Issuu on Google+

International Journal of Computer Science Engineering and Information Technology Research (IJCSEITR) ISSN 2249-6831 Vol. 3, Issue 2, Jun 2013, 295-302 © TJPRC Pvt. Ltd.

HIGH COMPACTION COARSE GRAINED JOB SCHEDULING IN GRID COMPUTING ABHINAV SRIVASTAVA, RITURAJ RATHORE & RAKSHA SHARMA Department of Computer Science, Institute of Technology, Guru Ghasidas Viswavidyalaya, Bilaspur, Chhattisgarh, India

ABSTRACT Grid computing is the federation of computer resources from multiple administrative domains to reach a common goal. The Main focus is on Uniform high performance access to computer resources and on demand creation of powerful virtual computer. In this paper we have proposed High Compaction based Coarse-Grained Scheduling: - A coarse-grained algorithm which is well focused to the compaction degree of the resource keeping in mind the various constraints including processing capability (in MIPS), bandwidth (in Mb/s), and memory-size (in Mb) of the available resources. The proposed algorithm introduces job scheduling strategy which schedules jobs in well formed group. Groups are formed in such a way that it utilizes the available resource to maximum extend. In addition to ensure maximum compaction it also maintain balance between processing time and resource utilization. The experimental results obtained using GridSim simulation toolkit demonstrate that the proposed scheduling policy efficiently reduces the processing time of jobs in comparison to grouping based scheduling algorithms.

KEYWORDS: Grid Computing, Job Scheduling, GridSim, Gridlets INTRODUCTION Grid computing is an advanced modern computing technique which enables optimal utilization of computer resources. In theseEnvironments, resources are geographically distributed, managed and owned by various organizations with different policies, and interconnected by wide-area networks or the Internet.Computer scientists in the mid-1990s, inspired by the electrical power grid’s pervasiveness and reliability, began exploring the design and development of a new IT (Information Technology) infrastructure exhibiting quality of seamless access to computing resources distributedacross different organizations [1]. Grid computing systems are loosely coupled, heterogeneous, and geographically dispersed having multiple administrative domains. Grid may be seen as a Cyber infrastructure for coupling and sharing distributed resources such as data, computational power. It may also serve as podium for Peers sharing ideas and collaborative, interpretation of data/results and Remote Visualization [2].In today’s world of high speed computing, computers have become extremely powerful, even home-based desktops are powerful enough to run complex applications. But still we have numerous complex scientific experiments, advanced modeling scenarios, astronomical research, genome matching, a wide variety of simulations, complex scientific as well as business modeling scenarios and real-time personal portfolio management, which require huge amount of computational resources. To satisfy some of these aforementioned requirements, Grid Computing is prodigy [3]. A scheduler is needed to locate the computers resources for execution of an application, and to assign the jobs required. This task involves prioritizing job queues, managing the load, finding idle machines. The Grid scheduling algorithm is the crux of the Grid scheduler. There exist several applications with a large number of lightweight jobs [4]. Job scheduling with light weight gives low performance in terms of processing time and communication time. In order to


296

Abhinav Srivastava, Rituraj Rathore & Raksha Sharma

achieve high performance, it is preferable to schedule light weight jobs in groups instead of light weight jobs. This paper mainly focuses on light weight job scheduling, grouping policies and allocation of these groups to resources in dynamic environment.

RELATED WORK Scheduling algorithm is the heart of the grid computing concept. Various scheduling algorithm have been proposed since 1990, lots of successful attempt have been made on individual job scheduling as well as grouping based coarse-grained job scheduling algorithm. Some of the representative research works on job scheduling in Grid computing have been studied. A Dynamic Job Grouping-Based Scheduling for deploying applications (DJGBSDA), group jobs according to MIPS of the resource. Scheduling framework for Bandwidth-Aware Job Grouping-Based strategy that groups the jobs according to MIPS and Bandwidth of the resource [5]. A Bandwidth-Aware Job Grouping-Based scheduling strategy that schedules the jobs according to MIPS and bandwidth of the resource and the model sends grouped jobs to the resource whose network bandwidth has highest communication or transmission rate[6]. Grouping based fine-grained job scheduling algorithm presents job scheduling algorithm that schedules the groups of jobs to resources according to MIPS and Bandwidth of resources [7]. Grouping-Based Job Scheduling Model (GBJS) added constraints on memory size in addition to MIPS and Bandwidth [8]. Allthese algorithms attempted to utilize resources efficiently. But a very few of them focused their attention on the degree of compaction for optimal grid utilization, which measures the percentage of time in which the resources are been used respect to the overall execution time [9]. Thus, the higher the compaction degree is, the more exploited and less idle the resources are. In this paper we have proposed High Compaction based Coarse-Grained Scheduling: - A coarsegrained algorithm which is well focused to the compaction degree of the resource keeping in mind the various constraints including processing capability (in MIPS), bandwidth (in Mb/s), and memory-size (in Mb) of the available resources.

PROPOSED MODEL This study presents the high compaction based coarse grained job scheduling techniques in grid computing. This scheduling techniques maximizes the utilization of grid resource to the optimized level, reduces the processing time of the jobs and network delay on the grid. This model converts jobs into group of jobs i.e. coarse grained jobs. Grouping Strategy Grouping strategy is based on processing capability (in MIPS), bandwidth (in Mb/s), and memory-size (in Mb) of the available resources. Jobs are grouped according to the capability of the selected resource. Therefore, the following conditions must be satisfied: Groupedjob_MI≤Resource_MIPS* Granularity size …….

(1)

Groupedjob_MS ≤ Resource_MS ……

(2)

Groupedjob_MS ≤ Resource_baud_rate * Tcomm

(3)

Where, MI (Million Instruction) is job’s required computational power, MIPS (Million Instruction Per Second) is processing capability of the resource and Granularity_size (time in seconds) for the job grouping activity, Groupedjob_MS is required Memory Size of group job, Resource_MS is the amount of Memory available at resource, Baud_rate is the bandwidth capacity of resource, and Tcomm is the job’s communication time [4].


High Compaction Coarse Grained Job Scheduling in Grid Computing

297

Equation (1) required computational power of grouped job shouldn’t exceed to the resource’s processing capability. Equation (2) Memory-size requirement of grouped job shouldn’t exceed to the resource’s memory-size capability. In Equation (3) Memory-size of the grouped job shouldn’t exceed to resource’s transfer capability within a specific time period. These are the main factors in job grouping strategy that influences the way job grouping is performed to achieve the minimum job processing time and maximum resource utilization of the Grid system. Implementation of Algorithm The job scheduler the service that resides in the user machine. Therefore, when the user creates a list of gridlets or jobs in the user machine, these jobs are sent to the job scheduler for scheduling arrangement. The job scheduler obtains information about the available resources from the Grid Information Service (GIS). Based on the information, the job scheduling algorithm is used to determine the job grouping and resource selection for grouped jobs. The size of a grouped job depends on the processing requirement length expressed in MI, Bandwidth expressed in Mb/s and Memory size requirement expressed in Mb. When the jobs are put into a group according to the selected resources, the grouped job is dispatched to resources for computation. Figure 1 depicts the architecture of the job scheduler used in the system. The system accepts total number of user jobs with specification such as Job_ID, Job_MI, Job_Memory-size and the available grid resources are specified by Resource_ID, Resource_MIPS, Resource_Bandwidth, Resource_Memorysize. Granularitysize and communication time are user defined in the grid environment (step 1-3). After gathering the details of user jobs and the available resources, the system periodically selects Jobs in First come first serve order after sorting it into descending order (4). The scheduler will then acquires resources in First come first serve order and multiplies the Resource_MIPS with the given granularity size and Resource_BW with communication time and resulting value is saved in grid information system (step 5). The jobs are then grouped according to the algorithm defined by the job scheduler. After grouping the jobs, the scheduler submits the grouped job to the selected resources for job computation After execution the jobs result goes to the user and resource is again available to the Grid and ready to execute the another job. This process continues till all the jobs are grouped and executed in the grid System.

Figure 1: Architecture of Proposed Model

PROPOSED ALGORITHM 

Initialize job: = 0, j: =0;

Initialize Resource: =0;

Input job_MI and job_MS for n jobs;

Sort jobs in descending order of their MI


298

Abhinav Srivastava, Rituraj Rathore & Raksha Sharma

Mini_job:= mini {job_MI and job_MS}

Input R_MIPSand R_MS for for all resources; //MI (Million Instructions) gives computational //capability

R_MI = R_MIPS * granularity_size

Sort resources in descending order of their R_MI //information about various parameters of resource is saved in GIS.

While (j<=n)

{

For i:=0 to (resource_no - 1){

If(R_MI >= group_MI&& R_MS >group_MS)

{

//resource_no is the no. of resources available.

Group_MI := Group_MI+ job_MI;

Group_MS : = Group_MS + job_MS;

J++;

}

Else

{

While( j<No of Jobs && R_MI >=Group_MI&& R_MS >=Group_MS&& R_MI >=Group_jobMI+job_MI&& R_MS >=Group_MS+job_MS&&R_baud_rate>=Group_MS&&(R_MI - group_MI>min_job_MI&&R_MS group_MS>min_job_MS) )

{

Search for max_ job from job []

If( R_MI>max_job + group_MI&&R_MS >max_job+group_MS)

{

Group_MI: = Group_MI + max_job;

Group_MS : = Group_MS + max_job;

}

}//end while

}

}//endfor}


High Compaction Coarse Grained Job Scheduling in Grid Computing

299

Figure 2: Example Illustrating Proposed Grouping Technique

EXPERIMENTAL EVALUATION Experimental Setup and Comparison GridSim simulation toolkit [10] has been used to create the simulationenvironment. A simulation is conducted in heterogeneousenvironment where each cluster has machines with differentcharacteristics such as MIPS, bandwidth and memory size.Constraints on MIPS, memory sizeand bandwidth are kept in mind togroup the jobs according to the available capability of theselected resource. The processing time is taken into accountto analyze the feasibility and to verify the improvement ofproposed model over other scheduling strategies. Table 1: Comparison between HCCG with AFJS and DJGBSDA Processing Time HCCG AFJS DJGBSDA 100 48 81 92 200 93 168 195 300 165 234 254 400 222 335 476 500 264 396 529 Note: for above experimental evaluation granularity_size is taken to be 10 seconds Number of Jobs


300

Abhinav Srivastava, Rituraj Rathore & Raksha Sharma

Figure 3: Comparison between HCCG with AFJS and DJGBSDA

Figure 4: HCCG Scheduling Snapshot for No. of Jobs=100

CONCLUSIONS AND FUTURE WORK High compaction coarse-grained scheduling strategy was proved to be effective in terms of resource utilization. This was due to its ability to have a global vision on the resources requirements for the whole set of jobs and scheduling jobs in groups. The simulation environment had shown that the proposedmodel is able to achieve the mentioned objectives in gridenvironment. Results of the comparative study shows that the proposed HCCG gives better performance than AFJS and DJGBSDA in terms of processing time. In future, this work can be extended to design a parallelscheduler in grid system to realize its performance.

REFERENCES 1.

Foster, I. and Kesselman, C. Computation Grids. Foster, I. and Kesselman, C. eds. The Grid: Blueprint for a New ComputingInfrastructure, Morgan Kaufmann, 1999, 2-48.

2.

RajkumarBuyya and Anthony Sulistio, Service and Utility Oriented Distributed Computing Systems: Challenges andOpportunities for Modeling and Simulation Communities.


High Compaction Coarse Grained Job Scheduling in Grid Computing

301

3.

Kleinrock, L. MEMO on Grid Computing, University of California, Los Angeles, 1969.

4.

N. Muthuvelu, Junyan Liu, N.L. Soe, S. venugopal, A. Sulistio, and R. Buyya, “A dynamic job grouping-based scheduling for deployingapplications with fine-grained tasks on global grids,” in Proc of Australasian workshop on grid computing, vol. 4, pp. 41–48, 2005.

5.

Ng WaiKeat, Ang Tan Fong, “SCHEDULING FRAMEWORK FORBANDWIDTH-AWARE JOB GROUPINGBASED SCHEDULINGIN GRID COMPUTING”, Malaysian Journal of Computer Science,Vol. 19, No. 2, pp. 117-126, 2006.

6.

T.F Ang, W. K. Ng, T.C Ling, “A Bandwidth-Aware Job Grouping-Based Scheduling on Grid Environment”, Information Technology Journal, Vol .8, NO.3, pp. 372-377, 2009.

7.

Quan Liu, Yeqing Liao, “Grouping-based Fine-grained Job Scheduling in Grid Computing”, IEEE First International Workshop on Educational Technology And Computer Science, Vol.1, pp. 556-559, 2009.

8.

Vishnu Kant Soni, Raksha Sharma, Manoj Kumar Mishra, “Grouping-Based Job Scheduling Model in Grid Computing”, World Academy of Science, Engineering and Technology 41 2010.

9.

H´ector Blanco, Fernando Guirado, JosepLluıs Lerida, and V. M. Albornoz, “MIP Model Scheduling for MultiClusters”.

10. Buyya, R. and M. Murshed, 2002. GridSim: A toolkit for the modeling and simulation of distributed management and scheduling for grid computing. Concurrency COmput: Practice Exp., 14: 1175-1220.Buyya, R. and M. Murshed, 2002. GridSim: A toolkit for the modeling and simulation of distributed management and scheduling for grid computing. Concurrency COmput: Practice Exp., 14: 1175-1220.



34 high compaction full