.831	7.541	5.212
40	9.978	9.524	6.812
50	16.021	15.254	7.521
60	17.854	16.541	8.924
70	18.012	17.114	9.652
80	18.972	18.921	10.955

The proposed Hybrid IPSO-IACO model shows a lesser waiting time due to the hybridization of PSO and ACO algorithms. This hybridization improves the drawback associated with standard PSO and ACO. It also considers nominal resource utilization without affecting efficiency. On the contrary, the existing models do not balance the load which causes higher waiting time as compared with the proposed model.

The overall performance comparison of proposed Hybrid IPSO-IACO, PSO, and ACO models under 80 tasks consideration is depicted in Table 6. From Table 6, it is evident that the proposed Hybrid IPSO-IACO model outperforms well when compared with existing models like PSO and ACO. This is owing to the proposed model makes use of all available resources to eliminate the longer waiting/response time to achieve higher efficiency.

Table 6

Performance comparison of Proposed, PSO and ACO models

Metrics	ACO	PSO	Hybrid IPSO-IACO
Average Response Time (s)	8.91	8.35	6.94
Makespan	154	145	125
Average Resource Utilization	0.392	0.401	0.496
Throughput	138	131	146.5
Average Waiting Time (s)	18.972	18.921	10.955

The resources are attained from pools and extracted from unused resources to provide the nominal outcome. Finally, for measuring statistical analysis, two factors are considered. They are population and workload. The statistical analysis can be carried out with the proposed Hybrid IPSO-IACO model. The value is set as α = 0.05 which is a standardized cutoff value. The p-value of the proposed Hybrid IPSO-IACO model is lesser while comparing with ACO and PSO.

5 Conclusion and future work

The primary objective of the cloud-based IoT environment is to schedule the task that determines the optimal mapping of relationships among virtual machines and tasks based on cloud systems and users’ goals. This research work focused on the Meta-heuristic approaches includes PSO, ACO, and Hybrid IPSO-IACO algorithms for effectual resource management and scheduling in the cloud-based IoT environment. The proposed Hybrid IPSO-IACO model has been assessed through the CloudSim simulator.

The performance results manifest that the proposed model is considered to be more effective in task scheduling in contrary to other existing models. Subsequently, the proposed model is more operative in resource allocation to VM because of managing the multi-objective task scheduling strategy in the cloud-based IoT system. The average waiting time of the proposed Hybrid IPSO-IACO model is 10.95 seconds which is 7.97 seconds and 8.02 seconds lesser than PSO and ACO. Finally, the proposed Hybrid IPSO-IACO model diminishes average response time and enhances resource utilization to a slighter extent as compared with existing state-of-the-art algorithms.

The major research constraint is the lack of adopting the proposed model in the heterogeneous environment and the workflow distribution is not extended for real-time applications. In the future, the current work will extend to meet the dynamic scheduling where task allocation can be carried out in the heterogeneous environment at various arrival times and in a diverse simulation environment.

References

and Xiao

, Design of cloud computing task offloading algorithm based on dynamic multi-objective evolution, Future Generation Computer Systems 122 (2021), 144–148.

Timea

, Miodrag

and Nebojsa

Multi-objective task scheduling in cloud computing environment by hybridized bat algorithm, Journal of Intelligent & Fuzzy Systems, (2021), 1–13. DOI: 10.3233/JIFS-219200

Cai

, Xing

and Deng

, Research on digital urban architecture design based on cloud computing data center, Environmental Technology & Innovation 22 (2021), 101543:1–15.

Lavanya

, Prasanth

and Jayachitra

, A Tuned classification approach for efficient heterogeneous fault diagnosis in IoT-enabled WSN applications, Measurement 183 (2021), 109771:1–16.

Prasanth

and Jayachitra

, A Novel Multi-Objective Optimization Strategy for Enhancing Quality of Service in IoT enabled WSN Applications, , Peer-to-Peer Networking and Applications 13 (2020), 1905–1920.

Prasanth

and Pavalarajan

, Implementation of Efficient Intra- and Inter-Zone Routing for Extending Network Consistency in Wireless Sensor Networks, Journal of Circuits, Systems, and Computers 29 (2020), 1–22.

Blum

S.C.

, An ant colony optimization algorithm for continuous optimization: application to feed-forward neural network training Neural, Comput Appl 16(3) (2007), 235–247.

Jena

U.K.

, Das

P.K.

and Kabat

M.R.

Hybridization of metaheuristic algorithm for load balancing in cloud computing environment, Journal of King Saud University – Computer and Information Sciences, (2020), 1–14.

Bagherzadeh

, Shahinzadeh

and Shayeghi

Integration of Cloud Computing and IoT (CloudIoT) in Smart Grids: Benefits, Challenges, and Solutions, In IEEE International Conference on Computational Intelligence for Smart Power System and Sustainable Energy, (2020), 1–8.

10.

Martinez

, Cordón

and Herrera

, A taxonomy and an empirical analysis of multiple objective ant colony optimization algorithms for the bi-criteria tsp, European J Oper Res 180(1) (2007), 116–148.

11.

Prasanth

and Pavalarajan

, Particle Swarm Optimization Algorithm Based Zone Head Selection in Wireless Sensor Networks, International Journal of Scientific & Technology Research 8 (2019), 1594–1597.

12.

Kephart

K.J.

, Hanson

, Kadasamy

and Jiang

, Power and performance management of virtualized computing environments via lookahead control, Cluster Computing 12(1) (2009), 1–15.

13.

Jafari

and Rezvani

M.H.

Joint optimization of energy consumption and time delay in IoT-fog-cloud computing environments using NSGA-II Metaheuristic algorithm, Journal of Ambient Intelligence and Humanized Computing, (2021), 1–24. DOI: 10.1007/s12652-021-03388-2

14.

Moradi

, Shahinzadeh

and Nafisi

Attributes of Big Data Analytics for Data-Driven Decision Making in Cyber-Physical Power Systems, In IEEE 14th International Conference on Protection and Automation of Power Systems, (2020), 83–92.

15.

Davoody-Beni

, Sheini-Shahvand

and Shahinzadeh

Application of IoT in Smart Grid: Challenges and Solutions, In IEEE 5th Iranian Conference on Signal Processing and Intelligent Systems, (2019), 1–8.

16.

Ajmal

M.S.

, Iqbal

and Khan

F.Z.

, Hybrid ant genetic algorithm for efficient task scheduling in cloud data centers, , Computers and Electrical Engineering 95 (2021), 107419:1–16.

17.

Elaziz

, Abualigah

and Attiya

, Advanced optimization technique for scheduling IoT tasks in cloud-fog computing environments, , Future Generation Computer Systems 124 (2021), 142–154.

18.

Dubey

and Sharma

S.C.

, A novel multi-objective CR-PSO task scheduling algorithm with deadline constraint in cloud computing, Sustainable Computing: Informatics and Systems 32 (2021), 100605:1–15.

19.

Zhang

J.X.

, ZM

and Zheng

, Survey of research progress on cloud computing, Application Research of Computers 27(2) (2010), 429–433.

20.

Xiao

, Zhang

W.H.

and Wang

D.H.

, Overview of workflow technology in scientific process, Application Research of Computers 28(11) (2011), 4013–4019.

21.

Masdari

, ValiKardan

, Shahi

and Azar

S.I.

, Towards Workflow Scheduling In Cloud Computing: A Comprehensive Analysis, , Journal of Network Computer Applications 66 (2016), 64–82.

22.

Badawi

A.A.

and Shatnawi

, Static scheduling of directed acyclic data flow graphs onto multiprocessors using particle swarm optimization, Computers Operations Research 40(10) (2013), 2322–2328.

23.

Wei

, Wang

and Wang

, Multi-resource balance optimization for virtual machine placement in cloud data centers, Computers & Electrical Engineering 88 (2020), 106866:1–16.

24.

Verma

and Kaushal

, Budget constrained priority based genetic algorithm for workflow scheduling in cloud, International Conference on Advances in Recent Technologies in Communication and Computing 50 (2013), 216–222.

25.

and Li

L.Y.

, Hybrid cloud scheduling method for cloud bursting, Fund Inform 138(4) (2015), 435–455.

26.

Gokul

, Karthikeyan

P.R.

and Kumar

, Test scheduling of System-on-Chip using Dragonfly and Ant Lion optimization algorithms, Journal of Intelligent & Fuzzy Systems 40 (2021), 4905–4917.

27.

Gokul

, Sakthivel

and Karthikeyan

, Minimization of test time in system on chip using artificial intelligence-based test scheduling techniques, , Neural Computing and Applications 32 (2020), 5303–5312.

28.

Liu

, Li

and Yang

, Research on cost-optimal algorithm of multi-QoS constraints for task scheduling in hybrid-cloud, J Softw Eng 9(1) (2015), 33–49.

29.

Gokul

, Sakthivel

and Karthikeyan

, Test scheduling for system on chip using modified firefly and modified ABC algorithms, SN Applied Sciences 1 (2019), 1–15.

30.

Marosi

, Kecskemeti

, Kertesz

and Kacsuk

FCM: an architecture for integrating IaaS cloud systems, The 2nd International Conference on Cloud Computing, GRIDs, and Virtualization, (2011), 7–12.

31.

Singh

and Gupta

, Towards energy saving with smarter multi queue job scheduling algorithm in cloud computing, J Eng Appl Sci 12(10) (2017), 8944–8948.

32.

Pop

, Dobre

and Cristea

, Deadline scheduling for aperiodic tasks in inter-Cloud environments: a new approach to resource management, J Super Comput 71(5) (2015), 1754–1765.

33.

Zhang

, Yang

, Li

and Yu

, A Particle Swarm Optimization Clustering-Based Attribute Generalization Privacy Protection Scheme, Journal of Circuits, Systems and Computers 27 (2018), 1850179:1–16.