457
Pso-Based Task Scheduling Algorithm Using
Adaptive Load Balancing Approach For Cloud
Computing Environment
Md Oqail Ahmad and Rafiqul Zaman Khan
Abstract: Task scheduling is one of the key problem in the cloud computing environment. It is responsible for mapping the tasks to the appropriate resources, keeps the whole system balanced and optimizing the performance of the overall system. In this paper, we proposed a PSO-based task scheduling algorithm using adaptive load balancing approach where the tasks are expected to be heterogeneous. The proposed PSO-ALBA algorithm enhanced the performance of the standard PSO algorithm using adaptive load balancing approach. Adaptive load balancing approach handles overloaded and under-loaded condition simultaneously. Load balancing method guaranteed to balance the load by measuring the capacity on each virtual machine. Tasks are relocating according to the status of each virtual machine based on the deadlines of the tasks. The implementation results carried out using CloudSim simulator. The performance evaluation shows that the proposed PSO-ALBA algorithm optimizes the makespan and throughput compared with other heuristics algorithms such as PSO and ACO.
Index Terms: Cloud computing, Heuristic algorithm, Adaptive scheduling, Performance parameters
————————————————————
1.
INTRODUCTION
CLOUD computing has become inspiring technology due to its powerful architecture to perform large scale and complex computing. It simplifies appropriate on-demand network access to a shared unlimited number of computing resources such as CPUs, memory and storage [1] [2]. These resources are quickly provisioned and released with the least management effort. The most critical services offer by cloud computing are reliability, scalability, elasticity and high availability, which make it a highly complex and large distributed system. To provide these facilities for processing of massive amount of data, cloud computing services are categorized as infrastructure, platform and software as service [3] [5-7]. These services are made available to customers based on the pay as use model [8-10]. As masses of users post their computing tasks on the cloud system, task scheduling methods plays a essential part in cloud computing environments [1][4][11]. The scheduling mechanism can perform dynamically provisioning, allocation and balancing of users demand to available resources [3] [12] [13]. Therefore, the scheduling of tasks deals to reduce makespan and maximize resource utilization [12]. In this paper, we emphasis on minimizing the makespan and maximizing the throughput. We achieve this by using a meta-heuristics concept called Particle Swarm Optimization (PSO) and Adaptive Load Balancing (ALB) approach [14]. The benefit of using an adaptive scheduling algorithm with the PSO technique is, it handles over-loaded and under-loaded condition concurrently. Load balancing method guarantees load balancing of the system by measuring the load on each virtual machine and then relocating the task according to the status of each virtual machine based on the deadlines of the tasks.
Our contribution of the proposed methodology is as follows:
We proposed a PSO-based task scheduling algorithm using adaptive load balancing approach for cloud computing to minimizing the makespan and maximizing the throughput.
The adaptive load balancing concept used with the PSO algorithm, i.e. the results generated from the proposed algorithm are more optimized and improved.
The results of performance evaluation shows that the proposed PSO-ALBA algorithm could outperform in term of makespan time and throughput compared with the existing heuristics algorithms such as PSO and ACO.
The rest of this paper is organized as follows: section 2 represents the literature review of existing task scheduling algorithm related to our research work. The description of the PSO standard and its working methodology with system architecture, problem formulation with parameters estimation and also explain the proposed PSO-ALBA algorithm in Section 3.Section 4 demonstrates the simulation results of PSO-ALBA and compares it with other heuristics algorithms. Finally, Section 5presents the conclusions and future direction.
2
RELATED
WORK
Heba Saleh et al. [1] have introduced the IPSO algorithm to afford the ideal allocation for a huge number of data in cloud computing environment. Allocation of task can be achieved by dividing the submitted tasks into bunches in a dynamic way. Each advent batches must consider resource utilization state. After receiving the most desirable solution of every batch, the algorithm attaches all most desirable solution for batches right into a very last allocation map. The main intent of this proposed algorithm is to balance the load over the last allocation map. The IPSO algorithm is compared with existing scheduling algorithms such as honey LBA-HB, ACO and RR algorithms. The proposed IPSO outperform in term the degree of imbalance, makespan as well as standard deviation of load. Mohit Kumar et al. [2] have introduced novel resource allocation framework that has capability to decide at run time to process the application at VMs using PSO algorithm with deadline constraints. The proposed scheduling algorithm name as PSO-BOOST that improved various performance ————————————————
MdOqail Ahmad is currently pursuing Ph.D. degree program in Department of Computer Science, Aligarh Muslim University, Aligarh, India, PH-08439243408. E-mail: oqail.jmu@gmail.com
458
metrics such as processing time, processing cost, task acceptance ratio and throughput by submitting the heterogeneous tasks. The results of implement were conducted using the CloudSim tool and show that proposed PSO-BOOST algorithm outperforms compared with existing PSO, adaptive PSO, ABC and improved min-min load balancing algorithm. Fei Luo et al. [4] suggested an improved PSO strategy based on adaptive loads, in which adaptive loads use to make the load change with the increase of the number of iterations as well as also introduces random weights in the later stage. To avoid condition when the particle swarm technique could be stuck in the native optimal that originates to the advanced stage. The proposed strategy applied to task scheduling to achieve better scheduling plan in cloud computing. The results of experiments conducted using CloudSim simulator show that the overall performance of the proposed algorithm better than standard PSO. Thanaa S. Alnusairi et al. [15] have proposed a bio-inspired the scheduling algorithm for load balancing names as Bin-LB-PSOGSA. This algorithm enables scheduling method to optimize the stability load level on VMs. This algorithm mapped the task to VMs according to the length of submitted load and processing speed of VM. The implementation outcomes show that the Bin-LB-PSOGSA increase load over time and decrease processing speed of VM, which shows that proposed algorithm, performs more efficiently in keeping the load balanced over time. Fatemeh Ebadifard et al. [16] introduced a method based on the PSO where tasks are independent and non-preemptive. The proposed method enhanced the efficiency of the PSO standard method using the concept of load balancing technique. The conducted results indicate that the proposed algorithm increases the makespan by 33% and decrease the resource utilization by 22% compared PSO standard method. Gabi Hua et al. [17] have proposed adaptive multi-objectives task scheduling algorithm based on PSO techniques to optimize the processing time and the transmission time. The proposed PSO-based AMTS outperform quasi-optimal solutions with compared to the genetic algorithm. Rajni Aron et al. [18] have introduced hyper-heuristics resource scheduling strategy on the ground of PSO techniques to secure scheduling jobs on suitable resources without disturbing any of the security standards. The simulations were conducted using the GridSim Toolkit. The performance of proposed strategy outperforms than the existing algorithms in term of makespan as well as the cost of the user’s application. In this paper, we proposed PSO-based task scheduling algorithm using adaptive scheduling approach to solve the limitation of existing discussed algorithms.
3 PROBLEM
DEFINITION
WITH
SOLUTION
FRAMEWORK
3.1 PSO Standard
PSO was proposed by Eberhart and Kennedy [19] in 1995 is a population based optimization method that finds a solution to a problem in a search by modeling and predicting the insect social behaviour in presence of goals. The term ―particle‖ is used to present as bees, birds or any other specific who parades social behaviour as a collection and cooperate with each other. Each particle hovers in the problem hunt space is observing for a possible solution. Each particle adjusts its best position based upon its experience and the experience of its neighbour. The particle is placed with respect to other
solutions in search space is known as particle position and to know the particle direction and movement to optimized its fitness is known as velocity. The particle fitness represents how close a particle is to optimum solution compared with other particles in the search space [20]. Each particle velocity is updated in each time to find out the best two position global-best Gbest and personal-best Pbest. Pbest presents the personal
best position of particle has visited and Gbest present the global best position of the particle and its neighbours have visited since starting the iteration [21]. The updated PSO equation for position and velocity are given in (1) and (2),
(1)
(2)
Where, present the current position of particle j, j present the number of particles and t present the current time. 𝑡�+1 present previous time of particle, presents new velocity of particle j, 𝜔�presents the inertia weight that is used to maintain the balance between exploration and exploitation. and are acceleration constant and , are a random number in the range between 0 and 1.
3.2 PSO working System
PSO working system is divided into four processes: initialization, updating of particle position and velocity, calculate fitness value and optimal solution as shown in Fig. (1) [21] [22]. Before initialization, particles must be designed to represent all constraints that will be used in the calculation of fitness value and optimal solution. After that random initialization of position and velocity particle. Particle position and velocity has to be updated the fitness value and comparing with others in their neighborhood. From these updated solutions, tasks are scheduled according to PSO algorithm and calculate fitness value to schedule the number of upcoming task and available resources. The fourth process
TABLE1
459
Fig. 1. Overview of PSO working System
is to selection of the optimal solution of the problem according to PSO algorithm. Tasks are heterogeneous in nature. PSO algorithm aim is to find the Gbest optimal solution, but each particle has contained with Pbest, So that the VNS approach is
applied to find the optimal solution. VNS process goes back to the second process, and this process repeat until the stopping criteria not met.
3.3 Problem Formulation
Scheduling of resources is essential in cloud computing. The main intent of scheduling the tasks to obtainable resources ground on their features. It means that a set of tasks, T = {T1, T2…Tn} has to be assigned of obtainable resources V = {VM1, VM2...VMn}, where n is the length of the sequence of tasks and m is the length of the sequence of VMs. On this observe of literature, PSO methodology is best for optimization of task scheduling as compare to different scheduling algorithms. The objective of our proposed scheduling algorithm is to optimize the makespan and throughput, according to equations (6) and (7).
: The capacity of VM can be calculated by using Equation (3);
(3)
where P is the processing speed of VM and Q is the number of CPUs.
: The particle's fitness value is measured with the help of equation (4);
(4) where is the length of the task.
: The execution time of a task can be calculated by using Equation (5):
= / (5)
Mp: Makespan specifies the total finishing time of every task and can be computed by equation (6):
(6)
where FT specifies the finishing time of task i.
Tp: Throughput means some tasks completed in a certain time period and measured with the help of equation (7).
Tp = ( ) (7)
where represent execution time of the task Ti.
3.4 Proposed Methodology
The main intent of our proposed methodology is to improve the scheduling results in terms of makespan time and throughput. The proposed algorithm starts to the initialized value of a particle randomly. The proposed PSO-ALBA algorithm, schedule the number of upcoming task corresponding to number available cloud resources based upon the fitness value. After that it calculates particles fitness value of each particle. The algorithm keeps updated Pbest and
Gbest during its execution. It compares the current fitness value
of each particle with its own Pbest value; the best position (Pbest)
particle will be distinguished. Then, it compare the current fitness value of the whole population together, the lowest fitness value will be specified as global-best (Gbest) position.
Resource manager monitor the status of VMs continuously and forward the information to task scheduler that decide task would be allocated to VM or not. If the utilization time of task Ti is less than the defined threshold value, then assign the task Ti into EDF list and schedule task Ti according to traditional EDF. When the utilization time of the task Ti is more than defined threshold value then assign the task Ti into AEDF list and schedule task Ti based on limitation of compatible tasks.
TABLE2 CLOUDSIM CONFIGURATION
460
The proposed algorithm aim is to find out optimal Gbest based
upon the defined fitness function. Compare the fitness value of Pbest with Gbest. If fitness value of Pbest is better than Gbest then
assigned it to Gbest. This procedure is continuing until all tasks
have not been scheduled to available VMs according to adaptive load balancing approach. PSO-ALBA scheduling algorithm optimizes the performance parameters like makespan time and throughput while considering deadline as a constraint. The complete steps of PSO-ALBA are expressed in Algorithm 1.
4 SIMULATION
IMPLEMENTATION
AND
RESULTS
In this section, simulation is conducted to evaluate the efficiency of the proposed scheduling approach to the optimized solution. The results of the proposed PSO-ALBA algorithm is compared with other heuristics algorithms such as PSO and ACO from performance parameters like makespan
and throughput. Simulation outcomes show that our proposed algorithm outperforms than other heuristics algorithm. Simulations were performed by using CloudSim [23] to estimate the efficiency of the proposed algorithm by changing the numbers of tasks while keeping a fixed VMs. Further, we have conducted results at changing number of tasks and VMs. The CloudSim simulation setup is shown in Table 2.
Comparison of makespan
The makespan of proposed PSO-RALB algorithm compared by PSO standard and ACO. This test has been implemented more than 50 times using timeshare policy at the independent nature of task, and the result has been presented. The makespan has been compared by varying the numbers of tasks 100 to 1000 while keeping a fixed number of VMs 50. Further, the results have conducted at varying number of tasks from 100 to 1000 and a varying number of VMs from 40 to 140. The experiment result has been shown in Table 3 and 4 and depicted in Fig. 3 and 4.
The makespan produced by PSO-ALBA algorithm is improved compared with makespan produced by PSO standard and ACO.
Comparison of Throughput
The comparison of throughput of proposed PSO-RALB, PSO standard and PSO is illustrated in Table 5 and 6 and depicted in Fig. 5 and 6. The performance parameter is computed for analyzing the maximum throughput. Table 5 and Fig. 5 shows that the performance of the PSO-ALBA algorithm improved even when numbers of tasks are increased from 100 to 1000 while fixed number of VMs 50. Table 6 and Fig. 6 shows the throughput of the proposed task scheduling algorithm is also improved when varying the number of tasks from 100 to 1000 and VMs from 40 to 140. The throughput of the proposed algorithm is much improved when compared to PSO standard and ACO algorithm.
TABLE5
THROUGHPUT COMPARISON OF FIX VM
TABLE6
THROUGHPUT COMPARISON OF NUMBER OF TASKS AND VMS
TABLE3
MAKESPAN COMPARISON OF FIX VM
TABLE4
461
Fig. 3. Makespan comparison of proposed PSO-ALBA with other algorithms at fix VMs.
Fig. 4. Makespan comparison of proposed PSO-ALBA with other algorithms at varying tasks and fix VMs.
Fig. 5. Throughput comparison of proposed PSO-ALBA with other algorithms at fix VMs.
Fig. 6. Throughput comparison of proposed PSO-ALBA with other algorithms at varying tasks and fix VMs.
5 CONCLUSIONS
In this paper, the proposed PSO-ALBA task scheduling algorithm capable of balancing the load among resources using adaptive load balancing approach. This adaptive approach scheduled the tasks based on deadline constraints to provide better scheduling solution for cloud computing environment. The proposed algorithm reduces the makespan and maximizes the throughput while comparing other heuristics algorithms. The result shows that the proposed PSO-ALBA algorithm performs much better than that of the other heuristic scheduling algorithms like PSO and ACO. To obtain a better result, the proposed algorithm should be developed and designed more accurate by considering problems associated to SLA-based resource scheduling, to perform hybridization of different optimization techniques. The results can also be implemented by using a workflow application and in real cloud environment.
REFERENCES
[1] H.Saleh, H..Nashaat, W.Saber, H. M.Harb,―IPSO Task Scheduling Algorithm for Large Scale Data in Cloud Computing Environment‖, in IEEE Access, Vol. 7, pp. 5412-5420, (2019).doi: 10.1109/ACCESS.2018.2890067
[2] M. Kumar,and S.C. Sharma, ―PSO-based novel resource scheduling technique to improve QoS parameters in cloud computing‖Neural Computing and Applications (2019).10.1007/s00521-019-04266-x.
[3] S.Pandey, L.Wu, S. M.Guru,and R.Buyya,―A Particle Swarm Optimization-Based Heuristic for Scheduling Workflow Applications in Cloud Computing Environments‖, 2010 24th IEEE International Conference on Advanced Information Networking and Applications, Perth, WA, 2010, pp. 400-407.doi: 10.1109/AINA.2010.31.
[4] F.Luo, Y.Yuan,and W.Ding,―An Improved Particle Swarm Optimization Algorithm Based on Adaptive Weight for Task Scheduling in Cloud Computing‖, CSAE2018, October 2018, Hohhot, China.
[5] R.Z. Khan,and M.O.Ahmad, ―Load balancing challenges in cloud computing: a survey‖, Proceedings of the International Conference on Signal, Networks, Computing and Systems (ICSNCS), Vol. 2, pp.25–32 (2016)
[6] R.N.Calheiros and R.Buyya,―Cost-Effective Provisioning and Scheduling of Deadline-Constrained Applications in Hybrid Clouds‖, In: Wang X.S., Cruz I., Delis A., Huang G. (eds) Web Information Systems Engineering – WISE. 2012. Lecture Notes in Computer Science,vol 7651. Springer, Berlin, Heidelberg.
[7] C.Sudhakar, M.Agroya,and T.Ramesh,―Enhanced Hyper-Heuristic Scheduling Algorithm for Cloud‖, 2018 International Conference on Computing, Power and Communication Technologies (GUCON)Galgotias University, Greater Noida, UP, India. Sep 28-29, (2018).
[8] M.O.Ahmad and R.Z.Khan, ―Load Balancing Tools and Techniques in Cloud Computing: A Systematic Review‖In: Bhatia S., Mishra K., Tiwari S., Singh V. (eds) Advances in Computer and Computational Sciences. Advances in Intelligent Systems and Computing, vol 554. Springer, Singapore. (2018) [9] S.K.Goyal, Singh andManpreet., ―Adaptive and Dynamic Load
Balancing inGrid Using Ant Colony Optimization‖, International Journal of Engineering and Technology (IJET), Vol 4 No 4, pp. 167- 174. (2012)
462
Computing, 2018/3/26, pp. 1-7.
[11] Z.Liu,and X.Wang, ―A PSO-Based Algorithm for Load Balancing in VirtualMachines of Cloud Computing Environment‖, ICSI 2012, LNCS 7331, pp. 142–147, (2012). [12] N.Manikandan, and A.Pravin,―LGSA: Hybrid Task
Scheduling in Multi ObjectiveFunctionality in Cloud Computing Environmen", 3dr Express,2019, pp. 1-16.
[13] D.Chaudhary,and B.Kumar, ―New Balanced Particle Swarm Optimisation for Load Scheduling in Cloud Computing‖,Journal of Information & Knowledge Management, Vol. 17, No. 1 (2018) 1850009 (23 pages).
[14] S. RamKumar, V. Vaithiyanathan, and V.Lavanya, ―Towards Efficient Load Balancing and Green it Mechanisms in Cloud Environment‖,World Appl. Sci. J., 29 (Data Mining and Soft Computing Techniques;. Pp.159-165, (2014)
[15] S. Thanaa, and Alnusairi,: Binary PSOGSA for Load Balancing Task Scheduling in Cloud Environment, (IJACSA) International Journal of Advanced Computer Science and Applications, Vol. 9, No. 5, 2018.
[16] F. Ebadifard, and S.M. Babamir, ―A PSO‐based task scheduling algorithm improved using a load‐balancing technique for the cloud computing environment‖, Wiley Online Library, Volume 30, Issue 12, (2017).
[17] D. Gabi, A S. Ismail, A. Zainal, Z. Zakaria and A. Al-Khasawneh, ―Hybrid cat swarm optimization and simulated annealing for dynamic task scheduling on cloud computing environment‖, Journal of Information and Communication Technology, 17(3), 435-467. (2018)
[18] R. Aron, I. Chana, and A. Abraham, ―A hyper-heuristic approach for resource Provisioning-based scheduling in grid environment", J Supercomputing, DOI 10.1007/s11227-014-1373-9.
[19] J. Kennedy and R. Eberhart, ―Particle swarm optimization‖, Proceedings of ICNN'95 - International Conference on Neural Networks, Perth, WA, Australia,. pp. 1942-1948, (1995). [20] S. Mohapatra and B. Majhi: ―An Evolutionary Approach for
Load Balancing in Cloud Computing‖, Handbook of Research on Securing Cloud-Based Databases with Biometric Applications. IGI Global, pp.433-463. (2015) doi:10.4018/978-1-4666-6559-0.ch020.
[21] M.. Kumar and S.C. Sharma, ―PSO-COGENT: Cost and Energy Efficient scheduling in Cloud environment with deadline constraint, Sustainable Computing‖, Informatics and Systems (2018), https://doi.org/10.1016/j.suscom.2018.06.002.
[22] N. Netjinda, B. Sirinaovakul and T. Achalakul, J Supercomput (2014) 68: 1579. https://doi.org/10.1007/s11227-014-1126-9