ISSN: 2277-4629 (Online) | ISSN: 2250-1827 (Print)

CPMR-IJT Vol. 2, No. 2, December 2012

Comparitive Study of Heuristics Techniques for Resource Allocation in Grid Computing Enviroment
Sudeepa Roy* Dr. Ajay Rana**

ABSTRACT
Grid is a collection of heterogeneous resources which may be computational or non-computational in nature. A grid computing system consists of machines that have different computational capabilities .Grid systems needs to be robust to deal with uncertainty and unpredictability as it operates in environment where system performance may degrade. Robustness of a system is determined by the fact that the system will show limited degradation in challenging environment. This paper discusses the different heuristic techniques for resource allocation and compares them on the basis of robusteness Keywords: Grid, Robustness, Heterogeneus, Heuristics

I. INTRODUCTION
Grid computing [1, 2] is considered to be wide area distributed computing as because it consists of machine sets with varying computing capabilities that can solve problems by allocating idle computing resources across geographically distributed area. Grid provides sharing, selection and aggregation of distributed resources and makes them use for computational purpose. The

important aspect of grid is the scheduling of jobs since there exist high heterogeneity of resources (PCs, Workstations, clusters, and supercomputers) which are geographically distributed and having different time zones, fabric management policies and scheduling, application requirement and design patterns. Scheduling is simple when it comes to single machine as it is just simply assigning the task to machine but, when it comes to grid computing because of their distributed, autonomous and heterogeneity property this strategy will not work. Grid computing systems should be able to assign the tasks of different users to the different avail resources efficiently and utilize the resources of unused devices (known as load balancing/job scheduling/ resource allocation). Purpose of resource allocation is to improve the performance of the grid computing system through an appropriate distribution of the users application tasks. Grid scheduling is NP complete problem [3]. Various heuristics have been developed to solve this Grid scheduling problem. A grid scheduler acts as an interface between the user and distributed resources. It hides the complexities of the computational grid from the grid user [3]. This paper presents a brief discussion on various heuristics and their importance in grid scheduling.

*Student, **Professor Department of CSE, Amity School of Engineering and Technology, Noida,UP, India

www.cpmr.org.in

CPMR-IJT: International Journal of Technology

11

ISSN: 2277-4629 (Online) | ISSN: 2250-1827 (Print)

CPMR-IJT Vol. 2, No. 2, December 2012

II. GRID SYSTEM TASK MODEL

A grid can be considered as a set of sites S which are independent and each having N computing nodes and M number of tasks to be executed at different sites. So, mathematically we can state the grid as G=<S, TM>. A site may have a single computing resources or a number of computing nodes connected in a distributed manner. Since the resources are not exclusively dedicated to

grid usage each site can freely participate in grid computing by offering resources. This interaction between these grid resources during the execution of a job requires the need for a scheduling layer that uses a different scheduling paradigm. The multi-site resources are assigned to the applications (user request) with the help of grid scheduler.

Fig. 1 Grid System

A. Role of Grid scheduler in task alocation

Each site can consists of a set of resources R of different types such as storage, computational and network resources. Every site also has some unique attribute associated with it viz. status of the site, maximum load/ capacity of the site. So, a site S can be represented as: S < {R}, status, max_load> Where, R = the set of resources available at site S, Status = status of the site S i.e., whether the status is working or not working, Max_load = the maximum capacity of the site S Again R can be represented as: R = {M} U {C} U {N}
www.cpmr.org.in

Where, M = set of resources of I/O type, C = set of resources of computational type, N = set of resources of data type. Grid scheduler performs a check on the workload and also the resource availability of each site .If any of the sites satisfies the job request then it is forwarded to that site for execution. Scheduling of jobs is done by grid scheduler and starting of job at site is done by the local scheduler. The process of assigning each task to a machine and scheduling the execution of the tasks on each machine is known as resource allocation/mapping/ resource management [3]. The goal of resource allocation is to achieve high system throughput.

CPMR-IJT: International Journal of Technology

12

ISSN: 2277-4629 (Online) | ISSN: 2250-1827 (Print)

CPMR-IJT Vol. 2, No. 2, December 2012

B. Role of Artificial Intelligence in grid schedulings

With the advent of Artificial Intelligence, resource allocation in Grid computing using Genetic algorithm has been the latest in scheduling algorithms. Genetic algorithm is based on natural law of selection using which jobs can selected from an initial population.

IV. HEURISTIC CLASSIFICATION OF GRID TASK SCHEDULING ALGORITHMS

Heuristic based techniques are based on experience based techniques to solve problems. The four basic heuristic are economic heuristic [4, 5, 6], metaheuristic [7, 8], population based heuristic [9, 3, 10, 11,12]hybrid heuristic [13, 14, 15, 16, 17].

III. PROBLEM STATEMENT

Grid Task scheduling is one of the NP-Complete problems used to find feasible solutions which minimizes cost. Let us assume T1, T2, T3, T4 and T5 are collection of indivisible and independent tasks. The tasks that have no dependency among each others are referred as meta task. Each task is assigned to the resource based on the order in which the tasks are arrived. The mapper controls the execution of all tasks and the tasks have no priority and deadline specified on them. In case of the heterogeneous environment, the size of the meta task and number of resources are known. The estimate of the expected execution time for each task on each resource is known prior to execution. Expected Time to Compute Matrix ETC (Ti, Rj) contains the execution of each task. Where Ti represents meta-task and Rj represents Resource Set. The Problem can be defined as follows: Let task set Ti= T1, T2, T3, T4, , Tn. Let Resource Set Ri = R1, R2, R3, R4,.,Rn. The main aim of the scheduling algorithm is to effectively utilize the idle time of the resources and minimizes the makespan. The makespan can be calculated as follows: Makespan = max (CT (Ti, Rj)) CTij = Rj + ETij CT = Completion Time Rj = Ready Time of Resource j. ETij = Execution time of Task i on Resource j Grid task scheduling is one of the NP-Complete Problem used to find the acceptable solution with fewer cost.
www.cpmr.org.in

A. Population Based Heuristic

It is computational method that optimises problem by taking population of individuals. This heuristic calculates the fitness of each individual and based on their fitness individual are takes out .The new population is used for next iteration of algorithm. When satisfactory fitness level is reached then algorithm stops.

B. Meta Heuristic
It defines a computational method that optimises a problem by iteratively trying to improve a candidate solution. Meta heuristics make few or no assumptions about the problem being optimised and can search very large spaces of candidate solutions. Meta heuristics are used for combinatorial optimisation in which an optimal solution is sought over a discrete search-space.

C. Hybrid Heuristic
Hybrid heuristic are used to optimise the problem by taking two or more heuristic for scheduling jobs in grid e.g. population heuristic + Meta heuristic.

D. Economic Based Heuristic

Resources are rare and there is always scarcity of it in here is always scarcity of resources. Economic heuristic deals with matching jobs to available resources in economical way such that resource provider and consumer get sufficient incentive to stay and play in competitive market..

E. Simple Heuristic
FCFS (First come First serve) means first come first serve. As the name specifies the scheduler executes the jobs in the order of their submission i.e. job submitted earlier will be executed earlier.
CPMR-IJT: International Journal of Technology

13

ISSN: 2277-4629 (Online) | ISSN: 2250-1827 (Print)

CPMR-IJT Vol. 2, No. 2, December 2012

Random: Tasks are selected randomly among all tasks that are submitted but not yet started for execution and this schedule is non-deterministic. Min-min: The Min-min heuristic requires two steps. In first step, machine with minimum completion time is selected for each task. Second step, from all tasks, task with minimum completion time for execution is sent for execution. Max-min: The Max-min heuristic methods first step is same as Min-mins but sends the task with maximum completion time for execution. This strategy is useful in a situation where completion time for tasks varies significantly.

as Simulated Annealing [19], hybrid heuristic such as Genetic Algorithm (GA) and Game Theory [17] and simple heuristic such as Triangular Pyramid Scheduling Model (TPS) [29] ensures minimization make span in grid task scheduling.

Minimize response time

The response time of a job is the time interval between this jobs arrivals and time taken by it into the system until it is completed. Response time includes the waiting time in the queues and the service time in the server [30]. Heuristic such as Fuzzy Model [27] minimizes response time when used in grid task scheduling.

Better workload balancing V. EMERGENCE OF HEURISTIC IN GRID TASK SCHEDULING Easier control of job allocation and improved efficiency:
Heuristic methods along with grid scheduler, ensures access to adequate resources so as to perform correct scheduling and removes any chances of deadlock due to situations where two or more jobs unable to complete and terminating jobs that for any reason are not performing as expected. Workload balancing is the routing of tasks among a group of sites according to the availability and capacity levels of sites. Population based heuristic such as Ant Colony Optimisation (ACO) [3, 10],meta heuristic such as Simulated Annealing [7], hybrid heuristic such as GA and an Ant Algorithm [15],economic based heuristic such as Macroeconomic Approach [4] achieve better load balancing in grid application.

VI. CONCLUSION
There are various scheduling algorithms used to minimize the overall completion time of the tasks. These algorithms find the most suitable resources to be allocated to the tasks in a heterogeneous system. In [12, 17, 18, 19, 20, 21] max-min and min-min algorithms estimate the execution and completion times of each task on all the heterogeneous resources. The min-min algorithm selects the task with minimum completion time and assigns it to the, resource on which minimum execution time is achieved. One of the problems with this algorithm is that it assigns the smaller tasks to the resources with relatively higher computational power. Max-min is one of the variations of min-min algorithm where task with minimum completion time is assigned to resource on which maximum execution time is achieved. Max-min shows better performance than min-min algorithm if the number of shorter tasks is much more than longer ones. In case of max-min algorithm the small tasks may wait
CPMR-IJT: International Journal of Technology

Improves utilization of resources

Resource utilization is defined as amount of resource that is busy in executing jobs. In order for grid resource to be efficiently used, the grid manager should increase their utilization rate so as to decrease their idle time. Population based heuristic such as Ant Colony optimisation (ACO), Particle Swarm Optimisation [12], Fuzzy Model [27] and Economic based heuristic viz. Continuous Auction (CDA) [6] have proven to improve resource utilization.

Minimize Make span

Make span is measure throughput of completion time. It can be calculated as maximum of completion time. Some population based heuristic such as genetic Algorithm [9], Ant Colony Optimisation (ACO) [10], Particle Swarm Optimisation [12], meta-heuristic such
www.cpmr.org.in

14

ISSN: 2277-4629 (Online) | ISSN: 2250-1827 (Print)

CPMR-IJT Vol. 2, No. 2, December 2012

for larger ones to be executed. The researchers in [3, 12] used max-max, a variation of min-min algorithm. They argued that it performs better for static and dynamic mapping problem. QoS (Quality of services) guided min-min technique; a variation of conventional min-min is used in [17, 18]

workshop on grid & cooperative computing (GCC02) Pages 442-450, 2002. [7] proc.15th International Conference on Grid and Cooperation Computing(GCC06), 2006. [8] Shoukat Ali,Anthony A. Maciejewski,Howard Jay Seigel, and Jong-Kook Kim Definition of a Robustness Metric for Resource Allocation in proc.17th International Parrallel and distributed processing symposium(IPDPS03), 2003. [9] R.L. Daniels, J.E Carrilo -robust scheduling for single-machine systems with uncertain processing times , IIE Trans.29 (11) (November-1997), 977-985. [10] A.J. Davenport, C. Gefflot, J.C Beck slackbased techniques for robust schedules, in: Sixth European conference on planning, September 1001, pp.7-18. [11] V.J.Leon, S.D. Wu, R.h. Storer,Robustness measures and robust scheduling for job shops, IIE Trans. 26(5) (September 1994) 32-34. [12] T. D. Braun, H. J. Seigel, N. Beck, L.Boloni, R.F. Freund, D. Hensgen, M.Maheswaran, A. I. Reuther, J. P Robertson, M. D. Theys, B.Yao A comparison of eleven static heuristics for mapping a class of independent tasks onto heterogeneous distributed computing systems Journal of Parallel Distributed computing.61 (6) (june 2001)810-837. [13] J. M. Schopf and F. Berman. Stochastic scheduling in proc of the 1999 ACM/IEEE conference on supercomputing, 1999. [14] A. Burns, S. Punnekkat, B. Littlewood,and D. Wright,Probabilistic Guarrantees for faultTolerant Real-time Systems Technical Report,Design for validation (De Va) TR No. 44,Esprit Long Term Research Project no. 20072,Dept. of Computer Science,Univ. of Newcastle upon Tyne,U.K.,1997.
CPMR-IJT: International Journal of Technology

VII. REFERENCES
[1] I. Foster, C. Kesselman, S. Tuecke, The Anatomy of the Grid: Enabling Scalable Virtual Organizations, International Journal Supercomputer Applications, 2001 [2] Satish Penmatsa & Anthony T. ChronopoulosJob allocation schemes in computational Grids Based on cost optimization. In Proc. 19th IEEE International parallel & distributed processing symposium (IPDPS05), 2005. [3] Prasanna Sugavanam, H.J.Seigel, Anthony A. Maciejewski, Mohana Oltikar, Ashish Mehta, Ron Pichel, Aaron Horiuchi,Vladimir Shestak, Mohammad AL-Qtaibi,Yogish Krishnamurthy, Syed Ali, Junxing Zhang, Mahir Aydin, Panho Lee, Kumara Guru, Michael Raskey, Alan Pippin Robust static allocation of resources for independent tasks under makespan and dollar cost constraints the journal of parallel and distributed computing (JPDC), 2005. [4] K. Q Yan, S. C. Wang, C. P. Chang and J. S. Lin A hybrid load balancing policy underlying grid computing environment in computer standards and interfaces,2006. [5] Ashish M. Mehta, Jay Smith, H.J. Seigel, Anthony A.Maciejewski, Arun Jayaseelan, Bin Ye Dynamic Resource Allocation heuristics that manage tradeoff between makespan and robustin Springer Science-Business Media LLC 2007. [6] Xiaoshan He, Xian-He Sun and Gregor Von Laszewski A QoS Guided Scheduling Algorithm for Grid Computing international
www.cpmr.org.in

15

ISSN: 2277-4629 (Online) | ISSN: 2250-1827 (Print)

CPMR-IJT Vol. 2, No. 2, December 2012

[15] P.Kouvelis, R.Daniels, G. Vairaktarakis Robust scheduling of a two-machine flow shop with uncertain processing times Iie Trans. 38 (5)(May 2000) 421-432. [16] P. Kouvelis,G. Yu Robust Discrete Optimization and its Applications Kluwer Academic Publisher,Dordrecht,1997. [17] Saeed Parsa and Reza Entezari-Maleki RASA: A new task scheduling algorithm in grid environment in World Applied sciences journal 7,152-160, 2009. [18] He, X., X-He sun and G.V. Laszewski, 2003 QoS guided Min-min heuristic for grid task scheduling Journal of computer science and technology, 18:442-451. [19] Maheswaran, M. Sh. Ali, H. Jay Siegel, D. Hensgen and R.F. Freund, 1999 Dynamic

Mapping of a class of independent Tasks onto Heterogeneous computing systems journal of Parallel and Distributed Computing, 59:107131. [20] Dong, F., J. Luo, L. Gao and L. Ge, 2006 A grid task scheduling algorithm based on QoS priority grouping in proc. Of the fifth international conference on grid and cooperative computing (GCC06), IEEE. [21] Etminani, K. and M. Naghibzadeh, 2007 A min-min Max-min Selective algorithm for grid task scheduling The third IEEE/IFIP international conference on internet. Uzbekistan. [22] Volker Hamscher, Uwe Schwiegelshohn, Achim streit, and Ramin Yashyapour evaluation of job scheduling strategies for grid computing in Grid-2000, volume: 1971, issue: 1, publisher: springer, pages: 191-202.

www.cpmr.org.in

CPMR-IJT: International Journal of Technology

16