An Optimized Robust Method for Scheduling and Load Balancing in Parallel Computing Systems

(1)

An Optimized Robust Method for Scheduling and Load Balancing in Parallel Computing Systems

Raj Kumar¹ Prof. (Dr.) Vijay Dhir²

1 Research Scholar, IKG PTU, Jalandhar, Punjab India

2 Professor CSE, Punjab, India

Abstract: Parallel computing is a very significant research area which is new and emerging field that holds contribution significantly in various areas of technology and industries. Main aim of this paper is to develop such an algorithm which can do the task scheduling and load baling in such a manner that the overall performance of the system is enhanced by generating the optimum results. As a result of this, larger problems can be solved in a effective manner in shorter span of time which is otherwise impossible for single serial machine. This paper deals with the efficient hybrid method using weighted round robin and minimum completion task scheduling for the successful execution of task with high throughput and efficient completion of jobs with less energy consumption of machines and the proposed performance is evaluated with traditional techniques named as minimum execution time, minimum completion time, round robin and opportunistic load balancing.

Keywords: Keywords- Parallel Computing, Load Balancing, Task Scheduling, MET, MCT, OLB.

_______________________________________________________________________________________

I. INTRODUCTION

Parallel computing deals with the computations where many calculations or the performance implementations of the tasks or processes are passed and executed instantaneously [1]. Huge difficulties can be separated into reduced ones, which are further solved in an efficient manner at the same interval of time. There are various distinct parallel computing systems. They are bit-by-bit, instruction by instruction and comprises of job parallelism process. The process of parallelism has employed to achieve high computing performance, but its acquisition is in high interest because of its physical restrictions that prevents scaling of the frequency [2]. The consumption of the power by computers is a big concern in the parallel computing. Such type of processors or systems has become the leading arrangement in the architecture of the computer which takes the form of high core processors. The tasks based on these computing are closely interrelated to the process of the synchronized computing which can be used frequently together, though these two are totally distinct [3][4]. It is conceivable to work parallel without the action of the concurrency like the process of multitasking by using time sharing arrangement with the use of single processing unit. In this arrangement, a computational process is broken in numerous phases, which can be less or more according to the application requirement that can be handled individually and whose outcomes are collective in nature on the time of completion [5]. In evaluation, in the arrangement of concurrent computing, the processes do not relate with the chores.

The separate processes can be varied in nature and often deal with interrelated communication of the tasks at the time of execution [6].

Research based on parallel computing is discovered in the mid 90's using some efforts which are basically continued or progressed parallel with each other. Cluster Computing is main concern of the parallel computing [7], [8]. The clusters deal with the high performance arrangement which is the choice of various industries with high elasticity, consistency, scalability and costing performance over different work stations [9].

(2)

Fig 1: Parallel computing generalized process

The process of clustering based on parallel computing arrangement, has one centralized node and more than one cluster nodes, which are the processing elements. Cluster performance can be effective by using addition of more efficient nodes, which are involved with the master node [10],[11]. Progressively, parallel handling is the cost-effective scheme for the efficient explanation or solutions of complex data handling difficulties. The appearance of low-cost parallel processors like multiprocessors based desktops and the workstations which have made parallel procedures applicable, as they have such software ethics for portable or movable parallel software design [12][13].

The advantages of having parallel computing in real time scenarios are as follows:

1. Time Saving

2. Solving complex problems

3. Execution of multiple tasks with in the same time intervals.

II. RELATED WORKS

Scheduling and load balancing is achieved on the knowledge of numerous performance evaluations, which will increase the performance of parallel tasks computations [14]. The author shows that the task deals with the data processing, software accessing and storage utilities. The efficient software classifies process reliable with the service- level arrangement and demanded sources [15]. Every process in the parallel processing is then allocated to the one of the available processing servers. The meta-heuristic task scheduling; delivers an outstanding amplification through which it practices the data for achieving scheduling results. Heuristic methods can work as static or dynamic [9] [10].

In addition, such cyclic methods like round robin deals with the FIFO method to perform such scheduling tasks. In addition, it works on the resource management for each task using specific time intervals [16] [17]. Subsequently the processes are queued till the probability for processing of the tasks is achieved. Also the balancing load is also one of the heuristic methodology based scheduling process which also deals with the scheduling of the processes to the following accessible apparatuses based on completion time of the execution [12][13].

Round Robin (RR) This method follow FCFS manner in time sharing environment for scheduling of jobs by giving each job same fixed time quantum on the resource. When quantum expires, current job is pre-empted so that resource is allocated to next job [16].

Opportunistic Load Balancing (OLB) it allocates the given job to the workstation that is ready next. Thus, it tries to keep maximum resources as busy as possible. But main drawback of this method is that poor makespan is resulted as this strategy does not take in to account execution-time of the process [17].

(3)

Minimum Execution Time (MET) That machine, which will take the lowest execution time for the process, will be the first to be scheduled. Because it does not take into consideration the availability of the machine before assigning job to it, load balancing is not achieved in effective way. Time-Complexity of this method is O (m) if we consider m as number of processes [18].

Minimum Completion Time (MCT) That machine, which will take the lowest completion time for the process, will be the first to be scheduled. Because it takes into consideration the availability of the machine before assigning job to it, load balancing is achieved in effective way. Time-complexity of MCT is similar to MET [18].

III. PROPOSED WORK

Round Robin Algorithm does not take into account different processing capabilities of the machines which results in load imbalance. This shortcoming can be taken care of by using the Weighted Round Robin (WRR) where larger processes are assigned to higher priority queues.

MCT covers the limitations of OLB and MET by considering the ready time of the resource to maximizing the overall resource utilization.

Proposed Algorithm (WRR+MCT) combines the benefits of WRR and MCT. It is pre-emptive in nature to balance the load among processors at run time in a better way. So, its dynamic algorithm in nature.

The proposed algorithm steps are as follows:

Step_1: Start.

Step_2: Set the processes so that Ji = 1 to n and Simulation time

Where J is No. of processes and n stands for limit of the total processes.

Step_3: Generate random initial energies Ex and the arrival time of the process Ts.

Step_4: Store the ID’s of processes in one array.

Step_5: Deployment of the queues Q [p] such that p = 1, 2, 3, 5 and Machines M[s] such that s = 1, 2, 3, 4, 5.

Step_6: Generate the burst_time of CPU to complete the processes.

Step_7: Initialization the array C[z] where z ≤1≤ n to complete the processes.

Step_8: Assigning weights W[x] to the queues and select the high priority queue having high weight.

Step_9: Assign jobs to the high queues and execution through machines having high bandwidth for the minimum completion time of jobs.

Step_10: Calculate time wanted for the process, store its Id for the current execution of the process.

Step_11: For i = 1: job count If C(t)< T (j)

Evaluate the power consumption, completed processes and completion time for the task.

End If End for

(4)

Where C (t) = current time and T (j) = time required per job.

Step_12: Evaluate the process for the incomplete jobs after the burst time.

Step_13: Evaluate the minimum completion time process to find the jobs which are executed and took minimum time to execute.

Step_14: Evaluate the power consumption to execute the number of jobs.

Step_15: Evaluate the latency, throughput and execution time taken by machines to execute the jobs parallel.

Step_16: Stop.

IV. RESULTS AND DISCUSSION

In this research we have attained whole simulation in MATLAB. We have used MATLAB because it’s an efficient strong technical computing tool which is used to analyze the algorithm and execution in an effectual manner. So below are the results and discussions of the proposed method.

Figure 2 shows the energy consumption of the system to execute the number of the tasks at given time simulations and exhibit that proposed methodology can attain low energy consumption which shows the robustness of the machines to execute the number of jobs.

Figure 3 Shows that the actual execution time taken by proposed method is less than the expected execution time considered by the other traditional methods.

Figure 4 shows that the latency in execution of all processes is lowest in proposed method to give real time solutions.

Figure 5 shows the highest throughput achieved by proposed method in terms of kilo bits per second (kbps) in evaluation to other methods.

Figure 6 shows the minimum completion time taken by hybrid method in execution of all the processes in given simulation time. As we know makespan is the total time taken by an algorithm for execution of given tasks from earliest start time of first process to latest finish time of last process. The makespan for given input data set is also obtainable from this figure and clearly proposed hybrid method is attaining lowest makespan than other traditional methods.

Thus, using MATLAB, Simulators for Round Robin (RR), Opportunistic Load Balancing (OLB), Minimum Execution Time (MET), Minimum Completion Time (MCT) and Proposed Hybrid Approach (WRR+MCT) are designed. Data Set consisting of random arrival times of processes, the power consumption by these processes and other data traffic is used as input for analyzing performance of all these simulators. Each Simulator needs input as number of processes for which scheduling is to be done so that the load is optimally balanced among various parallel processors and the appropriate simulation time needed by the simulator.

The experimental results obtained from these simulators are shown on next pages.

(5)

Fig 2: Energy Consumption Evaluation

Fig 3: Execution Time Evaluation (in ms)

(6)

Fig 4: Latency Evaluation (in ms)

Fig 5: Throughput Evaluation (in kbps)

(7)

Fig 6: Completion Time Evaluation

Method EC

(mJ)

ET (ms)

LT (ms)

TH (kbps)

CT

(ms) Makespan Reliability Deadline

RR 1271 41.96 597.6 2.812 65.81-

92.18 92.18

Less Reliable

Less Promising

MCT 241 34.96 5041 19.59 1.17-

50.41 50.41

MET 253 35.14 6.441 19.38 2.099-

54.63 54.63

OLB 615 34 3783 0.1969 1.494-

31.65 31.65 Proposed

(WRR+MCT) 9.609 4.96 0.128 3153 8.872-

6.457 6.457 More Reliable

More Promising Table A: Here different performance metrics like Energy Consumption, Execution Time, Throughput, Completion Time and Makespan are summarized quantitatively for all traditional algorithms as well as proposed algorithm. Based on these optimal quantitative results of proposed hybrid method, it can evident that qualitatively proposed method is more reliable than other methods in delivering the optimal results in given deadline of budget [19].

(8)

Where EC = Energy Consumption in mJ ET = Execution Time in ms

LT = Latency in ms TH = Throughput in kbps EC = Energy Consumption and Makespan

All these above five metrics are quantitative in nature while reliability and deadline are taken as qualitative ones From above table, we can see that the performance of our proposed method is coming far better and is achieving high efficiency in evaluating all the number of jobs by the machines to provide all the necessary resources to the users. The proposed hybrid method can attain optimal results which increase the efficiency of the parallel computing tasks in an effectual manner.

V. CONCLUSION

This paper deals with the parallel computing method using the hybridization method using weighted round robin and minimum completion time for the scheduling and the load balancing of the computing system to achieve less execution time, high throughput

Future work deals with optimization scenarios if the execution of the machines fails to implement the jobs assigned to it due to any halt condition

REFERENCES

1. V. Dhir, G. Kaur , “Execution of cloud using freeware Technology” , International Journal of Engineering Research in Computer Science and Engineering (IJERCSE), Vol 3, Issue 12, pp 22-29, 2016.

2. Saanjay Kumar Panda, “TBS: A Threshold Based Scheduling in Grid Environment”, 3^rd IEEE International Advanced Computing Conferenece (IACC), pp 23-27, 2013.

3. A. Kaminsky, “BIG CPU, BIG DATA: Solving the World's Toughest Computational Problems with Parallel Computing”, Create Space Independent Publishing Platform, 2016.

4. T. Deepa, Dhanaraj Cheelu, “ A comparative study of static and dynamic load balancing algorithms in cloud computing”, International Conference on Energy, Communication, Data Analytics and Soft Computing (ICECDS), 2017.

5. D. Kliazovich, J. E. Pecero, A. Tchernykh, P. Bouvry, S. U. Khan, A. Y. Zomaya. "CA-DAG: Modeling communication-aware applications for scheduling in cloud computing", Journal of Grid Computing 14, no. 1, pp:

23-39, 2016.

6. S. Singh, I. Chana. "A survey on resource scheduling in cloud computing: Issues and challenges." Journal of grid computing 14, no. 2, pp: 217-264, 2016.

7. S.J. Ahmad, C. S. Liew, E. U. Munir, T. F. Ang, S. U. Khan, "A hybrid genetic algorithm for optimization of scheduling workflow applications in heterogeneous computing systems", Journal of Parallel and Distributed Computing 87, pp: 80-90, 2016.

8. A. Haq, M. A. Shah, "Task Scheduling in Parallel Processing: Analysis", Journal of Biometrics & Biostatistics, 2015.

9. A. Sharma, S. Verma, "A Survey Report on Load Balancing Algorithm in Grid Computing Environment", Int. J.

Adv. Engg. Res. Studies/IV/II/Jan.-March 128, pp:132 -138, 2015.

10. R. M. Singh, S. Paul, A. Kumar, "Task scheduling in cloud computing: Review." International Journal of Computer Science and Information Technologies 5.6, pp: 7940-7944, 2014.

11. S. Nagadevi, K. Satyapriya, Dr.D.Malathy3, “A Survey on Economic Cloud Schedulers for Optimized Task Scheduling”, International Journal of Advanced Engineering Technology, 2013.

12. W. Huai, Z. Qian, X. Li, G. Luo, and S. Lu, “Energy Aware Task Scheduling in Data Centers, Journal of Wireless Mobile Networks, Ubiquitous Computing, and Dependable Applications”, volume: 4, number: 2, pp. 18-38, 2013.

13. D. Kliazovich, S. T. Arzo, F. Granelli, P. Bouvry, S. U. Khan, “e-STAB: Energy-Efficient Scheduling for Cloud Computing Applications with Traffic Load Balancing”, IEEE International Conference on Green Computing and Communications and IEEE Internet of Things and IEEE Cyber, Physical and Social Computing e-STAB, 2013.

(9)

14. S. H. Jang, T. Y. Kim, J. K. Kim , J. S. L. School, “The Study of Genetic Algorithm-based Task Scheduling for Cloud Computing”, International Journal of Control and Automation Vol. 5, No. 4, December, 2012

15. S. Sudhir, “Survey on Scheduling Issues in Cloud Computing”, Procedia Engineering, Elsevier, 2012.

16. A. mantri, S. Nandi, G. Kumar, S. Kumar, High Performance Architecture and Grid Computing, International Conference HPAGC 2011, Chandigarh, India, July, Proceedings, 2011.

17. V. Dhir, "Alchemi .NET Framework in Grid Computing", Proceedings of the 3rd National Conference;

INDIACom-2009 Computing For Nation Development at Bharati Vidyapeeth’s Institute of Computer Applications and Management, New Delhi. 2009.

18. M Hemamalini, M V Srinath, “Performance Analysis of Balanced Minimum Execution Time Grid Task Scheduling Algorithm”, Internaional Journal of Communication and Networking System, Vol 5, Issue 2, pp 96-100, 2017.

19. Shahbaz Afzal, Kavitha Ganesh, “A Taxonomic Classification of Load Balancing Metrics: A Systematic Review”, 33^rd Indian Engineering Congress, Udaipur, 2018: Technical Volume, pp: 85-90, 2018.

Authors

Mr. Raj Kumar pursued Bachelor of Technology from Beant College of Engineering and Technology, Gurdaspur, Punjab, India in 1999 and Master of Technology from Guru Nanak Dev University, Amritsar, Punjab India in year 2003. He is currently pursuing Ph.D from IKG-PTU Jalandhar Punjab India and currently working as Manager at SJVN Ltd Shimla HP India since 2004. He is a Life time member of Computer Society of India since 2010. He has published more than 5 research papers in reputed UGC approved International journals and conferences and it’s also available Online. His main research work focuses on Distributed Processing, Cloud Processing and Parallel Computing. He has 15 years of teaching/industrial experience and 3 years of research experience.

Author’s Name, Author’s profile. Dr Vijay Dhir pursued Bachelor of Technology, Master of Technology, and PhD from PTU Jalandhar. He is member of board of study in department of CSE IGK-PTU, Jalanadhar. He is a member of IEEE & IEEE computer society since 2009, ACM since 2010. He has published more than 85 research papers in reputed international journals including Thomson Reuters (ESCI & Web of Science) and conferences including IEEE and it’s also available online. His main research work focuses on Big Data Analytics, Image Processing, Digital Forensics, Distributed, Cloud and Parallel Computing. He has 19 years of teaching experience and 5 years of research experience.

Picture size should be absolute 3.18cm in height

and absolute 2.65cm in width