Efficient Load Balancing Grouping based Job Scheduling Algorithm in Grid Computing

Efficient Load Balancing Grouping based Job Scheduling Algorithm in Grid Computing Sandeep Kaur 1, Sukhpreet kaur 2 1,2 Sri Guru Granth Sahib World University Fatehgarh Sahib, India Abstract: Grid computing is a high performance computing environment to solve large-scale computational demands. Computational grids emerged as a next generation computing platform which is a collection of heterogeneous computing resources connected by a network across dynamic and geographically dispersed organizations, to form a distributed high performance computing infrastructure. In computational computational grid main emphasis is given on resource management and job scheduling. Job scheduling is a decision process by which application components are assigned to available resources to optimize various performance metrics. The main goal of scheduling is to maximize the resource utilization and minimize processing time of the jobs. Various research works has been done on job scheduling problem in grid, but still further analysis and research needs to be done to improve the performance of scheduling algorithm in computational grid. Different scheduling algorithms are suitable in different situation based on specific requirement. User jobs might be small and of varying lengths according to their requirements. Certainly, it is a real challenge to design an efficient scheduling strategy to achieve high performance in grid computing. Grouping strategy reduce the communication time of small scale jobs, but allocation of large number of jobs to one resource will increase the processing time and leads unbalancing processing load among the resources in grid computing environment. In this paper, an Efficient Load Balancing Dynamic Grouping based Job Scheduling (ELBDGBJS) approach has been proposed for grouping the fine-grained jobs in grid computing environment. In this scheduling algorithm jobs are scheduled based on number of jobs at particular time and resources capability. Independent fine-grained jobs are grouped together based on the dynamically specified group size and resources characteristics, for all resource utilization and minimize processing time. Hence in this paper, we have specifically focused on improving computational grid performance in terms of equal load balance and total computation time. A comparison of our proposed approach with other existing fine-grained job scheduling strategies is provided. Experimental results show, proposed algorithm (ELBDGBJS) performs better. Keywords: Grid computing, Job scheduling algorithm, Grouping Strategy, Load Balancing. 1. INTRODUCTION The popularity of the Internet and the availability of powerful computers and high-speed networks as low cost commodity components are changing the way we use computers today. These technical opportunities have led to the possibility of using geographically distributed and multi-owner resources to solve large-scale problems in science, engineering, and commerce. Recent research on these topics has led to the emergence of a new paradigm known as Grid computing [7] Grid computing is a high performance computing environment to solve large scale computational demands. The computational speed of individual computers has rapidly increased in the past thirty years. However, they are still not faster enough for large-scale scientific problems. This is why universities around the world are leading the development of the Grid computing which aims to link hundreds of major computing centers around the world. The term grid computing originated in the early 1990s as a metaphor for making computer power as easy to access as an electric power grid. The job scheduling and resource management is the central component of a grid system. However, grid performance can be improved in terms of job processing time by making sure all the resource available in the grid are utilized optimally using a good job scheduling algorithm [13]. An improved scheduling and efficient load balancing algorithm across the grid may lead to improve the overall system performance with less execution time. Although Grids have been used extensively for executing intensive jobs, but there are also exist several applications with a large number of lightweight jobs which needs small processing requirements. Sending/receiving each small job individually to/from the resources will increase the total communication time and cost. These applications involve high overhead time and cost in terms of job transmission to and from Grid resources and, job processing at the Grid resources. Therefore, there is a need for an efficient job grouping-based scheduling system to dynamically assemble the lightweight jobs of an application into a group of jobs, and send these grouped jobs to the grid resources. The dynamic grouping should be done based on the computational requirements of each job with respect to availability, processing capability of computational grid resources. 2. LITERATURE SURVEY The author has [7] proposed an algorithm that combines the advantages of three algorithms first shortest processing time, second, longest processing time, and third, earliest deadline first. The author realize that virtualization technique is required to address the problem of resource unavailability and software mismatch, With the help of virtualization a dynamic user Volume 2, Issue 4 July August 2013 Page 138

defined environment is created, so there is less possibility of rejecting a job due to unavailability of resource. Since this is the hybrid of virtual and physical machine scheduling, overhead in creating virtual machine in the presence of physical resource is reduced. The author in [2] proposed ORC scheduling strategy in Grid environment which includes the best fit followed by round robin scheduling for distribute the jobs among the available processors. The remaining un-allotted jobs are queued for next execution. Thus the order of jobs execution increases the performance of the processor and distribute load effectively across the network. This will reduce the waiting time of jobs in the queue and avoid the starvation. Many developed algorithms either considered the resource characteristics or application characteristics. But the adaptive fine grained job scheduling algorithm considered both characteristics. The authors in [3] feel there is a need to reduce the communication time, processing time and enhance resource utilization in case of scheduling the light-weight or small jobs. There are many applications in which Consist a large number of lightweight or less processing requirement jobs. Scheduling with light weight gives low performance in terms of processing time and communication time. So to achieve high Performance, less processing requirement jobs are grouped before allocation of resource. This grouping algorithm integrated Greedy algorithm and FCFS algorithm to improve the processing undertake of fine-grained jobs. The algorithm considers the dynamic characteristic of the grid environment. The time complexity scheduling algorithm is very high. It does not pay any attention to memory size constraint and pre-processing time of job grouping is high. The proposed a Dynamic Job Grouping-Based scheduling [4] maximizes the utilization of resource, and reduces the overhead time. This strategy dynamically assembles the individual Fine-grained jobs of an application into a group, and sends these coarse-grained jobs to the Grid resources. This dynamic grouping strategy based on the processing requirements of each application, Grid resources s availability and their processing capability and granularity size. Granularity size is the job processing time at the resources is used to measure the total amount of jobs that can be completed within a specified time in a particular resource. Relationship between the total number of jobs, processing requirements of those jobs, total number of available Grid resources, processing capabilities of those resources and the granularity size should be determined in order to achieve the minimum processing time. The author described a framework which is [5] comparatively better than nonbandwidth- aware job grouping scheduling framework. After creation a list of jobs, sent to the job scheduler for scheduling arrangement. The job scheduler obtains information on available resources and their network information from the Grid Information Service (GIS), and routing information from routers. Based on the information, the job scheduling algorithm is used to determine the job grouping and resource selection for grouped jobs. Once all the jobs are put into groups with selected resources, the grouped jobs are dispatched to their corresponding resources for computation. The bandwidth aware algorithm [6] improves the performance by reducing the delaying factors in network environment and maximizing the utilization of grid resources. Bandwidth-awareness strategy combined with the dynamic job grouping strategy to improve transmission delay and overhead. After knowing the processing capability and the bottleneck bandwidth, the scheduler selects the resource with the largest bottleneck bandwidth and started grouping independent fine-grained jobs together based on the chosen resource`s processing capability. Job groups are sent based on Largest Job First (LJF) manner to the corresponding resource. Longer execution time jobs executed concurrently with shorter execution time jobs. This strategy improves the overall execution time of the jobs. The grouping depends on a granularity size. It is the time within which a job is processed at the resources. This value is used to measure the total amount of jobs that can be completed within a specified time in a certain resource. It is one of the main factors in job grouping strategy that influences the way of job grouping to achieve the minimum job execution time and maximum utilization of the Grid resources. The author in [8] calculated the percentage of the processing capability of a resource on the total processing capability of all the resources. Using this percentage, the processing capability of a resource based on the total length of all tasks to be scheduled is calculated. The scheduler groups the jobs according to the calculated processing capability. The new job group is scheduled to execute in the available resources. This process of grouping and scheduling is repeated until all the user jobs are grouped and assigned to selected grid resources. Due to job grouping this approach optimizes computation/communication ratio and enhances utilization of resources. The Authors state that [12] allocating large number of jobs to one resource will increase the processing time. So to avoid this situation during job grouping activity, the total number of jobs group should be created such that the processing loads among the selected resource are balanced. There are some defects in the above scheduling algorithms. First, the algorithms don t take the dynamic resource characteristics into account. Second, the grouping strategy can t utilize resource sufficiently. And finally, it doesn t pay attention to the network bandwidth and memory size. To solve the problems mentioned above in [9] an adaptive fine grained job scheduling mechanism is presented. This algorithm is divided into two parts. In the first part, the In the second part after gathering the details of user jobs and the available resources, the system selects Jobs in FCFS Volume 2, Issue 4 July August 2013 Page 139

order to form different job groups. The scheduler selects resources in FCFS order after sorting them in descending order of their MIPS. Jobs are put into a job group one after another until sum of the resource requirements of the jobs in that group is less than or equal to amount of resource available at the selected resource site. A described model in [11] obtains the information about resources and job. This information is used for job grouping and for resource selection. When the jobs are put into a group according to the selected resources, the grouped job is dispatched to resources for computation. GBJS gives better performance than AFJS and DJGBSDA in terms of processing time. The proposed TMDGJS in [15] terms of processing time and overhead time, gives better performance, by minimizing overhead time and computation time. This strategy play important role in reducing overall processing time of jobs. Jobs are put alternatively from shortest list and added into job group according to the processing capability of the selected resource. Various job grouping based scheduling algorithms in [16] grid computing has been surveyed. A Grouping strategy based on the processing capability of selected resource plays a vital role to improve the overall performance of grid computing environment. However allocation of large number of jobs to one resource will increase the processing time and leads unbalancing processing load among the resources in grid computing environment. So, there is an efficient job grouping-based scheduling strategy is required during the job grouping activity, so that the processing loads among the selected resource are balanced. The term "load balancing in [17] refers to the technique that tries to distribute work load between several computers, network links, CPUs, hard drives, or other resources, in order to get optimal resource utilization, throughput, or response. To minimize the time needed to perform all tasks, the workload has to be evenly distributed over all nodes which are based on their processing capabilities. This is why load balancing is needed. The load balancing problem is closely related to scheduling and resource allocation. Grouping strategy based on the processing capability of selected resource. After resource selection, Jobs are started to put into the group. This process continues until the resource capability is less to the sum of grouped job requirement. This technique is also useful in maximum resource utilization, as jobs are grouped based on resource capability, but some system are fully occupied whereas other might be remains unutilized or idle in grid environment. 3. GROUPING STRATEGY Jobs are received and stored in job list. Then resources are sorted out in descending order of their processing capabilities in MIPS and jobs according to the job length.after that total number of jobs and resources are calculated and number of jobs in each group is specified. G_size = JList_Size / RList_Size Jobs are added into group based on group size by alternatively taking jobs from front end i.e. job with highest length and then rear end of the job list i.e. job with smallest length in job list. While grouping the jobs taken in order of the grouping strategy from the job list, the above strategy falls into cases given below. G (i) _size<=g_size When above condition fails while adding a job alternatively from job list during the grouping operation, then it stops grouping and job group send to the corresponding resource for computation. The above job grouping process is repeated as long as jobs exist. The advantage of this grouping strategy is as follows. 1) Balance the processing load among the selected resources. 2) To start the grouping process with the addition of a job that involves bigger computational need. 3) To ensure a well combination of bigger and smaller jobs added into the group. 4. SCHEDULING FRAMEWORK The scheduling framework illustrated in Figure 1 depicts the design of the job scheduler and its interactions with other entities. Therefore, when the user creates a list of jobs, these jobs are sent to the job scheduler for scheduling arrangement. The job scheduler obtains information on available resources. After gathering details, the scheduler specifies the group size based on job_count and resource_count. Based on the information, the job scheduling algorithm is used to determine the job grouping. Then scheduler submits the job groups to their respective resources for computation. After executing the group job, results goes to back to the corresponding users and the resource is again available to scheduler system. Figure 1 Job Grouping-based Scheduling Framework 5. EFFICIENT LOAD BALANCING DYNAMIC GROUPING BASED JOB SCHEDULING ALGORITHM 1. The scheduler receives the job _List created by the user. 2. Sort the job_list, according to their MI in decreasing order. 3. Count the jobs and store in variable JList_Size. Volume 2, Issue 4 July August 2013 Page 140

4. The scheduler receives the Resource_List, R [j]. 5. Count the number of resources and store in RList_Size variable. 6. Sort the Resources in decreasing order according to their MIPS. 7. Get the MIPS of the Resource j. 8. Resource_MI=R[j] x Chunk_Size. 9. Define the total size of each group TG_size = JList_Size / RList_Size 10. Set the Group (i) _size to zero. (i=0, 1, 2, 3 n) 11. Set the Grouped_Gridlet_Length to zero. If Group (i) _size<=tg_size and Grouped_Gridlet_Length is less than Resource[j] _MI Job are selected from sorted out job_list alternatively and added to summation of previous Grouped_Gridlet_Length and Group (i) _size is increment Else Deduct the last job added to Group (i) _size 12. Submit the Group (i) _size to Resource, R[j]. 13. Increment i (for next group) 14. Increment j (for next resource in Resource-List) and goto step10. Algorithm Description In this section we describe a Load Balancing Dynamic Grouping based Job Scheduling algorithm. The algorithm mainly consists of two phases: (1) Create available resource list and job list (2) Job grouping and scheduling. Grouping of job is done based on the total group size and available resources. First user submitted jobs are collected and one job list is created, then scheduler collects available resource characteristics. And then jobs are sorted out. After that group size is calculated based on job count and available resource.scheduler select job alternatively and job group are created for scheduling based on group size. The proposed job grouping and scheduling algorithm presented next is illustrated through an example. In this example 15 user jobs with varying processing requirements (MI) are grouped into five job groups taking into account the number of jobs group and according to the processing capabilities (MIPS) of the available resources and the Chunk_size. Resource MI is calculated as: RMI=MIPS*Chunk_Size and the Chunk_Size is taken as 10. Before grouping, the group_size is specified. The jobs are added into according to the grouping strategy as mentioned and (the first job is always taken from the front end of the reverse sorted job list. Then, the next job is added from the rear end of the list. 6. EXAMPLE Job_list with job length (MI Table 1: User Jobs R_list with MIPS Table 2: Resource List Chunk Size=10 RMI=MIPS*chunk size Table 3: Comparison of TMDGJS and ELBDGJS based on processing load Figure 2 Comparison of TMDGJS and ELBDGJS based On processing load Volume 2, Issue 4 July August 2013 Page 141

7. SCHEDULING ACTIVITY Figure 3 Activity Diagram for Job Grouping-based Scheduling Framework execution time, memory-size requirement. In this experiment, jobs and resources are randomly generated and the number of jobs varies from 100 to 500. The processing time is taken into account to analyze the feasibility of the proposed scheduling algorithm. Our algorithm can reduce the execution time and also the job completion success rate is high. The results are obtained through various observations with different Chunk_Size. Simulation Environment :-( Jobs are created with deviation of 20%) In this simulation jobs are created with average MI of 200 and deviation of 20%. Chunk_Size of 10, 20 and 30 are taken for observation 1, observation 2 and observation 3 respectively to analyze the processing time of submitted jobs. Observation:-1 The Figure-4 given below represents the processing time of two algorithms for the Chunk_Size 10 with different user jobs of average MI 200 with deviation of 20%... Observation:-2 The Figure-5 corresponds to the processing time of two algorithms for the Chunk_Size 20 with different user jobs of average MI 200 with a deviation of 20%. Observation:-3 The observation in Figure-6 signifies the processing time of two algorithms for the Chunk_Size 30 with different user jobs of average MI 200 with deviation of 20%. 8. SIMULATION AND RESULT Extensive simulation has created. The inputs to the simulations are total number of randomly generated jobs with MI deviation percentage, resource MIPS and different Chunk_Size. Simulations are conducted using ten resources of different MIPS. Resources associated with different Chunk_Size such as 10, 20, and 30. The details of the resource list are shown in Table 4. Table 4 Resource List Figure 4 Processing time with Chunk_Size=10 In this simulation, the total processing time is obtained in seconds. The purpose of this Simulation is to analyze and compare the performance difference between two scheduling algorithms: A Time-Minimization Dynamic Job Grouping-Based Scheduling algorithm (TMDGBJS), and proposed An Efficient Load balancing Dynamic Job Groupong-Based Scheduling algorithm. The scheduling and grouping strategy of TMDGBJS is adopted from [15]. Simulation: In our simulation, each resource is characterized by its MIPS and chunk_size. The jobs are characterized by their amount of computations, expected Figure 5 Processing time with Chunk_Size=20 Figure 6 Processing time with Chunk_Size=30 Volume 2, Issue 4 July August 2013 Page 142

9. Conclusion and Future Scope Extensive simulation environments are conducted to study and compare the behavior of the proposed TMDGJS and ELBDJGBS. The simulation results have shown that the proposed scheduling algorithm is able to achieve the desired objectives in grid environment. The comparative study done through various observations shows that the proposed ELBDGJS gives better performance than TMDGJS. The proposed Scheduling is a grouping based job scheduling strategy that has taken Group size constraint for grouping the jobs into account. The grouping algorithm improves the processing time of fine grained jobs. Experimental result demonstrates efficiency and effectiveness of the proposed algorithm. This scheduling algorithm balances the processing load among the resource and gives the better result in execution time. Advantages of this algorithm compared with other are: It reduces the total processing time of jobs. Grouping the jobs fine-grained into group will reduce the network latencies. Balance the processing load among the selected resource. To further test and improve the algorithm, some dynamic factors such as network delay and QoS constraints can be taken into account. References [1] Ian Foster, Carl Kesselman, Jefrey M. Nick, and Steven Tuecke, Grid services for distributed system integration. Computer, 35:37 46, June 2002. (journal [2] K.Somasundaram, and S.Radhakrishnan (2008), Node Allocation In Grid Computing Using Optimal Resource Constraint (ORC) Scheduling, IJCSNS International Journal of Computer Science and Network Security, VOL.8 No.6, June 2008.. (journal [3] Q. Liu, Y. Liao, Grouping-Based Fine- grained Job Scheduling in Grid Computing, Vol.1, pp.556-559, IEEE First International Workshop on Education Technology and Computer Science, 2009. (journal [4] N. Muthuvelu, J. Liu, N.L Soe, S.Venugopal, A. Sulistio,A and R.Buyya, A Dynamic Job Grouping- Based Scheduling for Deploying Applications with Fine-Grained Tasks on Global Grids, Conferences in Research and Practice in Information Technology, Vol. 44. (conference [5] N.W keat, A.T Fong, L.T Chaw, L.C Sun, Scheduling Framework for Bandwidth-Aware Job Grouping-Based Scheduling in Grid Computing Malaysian Journal of Computer Science, Vol. 19(2), 2006. (Journal [6] T.F Ang, W.K Ng, T.C. Ling, L.Y. Por and C.W Liew, A Bandwidth -Aware Job Grouping Based Scheduling in Grid Environment. InformationTechnology Journal, 8: 372-377. Information Technology Journal, 8: 372-377. (Journal [7] S.T Selvi, P.R SathiaBham, S. Architha, T. Kaarunya and K.Vinothini, "Scheduling In Virtualized Grid Environment Using Hybrid Approach", International Journal of Grid Computing & Applications Vol.1. No.1, September 2010. (Journal [8] S.Selvarani, and G.S.Sadhasivam, Improved Job Groupin based PSO Algorithm for Task Scheduling in Grid Computing International Journal of Engineering Science and Technology Vol. 2(9), 2010, 4687-4695.(Journal [9] S. Gomathi, and D.Manimegalai, An Analysis of MIPS Group Based Job Scheduling Algorithm with other Algorithms in Grid Computing IJCSI International Journal of Computer Science Issues, Vol. 8, Issue 6, No 3, November 2011. (Journal [10] R.Sharma, V.K. Soni, M.K Mishra, P. Bhuya A Survey of Job Scheduling and Resource Management in Grid Computing World Academy of Science, Engineering and Technology 40 2010.. (conference [11] V.K Soni, R.Sharma, M.K Mishra, Grouping-Based Job Scheduling Model in Grid Computing. World Academy of Science, Engineering and Technology 41 2010(conference [12] P. Rosemarry, R.Singh, P.Singhal, And D. Sisodia, Grouping Based Job Scheduling Algorithm Using Priority Queue And Hybrid Algorithm in Grid Computing. International Journal of Grid Computing & Applications (IJGCA) Vol.3, No.4, December 2012. (Journal [13] A. Mukherjee and P.M.Khilar, An Efficient Job- Grouping Based Scheduling Algorithm for Fine- Grained Jobs in Computational Grids.2011. (Thesis l [14] G.Sharma, and J.K Bhatia, A Review on Different Approaches for Load Balancing in Computational Grid Journal of Global Research in Computer Volume 2, Issue 4 July August 2013 Page 143

Science, Volume 4, No. 4, April 2013.. (Journal [15] M.K.Mishra, P. Mohanty, G.B Mund, (2012) A Time-minimization Dynamic Job Grouping based scheduling in Grid Computing, International Journal of Computer Applications (0975 8887) Volume 40 No.16, February 2012.. (Journal [16] K.Sandeep, K.Sukhpreet, Survey of Resource and Grouping Based Job Scheduling Algorithm in Grid Computing, International Journal of Computer Science and Mobile Computing, IJCSMC, Vol. 2, Issue. 5, May 2013, pg.214 218 (Journal [17] A.Revar, M.Andhariya, D.Sutariya, Prof. M.Bhavsar, Load Balancing in Grid Environment using Machine Learning Innovative Approach, International Journal of Computer Applications (0975 8887) Volume 8 No.10, October 2010. (Journal Volume 2, Issue 4 July August 2013 Page 144