3.1 Introduction

Cloud computing is well known computing paradigm that consists pool of heterogeneous resources and provision to end users over the internet pay per uses basis. Cloud computing offers on-demand resource provisioning, multi-tenant environment for the services, elasticity and scalability at run time, measured the services types of significant characteristics. Cloud computing offers several users beneficiary services through a network where cloud users have the flexibility of accessing the cloud resources, pre-build applications, design own framework, etc. [1]. There are several enterprises that are deploying their services in cost effective manner using cloud platform. Nowadays, cloud computing is becoming an efficient paradigm that provides high-performance computing resources over the Internet to solve large-scale scientific and engineering problems. Several industries and business are shifting their platform from traditional approach to cloud, due to which end users and deployment of web applications are increasing exponentially day by day. On the contrary, expectation of end users (time and cost-effective service) and vendors (maximum profit by utilizing the cloud resources) is raised as cloud-based services become common. However, scheduling of cloudlet over the resources becomes the key challenging issue and it can degrade performance, if it failed to achieve the objectives of either vendors or end users. Cloudlet schedule is char-acterized by the presence of several cloudlets (i.e. the objects that refer to tasks) that have to be executed on a limited number of virtual machines. Therefore, management of cloud resources is critical especially when many cloudlets are submitted simultaneously to cloud computing [2].

Cloud Computing models utilize the virtualization technology in data center infrastructure for delivering the SaaS, PaaS, and IaaS services to the cloud users [3]. Virtualization allow the decouples of physical machine from the operating system and created an abstraction layer between the host machine and virtual machine to run multiple Operating System (OS) on a single hardware resource [4]. Hypervisor is solely responsible for the creation of abstraction layer which provide the flexibly to run multiple OS concurrently on a single physical machine for efficient utilization of resources. Virtualization technology also supports the live virtual machine migration, dynamic resource allocation, and workload isolation features. Cloud data centers have a large number of computational capabilities with a varying capacity of physical machines which are tightly coupled. There are several expenses like resource management, manpower for maintaining the servers, license for running the software, space, cooling system, power and energy consumption are required to establish a cloud data center [5].

Figure 3.1 shown that the managing and provisioning services of cloud data centers performed in two different levels: task scheduling level and VM application policies level. However, researchers often deal with the above levels separately by developing the task scheduling algorithm for the cost benefits of cloud users and designed VM allocation policies for the benefits for cloud services providers. The major challenge of cloud data center is the efficiently utilizing the infrastructure by cooperative between the user’s submitted tasks and generate an optimal allocation sequence of virtual machine to the physical host; therefore, the requirement of an efficient task scheduling algorithm along with an optima VM allocation policy.

Many researchers have developed different techniques to solve scheduling problems [3–29]. Some researchers reduce the makespan time of applications, but they ignore other important constraints like requirements of different cloudlets, availability of computing resources, and time complexity of scheduling algorithm. Whenever an algorithm focused at only one parameter, it takes less time to compute the applications. As the number of parameters is increased, the time of algorithms is also increase due to conflicting parameters. Hence, it is tough to find the optimal solution of conflicting parameters in cloud computing. Most of the developed traditional approaches failed to find the solution of contradictory objectives. For example, heuristic algorithms often have low time complexity, but they provide a large schedule length as they provide near-optimal solutions. On the other hand, meta-heuristic algorithms like Genetic Algorithm (GA) and Simulated Annealing (SA) take more time to schedule a set of cloudlets (i.e. tasks) but they provide a small schedule length. This paper presents a new Cloudlet Scheduling Approach (CSA) for the cloud computing environment. The CSA enhances the overall performance of cloud-let scheduling by taking into account both the cloudlets requirements and resources available, and improving schedule length, time complexity, and load balancing degree. The contributions of this research work are divided into following two parts:

• Build the several small size clusters by converting the large size scheduling problems into small size clusters.
• Developing a new efficient Cloudlet Scheduling Approach (CSA) for the cloudlets allocation.

The remainder part of the research work is organized in six sections. Literature survey is done in Section 3.2. Basic details about cloud computing and cloudlet scheduling problem is covered in the section 3.3. Section 3.4 illustrates the formulation of scheduling problem as an optimization problem. Section 3.5 presents some of the popular heuristic and meta-heuristic techniques to overcome the scheduling problem. The proposed CSA is developed in section 3.6. Section 3.7 presents the simulation results while section 8 presents the conclusion of this research work.

3.2 Literature Survey

There have been so many task scheduling algorithms presented in the literature to overcome the scheduling issues in the cloud environments. Several algorithms inspired with the metaheuristic based optimization technique to achieve the desired results in terms of various Quality of Services (QoS) parameters. Guo [6] developed a multi-objective task scheduling algorithm for efficiently and optimally placement of the cloud request to the available number of virtual machines. The developed techniques consider multiple parameters for evaluating the performance. The proposed algorithm is based on the fuzzy self-defense for designing the scheduling scheme. A simulation based experiment performed to validate the effectiveness of the proposed work and compared the results with the existing techniques. Alsaidy et al. [7] developed a scheduling algorithm which is inspired from the Partial Swarm Optimization (PSO) to tackle the issue of task scheduling in the cloud environment. The effectiveness of the proposed algorithm was checked by implemented it with the help of CloudSim in cloud environment. The simulation results shown some enhancements in the various parameters compared to the other standard task scheduling algorithms.

NoorianTalouki et al. [8] uses optimized cost table to produce the optimal results by placing the higher priority task first and lower priority task on the below order for the execution. The proposed approach follows the HEFT algorithm for reducing the makespan time of dependent tasks. To analyze the performance of proposed technique, a different scientific workflow dataset is used. A significant increment in the performance of the developed technique is notice while comparing other heuristic and metaheuristic based techniques. The performance of the proposed work is checked on the makespan, Schedule Length Ratio (SLR), efficiency, and speed up parameters. Oudaa et al. [9] designed a model which is based on the multi-agent system for finding the optimal schedule sequence to allocate the task to the available number of virtual machines. The proposed model uses the Deep Enforcement Learning (DER) for the generation optimal schedule sequence of tasks. A number of experiments performed which shows the effectiveness of the proposed model in the cloud environment. Yang et al. [10] focused on the realization of rapid reconfiguration of cloud resources to make the scheduling more flexible and enhancement in the utilization of these resources. A new Software Defined Cloud Manufacturing (SDCM) method for handing the task allocation problem in the cloud environment formalizes the time constraint data traffic problem and divided the tasks into the sub tasks and allocated them to generate an optimal salutation to overcome the data traffic issues. The proposed algorithm combined the two metaheuristic-based approaches for generating the safe schedule sequence. The experiment results demonstrate the usefulness of the algorithm in terms of preventing network congestion and successful to reduce the network latency.

Sanaj et al. [11] developed a nature multi objective task scheduling algorithm which use the features of chaotic squirrel search for resolving the task scheduling problem for the independent tasks in the cloud environment. The proposed technique synthesized with the messy local search to enhance the optimal searching ability. To check the performance of the proposed technique, a cloud simulator toolkit is used to implement the proposed technique along with the other standard task scheduling methods. The simulation results proof the effectiveness of the research work. Dubey et al. [12] designed a hybrid multi-objective task scheduling framework for the dependent task which combined the features of two metaheuristic- based optimization techniques named as CRO and PSO to overcome the scheduling issue in the cloud. The proposed framework also considers the security aspect during submission of task to the cloud. The proposed framework tested over the CloudSim with the synthetic data set and compared with the other existing technique. The comparison result proves that the proposed framework achieved the desired result.

3.3 Cloud Computing and Cloudlet Scheduling Problem

Cloud computing technology contains four main components namely, client, cloud information system, cloud datacenter broker, and virtual machines. Figure 3.2 depicted the major component bs of cloud computing model. Clients are the end users and submitted work in the form of tasks or cloudlet to the cloud datacenter broker for the processing the cloudlet based on the resource requirements. Cloud information system is a repository which stores all the critical information about the cloudlets like cloud-let identification numbers, length, number of resource requirement, and arrival time of cloudlets. This information is used by the datacenter broker for the selection of resources to the cloudlets. Scheduler is responsible for finding the schedule sequence for the submitted cloudlets. Scheduler is the backbone of datacenter broker and determines the order according to which each cloudlet is executed. Virtual machines are the main components in cloud computing environment that responsible for execution of cloudlet and return the results.

In cloud computing environment, several cloudlets arrive to the system at the same time. Each cloudlet needs to be assigned into a suitable VM to be executed in shortest time. However, because the number of available VMs is less than the number of submitted cloudlets, a scheduling algorithm is required to schedule the cloudlets onto the available VMs. This problem is called cloudlets scheduling problem. Briefly, given a set of n cloudlets to be executed on m of VMs in the cloud computing environment, the cloudlets require certain resources and have computational capacity requirements. On the other hand, VMs are also capacitated. Therefore, need to find the efficient cloudlet execution order so that the cloudlet completion time can be minimized whereas the necessary requirement of cloudlets is met and the virtual machines capacity are not compromised.

3.4 Problem Formulation

The cloudlet-scheduling problem may be formulated as an optimization problem to be solved by optimization approaches. Designing amathematical model to the cloudlet scheduling problem involves two steps: (i) formulate a cost function to represent the objective of the cloud-lets scheduling, (ii) formulate set of constraints in terms of the cloudlets requirements and the availability of the VMs resources.

To formulate the scheduling problem, let n be the number of cloudlets, m is number of VMs, e_iv be the execution time of cloudlet i on machine v, and x_iv is a binary variable such that x_iv is 1 if the cloudlet i is assigned to machine v and 0 otherwise as shown in eq. (3.1).

(3.1)

The scheduling problem may be formulated mathematically as:

(3.2)

Subject to

(3.3)

(3.4)

(3.5)

In this model, the objective of scheduling is formulated, in eq. (3.2), to minimize the execution time. Indeed, several constraints are modeled to meet the requirements of cloudlets and not violate the availability of virtual resources. The first constrain, eq. (3.3), guarantees that each cloudlet i is assigned to exactly one VMv. The second constraint, eq. (3.4), guarantees that the requirement of cloudlet_i doesn’t exceed the capacity of VM_v. The third constraint, eq. (3.5), guarantees that total memory requirements should not exceed the maximum available memory.

We can define an optimal solution S_opt as a best solution that achieves the lowest makespan. According to reference [3], page (106), the system can achieve the lowest makespan (i.e. optimal solution) if and only if the next conditions are met:

1. Each cloudlet is assigned to distinct number VM.
2. Each cloudlet starts execution as soon as possible.

By another way, we distributed cloudlets among the VMs to balance the workload. Thus, high load balancing is achieved, when optimal solution is founded. We can say that a solution with maximum balancing degree is the near optimal solution.

3.5 Cloudlet Scheduling Techniques

Recently, several cloudlet-scheduling techniques were developed [13]. They may be classified into static scheduling and dynamic scheduling. The distinction indicates the time at which the scheduling decisions are made. In dynamic scheduling, cloudlets are submitted at different time slots, and scheduled during run time (on the fly) based on the current state of VMs. Jing et al. [14] is an example of dynamic approaches. In this paper, authors developed a new hybrid meta-heuristic exploiting the features of simulated annealing and particle swarm optimization techniques for maximization of cloud service providers profit and minimizes the execution time. Yuan et al. [15] proposed an effectively dynamic task scheduling algorithm for dispatch all arriving tasks to private CDC and public clouds. The disadvantage of scheduling is the overhead incurred to determine the schedule while the program is running. This study concerned with static scheduling. In static scheduling, all cloudlets arrive simultaneously; the number of them is more than the number VMs and information regarding the cloudlets is known prior to execution. In this case, all cloudlets are scheduled first at compile time, before running process. Furthermore, it is easy to implement on the cloud environment while archiving the desired goal of cost saving, makespan time, energy consumption. We can classify a static scheduling into two classes: Heuristic methods and meta-heuristic methods.

3.6 Cloudlet Scheduling Approach (CSA)

3.7 Simulation Results

This section discuss the performance analysis of CSA, which is evaluated by considering schedule length, balancing degree, and running time of the CSA., In addition, the simulation results are compared with the most recent heuristic technique, conductance algorithm [26], and with the well-known meta-heuristic algorithms, MGA [32], and MSA [35]. The conductance algorithm is selected because it provides low time complexity. In addition, it finds a good solution than FCFS as the default scheduler of cloud computing model. While, the MGA and MSA algorithms are selected as the meta-heuristic scheduling algorithms due to (i) firstly, the MGA algorithm is the update of the most famous meta-heuristic GA technique. Secondly, it gives better solution than Max-min and Min-min algorithms. (ii) The MSA algorithm provides low time complexity and better solutions than greedy algorithm.

3.7.2 Evaluation Metrics

In this study, two evaluation parameters are used to compare the proposed CSA with the other algorithms, namely schedule length and balancing degree.

1. The schedule length (i.e. makespan) is the maximum execution time for all used virtual machines.
2. The Balancing Degree (BD) is the degree of load balancing between the VMs that achieves after scheduling. By using the optimal solution definition in section (3.4), we can use eq. (3.7) to calculate the balancing degree.
   (3.7)

From Eq. (3.7), we noted that the algorithm with high balancing degree is the near optimal solution.

3.7.2.1 Performance Evaluation with Small Number of Cloudlets

Table 3.2 shows the simulation results of scheduling 10, 20, 40, 60, 80, and 100 cloudlets into three VMs by the proposed CSA, MGA [32], MSA [35], and conductance algorithm [26]. From Table 3.2, we see that the proposed CSA provides the best results. It gives low schedule length at low running time. This is because CSA algorithm assigns each cluster into the VM that minimizes its execution time. Since CSA algorithm gives only one solution using clustering idea, it schedules large number of cloudlets in low running time. Unlike MGA and MSA, they find many solutions. And then they compare the solutions to find the best of them. This operation spends more time.

From the results SL of conductance algorithm, we can see that our algorithm has lower SL than the conductance. This is because the conductance algorithm assigns largest cloudlets in one VM. Thus, it doesn’t balance between the requirement of computing power and available VMs capacity.

3.7.2.2 Performance Evaluation with Large Number of Cloudlets

Tables 3.3, 3.4, 3.5, and 3.6 show the simulation results of scheduling largest number of cloudlets on different VMs. From these tables, we noted that the CSA algorithm minimizes the schedule length and running time. The algorithm distributes large number of cloudlets in small number of clusters. After that, it assigns the clusters instead of the cloudlets. This operation takes very small time. Thus, CSA algorithm has low running time. The CSA provides also the heights BD values. This means that it allocates the cloudlets into the available resources with high balancing degree. This balancing increases the system utilization. In addition, the makespan of CSA is less than that resulting from the other algorithms. This is because the CSA conceders a VM capacity. It calculates CML and CTL values to each cluster. This values balance VMs capacity with cluster requirements.

Tables 3.5, 3.6, and 3.7 indicate that the CSA is more efficient with large number of cloudlets. It can schedule 2000 cloudlets at only one second. This time is very small compared with the MGA and MSA algorithms. We see also that the running time with red color of MGA algorithm is more than schedule length of it. This means that MGA algorithm can’t use with large number of cloudlets, because it is not useful. As an example, in Table 3.5, the running time of MGA algorithm of 1000 cloudlets at 100 VMs equals 111. This value is larger than the schedule length (i.e. 106). Similarly, we see SL of MGA is larger than running time of all red values in Tables 3.6 and 3.7.

From all tables, the proposed CSA can achieve lower schedule length than the MGA, and SA. In addition, it finds a near optimal solution faster than both of them by 99.5% with MGA and 92.9% with MSA.

Although the conductance algorithm gives low time complexity as shown in all of tables, the CSA gives very low schedule length compared the conductance algorithm. It gives lower makespan than conductance scheduling algorithm by 41%. The proposed CSA not only minimize the schedule length, but also makes the system more balancing than any algorithm. In addition, the CSA appears like heuristic algorithm in result stability. It generates only one solution at all run times for specific problem, while the other meta-heuristic algorithms may give more than different solutions at each run for the same problem. In addition, it appears like meta-heuristic in minimizing schedule length.

3.8 Conclusion

In this research work, a new efficient Cloudlet Scheduling Approach has been proposed to solve the cloudlet scheduling problem in cloud computing. The main idea of the proposed approach is to minimize the overall time complexity by converting large size cloudlet scheduling problem into small size cluster scheduling problem and increases the system utilization. The proposed CSA enhances the overall system performance. It increases the balancing degree between the different VMs in the system. The new approach is more efficient than the most recent heuristic and meta- heuristic algorithms. The simulation results depicted that the CSA reduces the makespan than MGA and SA techniques and finds good solutions faster than them by 99.5% with MGA and 92.9% with SA. In addition, the CSA gives lower makespan than heuristic conductance scheduling algorithm by 41%.

References

1. 1. https://www.ibm.com/cloud‐computing/learn‐more/what‐is‐cloud‐computing/.
2. 2. https://en.wikipedia.org/wiki/Cloud_computing.
3. 3. Dubey, K., Nasr, A.A., Sharma, S.C., El-Bahnasawy, N., Attiya, G., El-Sayed, A., Efficient VM placement policy for data centre in cloud environment, in: Soft Computing: Theories and Applications, pp. 301–309, Springer, Singapore, 2020.
4. 4. Swathi, T., Srikanth, K., Raghunath Reddy, S., Virtualization in cloud computing. Int. J. Comput. Sci. Mobile Comput., 3, 5, 540–546, 2014.
5. 5. Dubey, K. and Sharma, S.C., An extended intelligent water drop approach for efficient VM allocation in secure cloud computing framework. J. King Saud Univ.-Comput. Inf. Sci., 2020.
6. 6. Guo, X., Multi-objective task scheduling optimization in cloud computing based on fuzzy self-defense algorithm. Alexandria Eng. J., 60, 6, 5603–5609, 2021.
7. 7. Alsaidy, S.A., Abbood, A.D., Sahib, M.A., Heuristic initialization of PSO task scheduling algorithm in cloud computing. J. King Saud Univ.-Comput. Inf. Sci., 2020.
8. 8. NoorianTalouki, R., Shirvani, M.H., Motameni, H., A heuristic-based task scheduling algorithm for scientific workflows in heterogeneous cloud computing platforms. J. King Saud Univ.-Comput. Inf. Sci., 2021.
9. 9. Oudaa, O., Gharsellaoui, H., Ahmed, S.B., An agent-based model for resource provisioning and task scheduling in cloud computing using DRL. Proc. Comput. Sci., 192, 3795–3804, 2021.
10. 10. Yang, C., Liao, F., Lan, S., Wang, L., Shen, W., Huang, G.Q., Flexible resource scheduling for software-defined cloud manufacturing with edge computing. Engineering, 2021.
11. 11. Sanaj, M.S. and Joe Prathap, P.M., Nature inspired chaotic squirrel search algorithm (CSSA) for multi objective task scheduling in an IAAS cloud computing atmosphere. Eng. Sci. Technol., Int. J., 23, 4, 891–902, 2020.
12. 12. Dubey, K. and Sharma, S.C., A novel multi-objective CR-PSO task scheduling algorithm with deadline constraint in cloud computing. Sustain. Comput.: Inform. Syst., 32, 100605, 2021.
13. 13. Mathew, T., Sekaran, K., Jose, J., Study and analysis of various task scheduling algorithms in the cloud computing environment. International Conference on Advances in Computing, Communications, and Informatics (ICACCI), pp. 658–664, 2014.
14. 14. Bi, J., Yuan, H., Tan, W., Zhou, M.C., Zhang, J., Li, J., Application- Aware dynamic fine-grained resource provisioning for virtualized cloud data centers. IEEE Trans. Autom. Sci. Eng. (TASE), 14, 2, 1172–1184, 2015.
15. 15. Yuan, H., Bi, J., Tan, W., Zhou, M.C., Li, B.H., Li, J., TTSA: An effective scheduling approach for delay bounded tasks in hybrid clouds. IEEE Trans. Cybern. (TCYB), 47, 11, 3658–3668, 2016.
16. 16. Bi, J., Yuan, H., Tan, W., Li, B.H., TRS: Temporal request scheduling with bounded delay assurance in a green cloud data center. Inf. Sci. (INS), 2016.
17. 17. Kimpan, W. and Kruekaew, B., Heuristic task scheduling with artificial bee colony algorithm for virtual machines. The 8^th International Conference on Soft Computing and Intelligent Systems (SCIS) and 17^th International Symposium on Advanced Intelligent Systems (ISIS), 2016.
18. 18. Nasr, A., EL-Bahnasawy, N., El-Sayed, A., Performance enhancement of scheduling algorithm in heterogeneous distributed computing systems. Int. J. Adv. Comput. Sci. Appl.(IJACSA), 6, 5, 88–96, 2015.
19. 19. Nasr, A., EL-Bahnasawy, N., El-Sayed, A., A new duplication task scheduling algorithm in heterogeneous distributed computing systems. Bull. Electr. Eng. Infor. (BEEI), 5, 3, 373–382, September 2016.
20. 20. Pavithra, B. and Ranjana, R., A comparative study on performance of energy efficient load balancing techniques in cloud. International Conference on Wireless Communications, Signal Processing and Networking (WiSPNET), pp. 1192–1196, 2016.
21. 21. Chen, H., Wang, F., Helian, N., Akanmu, G., User-priority guided min-min scheduling algorithm for load balancing in cloud computing. National Conference on Parallel Computing Technologies (PARCOMPTECH), pp. 1–8, 2013.
22. 22. Kokilavani, T. and George Amalarethinam, D.I., Load balanced min-min algorithm for static meta-task scheduling in grid computing. Int. J. Comput. Appl., 20, 2, 43–49, 2011.
23. 24. Etminani, K. and Naghibzadeh, M., A min-min max-min selective algorithm for grid task scheduling. IEEE/IFIP International Conference in Central Asia on Internet, pp. 1–7, 2007.
24. 25. Devipriya, S. and Ramesh, C., Improved max-min heuristic model for task scheduling in cloud. International Conference on Green Computing, Communication and Conservation of Energy (ICGCE), pp. 883–888, 2013.
25. 26. Chatterjee, T., Ojha, V.K., Adhikari, M., Banerjee, S., Biswas, U., Snáše, V., Design and implementation of an improved datacenter broker policy to improve the QoS of a cloud. Proceedings of the 5^th International Conference on Innovations in Bio-Inspired Computing and Applications IBICA, Springer International Publishing, pp. 281–290, 2014.
26. 27. Adil, S.H., Raza, K., Ahmed, U., Ali, S.S.A., Hashmani, M., Cloud task scheduling using nature inspired meta-heuristic algorithm. International Conference on Open Source Systems & Technologies (ICOSST), pp. 158–164, 2015.
27. 28. Tawfeek, M.A., El-Sisi, A., Keshk, A.E., Torkey, F.A., Cloud task scheduling based on ant colony optimization. 8^th International Conference on Computer Engineering & Systems (ICCES), pp. 64–69, 2013.
28. 29. Sindhu, S. and Mukherjee, S., A genetic algorithm based scheduler for cloud environment. 4th International Conference on Computer and Communication Technology (ICCCT), September 2013, IEEE, pp. 23–27.
29. 30. Omara, F.A. and Arafa, M.M., Genetic algorithms for task scheduling problem. J. Parallel Distrib. Comput., 70, 1, 13–22, 2010.
30. 31. Kar, I., Parida, R.N.R., Das, H., Energy aware scheduling using genetic algorithm in cloud data centers. International Conference on Electrical, Electronics, and Optimization Techniques (ICEEOT), IEEE, 2016.
31. 32. Singh, S. and Kalra, M., Scheduling of independent tasks in cloud computing using modified genetic algorithm. International Conference on Computational Intelligence and Communication Networks (CCIN), Nov. 2014, IEEE, pp. 565–569.
32. 33. Attiya, G. and Hamam, Y., Task allocation for maximizing reliability of distributed system: A simulated annealing approach. J. Parallel Distrib. Comput., 66, 1259–1266, 2006.
33. 34. Kashani, M. and Jahanshahi, M., Using simulated annealing for task scheduling in distributed systems. International Conference on Computational Intelligence, Modeling and Simulation, pp. 265–269, 2009.
34. 35. Liu, X. and Liu, J., A task scheduling on simulated annealing algorithm in cloud computing. Int. J. Hybrid Inf. Technol. (IJHIT), 9, 6, 403–412, 2016.
35. 36. Kahraman, C., Engin, O., Kaya, İ., Öztürk, R.E., Multiprocessor task scheduling in multistage hybrid flow-shops: A parallel greedy algorithm approach. Appl. Soft Comput., 10, 4, 1293–1300, September 2010.
36. 37. Awad, A., EL-Hefnawy, N., Abdel_Kader, H., Enhanced particle swarm opti-mization for task scheduling in cloud computing environment. International Conference on Communication, Management and Information Technology (ICCMIT), pp. 920–929, 2015.
37. 38. Tumeo, A., Pilato, C., Ferrandi, F., Sciuto, D., Lanzi, P., Ant colony optimization for mapping and scheduling in heterogeneous multiprocessor systems. International Conference on Embedded Computer Systems, Architectures, Modeling, and Simulation, SAMOS, pp. 142–149, 2008.
38. 39. Xu, J., Lam, A., Li, V., Chemical reaction optimization for the grid scheduling problem. IEEE International Conference on Communications, ICC, pp. 1–5, 2010.
39. 40. Calheiros, R.N., Ranjan, R., Beloglazov, A., De Rose, C.A., Buyya, R., CloudSim: A toolkit for modeling and simulation of cloud computing environments and evaluation of resource provisioning algorithms. Softw.- Pract. Exp., 41, 1, 23–50, 2011.