A Hierarchical Resource Allocation Architecture for Mobile Grid Environments

Document Sample
A Hierarchical Resource Allocation Architecture for Mobile Grid Environments Powered By Docstoc
					                                                          (IJCSIS) International Journal of Computer Science and Information Security,
                                                          Vol. 8, No. 5, August 2010




  A Hierarchical Resource Allocation Architecture for
              Mobile Grid Environments

                       S. Thenmozhi                                                                A.Tamilarasi
Assistant Professor, Department of Computer Applications                                 Professor, Department of MCA
    Chettinad College of Engineering and Technology                                     Kongu Engineering College, Erode
                     Tamilnadu, India.                                                          Tamilnadu, India
               sthenmozhiphd@gmail.com


Abstract— The mobility issue in grid environments has                    secure and efficient way. Moreover, it has the capability to
established new challenges to the research communities                   organize the underlying ad hoc networks. It forms arbitrary
particularly in the areas of scheduling, adaptation, security and        and unpredictable topologies by providing a self-configuring
mobility. Especially, the resource allocation becomes more               grid system of mobile resources which are connected by the
challenging when mobility is considered in grid environment.             wireless links [1].
Hence it is necessary to consider the mobility of users along with
the resource availability while scheduling the resources for the             The mobile grid uses the advanced capabilities of wireless
execution of jobs. In this paper, we propose to design a                 networks and lightweight thin devices. Though grid computing
Hierarchical Resource Allocation Architecture (HRAA) which               integrates geographically dispersed resources and users to
includes resource monitoring and scheduling operations for               create a dynamic virtual organization, most of the resources
mobile grid. In this architecture, the Mobile Grid is divided into       are static in nature. The user and the resource participating in
clusters. Each cluster has one cluster head (CH). A master server        problem solution are the two basic units of the processing
(MS) controls each local clusters and has frequent updates of all        environment. [2].
the CH information. Each CH has a monitoring agent (MA)
which will periodically predict the mobility of the cluster nodes            For many large scale applications which are dynamic in
and monitor the resource availability and update their values.           nature and require transparency for users, Mobile Grid is
When the MS forwards the job request of a user to the ideal CH,          considered as the best solution. Grid will increase the job
the CH schedules the jobs based on the predicted time for                throughput and performance of the corresponding applications
resource availability and sufficiency of the resources. By               by applying efficient mechanisms for resource management.
simulation results, we show that our proposed architecture               Moreover, it will enable the advanced forms of cooperative
achieves good success ratio and throughput with reduced delay            work by allowing the seamless integration of resources, data,
and energy consumption.                                                  services and ontologies [1]. Some of the applications of the
                                                                         mobile grids are scientific, public services and commercial
   Keywords-Resource Allocation, monitoring agent (MA),                  businesses. Mobile grids integrate the mobile devices like
Hierarchical Resource Allocation Architecture (HRAA), cluster
                                                                         laptops, PDAs (Personal Digital Assistants) [2].
head, master server (MS)
                                                                         B. Resource Allocation (or) Management in Mobile Grid
                         I. INTRODUCTION
                                                                             Resource allocation is a basic issue to achieve high
A. Mobile Grid                                                           performance on a grid workflow [3]. Resource allocation can
     The Grid is a distributed, high performing computing and            be classified into resource selection (discovery) and resource
data handling infrastructure. It provides common interfaces for          binding (acquiring). The resource selection is separated from
all the resources by using standard, open, general-purpose               resource binding based on the common architecture of
protocols and interfaces by incorporating the geographically             conventional resource brokering systems. These systems
and organizationally dispersed, heterogeneous resources. But             mainly concentrate on the resource selection for providing
it is the basis and the enabling technology for the persistent           complex resource specification languages and resource
and utility computing [1]. Multiple administrative domains,              selection algorithms. On providing a resource specification by
autonomy,          heterogeneity,          scalability      and          using available resource information, a resource selection
dynamicity/adaptability are the important features of the Grid.          algorithm first discovers a matching set of resources and
                                                                         negotiates with an individual local resource manager. Then the
   The mobility issues are handled by enabling both fixed and            application attempts to acquire the resources [4].
mobile users, in the mobile grid environment. By using the
underlying technologies transparently and efficiently, the               C. Resource Allocation challenges in Mobile Grid
access for both fixed and mobile grid resources are provided.               When mobility is considered, the resource selection
Mobile Grid is derived from Grid with the additional support             becomes more challenging. Therefore, it is necessary to
of mobile users and resources in a seamless, transparent,                consider the mobility of users with the resources in resource




                                                                     6                               http://sites.google.com/site/ijcsis/
                                                                                                     ISSN 1947-5500
                                                         (IJCSIS) International Journal of Computer Science and Information Security,
                                                         Vol. 8, No. 5, August 2010



selection. The mobility issue in grid environments has                  resources, move between WLANs. This willingness of the
established new challenges to the research communities                  mobile node is based on reciprocity criteria. They have also
particularly in the areas of scheduling, adaptation, security and       considered the divisible load applications (DLA) which
mobility. Mostly, the behavior of the user and/or mobile                divides the load of computation into parts and made to carry
device is highly unpredictable which produces disconnection             out independently. Moreover, they have described an
problems                                                                architecture for the realization of a Mobile Grid and
                                                                        investigate key design decisions and optimizations.
    In grid environments, new challenges are introduced to the
research communities especially in the fields of mobility,                  Lei Zhang et al [9] used the Particle Swarm Optimization
scheduling, adaptation and security due to issues like mobility,        (PSO), the latest evolutionary optimization technique to solve
power consumption and size of devices. The secondary                    the task scheduling problem in grid (computational grids)
problems are the small screen size and difficult input                  environment. Here each particle is represented as a possible
mechanisms. Peer-to-peer computing provides many useful                 solution, and the position vector is transformed from the
technologies and ideas, for developing scalable and reliable            continuous variable to the discrete variable. They have also
mobile grids. In mobile environment, the most challenging               aimed to generate an optimal schedule to get the minimum
problem is the disconnection problem [2].                               completion time while completing the tasks.
   The problems of resource allocation are:                                 Abdullah M. Elewi et al [10] have addressed the problem
                                                                        of energy efficient real time task scheduling where the tasks
   1) Identification of an appropriate service and resources.           are dependent due to exclusive access shared resources.
   2) Based on certain criteria, allocating the resources,              Moreover, they have proposed about the enhancements over
      such as pricing or priority.                                      the existing dual speed switching algorithm (DSA) where their
                                                                        proposed algorithm achieves more energy saving and has the
   3) Dynamically allocating and updating the state of the              capability to function with both SRP and DPCP protocols.
      resources [5].
                                                                           Homam Reda El-Taj et al [11] have given a survey about
    A centralized allocation manager is not possible because            mobile computing, the mobile grid computing, mobile agents
the portions of the Grid may apply different allocation                 and how to apply mobile agents on mobile grid computing and
strategies due to decentralized Grid policies. The lack of              what has been done to solve the issues in these areas of study.
accurate resource status information at the global scale is the
additional challenge to the Grid resource allocation. At their              Ming Wu et al [12] have proposed a prototype of Grid
removal, the knowledge of real time environment has been                Harvest Service (GHS) which provide dynamic and self-
limited by the allocation strategies which are utilized by the          adaptive task scheduling. Their study is made upon task
users and brokers. Therefore the possible allocation                    scheduling of a parallel or distributed application with a
mechanisms should not depend on the availability of current             divisible workload in a heterogeneous environment. It also
global knowledge [6].                                                   shows the possibility of integrating the three parts of task
                                                                        scheduling, that is the task allocator, scheduler and predictor
    In this paper, we develop a resource monitoring and                 into existing toolkits for better service.
scheduling scheme for mobile grid. In this scheme, the user
submits the job request for a job to be executed to a server.              Hesham A. Ali et al [13] have introduced a "self ranking
The job request contains job description, number of resources           algorithm", which will be used to build a mobile computing
required, expected job completion time and the quoted price             scheduling mechanism to schedule the tasks on the mobile
allotted for it.                                                        devices, which will maximize the profit of the mobile devices
                                                                        which are integrated within the grid using their computational
                       II. RELATED WORK                                 power as an addition to the system overall power.
    Xiaozhi Wang et al [7], have proposed a layered structure               Gurleen Kaur et al [14] have addressed the promising and
of Grid QoS. Based on the analysis of the content of grid               bright side of the grid computing technology. They have
resource allocation management (GRAM) based on QoS, their               explored the grid capabilities further by organizing the grid
work puts forward the architecture of GRAM based on QoS.                computing concept from two broad perspectives such as
Through mapping, converting and negotiating the QoS                     User’s Perspective and Administrator’s Perspective.
parameters, it can implant the user's requirement about QoS in
the process of resource allocation management, and connect                III. HIERARCHICAL RESOURCE ALLOCATION ARCHITECTURE
Grid QoS with GRAM very well. All these provide a
reasonable consulting model for QoS and resource allocation             A. System Design
management in grid. Their work raised the performance of                    In our system design, a set of machines and a cluster head
GRA from different aspects. In their process of searching,              (CH) are included in each local cluster. A master server (MS)
system may negotiate with the user, then get final result: not          controls and groups many local clusters. The MS collects
being able to supply, being able to supply or reducing QoS              information about the resources in its local clusters. Then it
demands to supply.                                                      stores the information in its own database.
   Konstantinos Katsaros et al [8] have discussed a campus-                The proposed scheme divides a given task that is submitted
wide hierarchical Mobile Grid system architecture in which              by a user into subtasks. Then it finds spare processors and
mobile nodes (MNs), willing to offer their computational                other critical resources on the network, distributes the



                                                                    7                               http://sites.google.com/site/ijcsis/
                                                                                                    ISSN 1947-5500
                                                          (IJCSIS) International Journal of Computer Science and Information Security,
                                                          Vol. 8, No. 5, August 2010



subtasks, monitors the progress of the subtasks, and restarts                Poweri is the power of the node ni and Jobsize j is the
the subtasks that fail.
                                                                         size of the Job j .
   It consists of the following steps:
                                                                         D. Scheduling Strategy
   1) The MS divides a given task into a sequence of
      subtasks and allocates the subtasks to the local cluster               After estimating the Ptime and WL values, the MA sends
      heads (CH).                                                        these values to its cluster head CH 1 . The CH 1 then schedules
   2) The CH finds the available processing power and                    the jobs if their grid node satisfies the following condition
      average mobility on the local machines and then
      distributes the subtasks over these machines.                                   If ( Ptime / WL ) > Th                                (4)
   3) The CHs gather the completed subtasks from the local
      machines and then send the data back to the MS.                        where Th is a threshold value (which can be fixed based
                                                                         on the job request).
   4) The MS aggregates the completed subtasks and then
      stores the results in its own database, and                            From 4, we can observe that if the Ptime is less and if the
                                                                         work load is more then the gird nodes are unable to execute
   5) MS also sends back the results to the application of               the job request. Therefore, the jobs are executed by the nodes
      the respective users.                                              only when Ptime is high and the work load is low. If the CH
B. Mobility Metric                                                       is unable to allocate the resources in its cluster, it resubmits
                                                                         the job request information to the MS.
    User Range is a given fact about the initiator coverage, in
which user can communicate with mobile devices. Average                  E. Functions of HRAA
Mobility, a derived parameter, represents the average mobility
of a resource and/or user (based on user and resource
mobility). Average mobility is calculated based on two recent
communications between user/initiator and resource with
respect to the user/initiator.
   Average Mobility is calculated as

             Average Mobility = f 1 − f 2                      (1)

    Where, f 1 and f 2 are the first history and second history
respectively. The history is simply the distance between user
and resource and it can be calculated by finding difference
between the two recent interactions.
    Ptime shows the predicted time for resource availability
within the user’s range and is calculated by the following                                   Figure 1. Functions of HRAA
equation
                                                                         The sequence of operations in HRAA is shown in Figure. 1. In
 Ptime = (User Range – Distance) / Average Mobility            (2)       this figure, the arrows represent the communication messages
                                                                         and the nodes represent the agents/servers. The sequence is as
   The “Distance” is the net difference between the new                  follows.
locations of user and resource.                                             1) The MAs of each node in the local cluster send the
                                                                               resource status information to the CHs.
C. Resource Availability Metric
    The monitoring agents (MA) estimates the workload of its                2) The CHs send this information to the MS.
grid nodes (ni , i = 1,2, L , k ) present in the cluster CL1 using          3) The MS then create a database which contains
the following formula:                                                         information about the status and the price of each
                                                                               resource.
                        k                                                 4) A user submits its job details and the resource
                                       
                CWLi +     ∑
                        j =1
                              Jobsize j 
                                        
                                                                               requirements to the MS.
                                                                          5) The MS sends the job request information to the local
          WLi =                                                (3)             CH.
                      Poweri
                                                                            6) The local CH allocates the resources in their control
   Where CWLi is the current workload of ni , WLi is the                       depending on the predicted time, the average
                                                                               processing power and average load of its local
work load of ni                                                                machines



                                                                     8                               http://sites.google.com/site/ijcsis/
                                                                                                     ISSN 1947-5500
                                                      (IJCSIS) International Journal of Computer Science and Information Security,
                                                      Vol. 8, No. 5, August 2010



   7) If any CH unable to allocate the resources in its              B. Performance Metrics
      cluster, it resubmits the job request information to the          In our experiments, we measure the following metrics.
      MS.
                                                                        Average Execution Delay: It measures the average delay
   8) The MS forwards the job request information to other
      CH. The process is continued until the job is                  occurred while executing a given task.
      successfully assigned                                              Average Success Ratio: It is the ratio of the number of
   9) The CHs gather the completed subtasks from the local           tasks executed successfully and the total number of tasks
      machines and then send the data back to the MS.                submitted.
   10) The MS aggregates the completed subtasks and then                 Average Energy: It is the average energy consumption of
       stores the results in it’s own database.                      all nodes in executing the tasks.
   11) MS also sends back the results to the application of             Throughput: It is the number of tasks finished
       the respective users.                                         successfully.
                   IV. SIMULATION RESULTS                            C. Results
                                                                     A. Based on Rate
A. Simulation Model and Parameters
                                                                        In this experiment, we vary the execution rate as 250Kb,
                                                                     500Kb, 750Kb and 1000Kb.

                                                                                                           Rate Vs Delay

                                                                                           10
                                                                                            8
                                                                                            6




                                                                            Delay
                                                                                                                                             ARAM
                                                                                            4                                                HRAA
                                                                                            2
                                                                                            0
                                                                                                 250      500       750      1000
                                                                                                           Rate (Kb)


                                                                                                        Figure 3. Rate Vs Delay

                    Figure 2. Simulation Setup
                                                                                                       Rate Vs SuccessRatio
    In this section, we examine the performance of our
Hierarchical Resource Allocation Architecture (HRAA) with                                   1
an extensive simulation study based upon the ns-2 network
                                                                            SuccessRatio




                                                                                           0.8
simulator [15]. The simulation topology is given in Figure 2.                              0.6                                               ARAM
We compare our results with Agent-based Resource Allocation                                0.4                                               HRAA
Model (ARAM) [5].Various simulation parameters are given in                                0.2
table 1.
                                                                                            0
                                                                                                 250      500        750     1000
                  TABLE I. SIMULATION SETTINGS
                                                                                                            Rate (Kb)
     Mobile Nodes              9
     Users                     3
                                                                                                   Figure 4.Rate Vs Success Ratio
     Clusters                  3
     Area Size                 1000 X 1000
     Mac                       802.11                                                                      Rate Vs Energy
     Radio Range               250m
     Routing Protocol          DSDV                                                        10

     Simulation Time           50 sec                                                       8
                                                                            Energy




     Traffic Source            CBR                                                          6                                                ARAM
     Packet Size               512                                                          4                                                HRAA
     Rate                      250kb,500kb,….1000kb                                         2
     No. Of tasks              2,4,6,8 and 10                                               0
     Speed                     5m/s                                                              250      500       750      1000
     Transmit Power            0.660 w                                                                     Rate (Kb)
     Receiving Power           0.395 w
     Idle Power                0.335 w
     Initial Energy            10.1 J                                                                   Figure 5.Rate Vs Energy




                                                                 9                                              http://sites.google.com/site/ijcsis/
                                                                                                                ISSN 1947-5500
                                                                             (IJCSIS) International Journal of Computer Science and Information Security,
                                                                             Vol. 8, No. 5, August 2010




                                              Rate Vs Throughput                                                                  No. of Tasks Vs SucessRatio

                           3000                                                                                         1
       Throughput(pkts.)


                           2500                                                                                        0.8




                                                                                                   SuccessRatio
                           2000                                                                                        0.6
                                                                          ARAM                                                                                                   ARAM
                           1500
                                                                          HRAA                                         0.4                                                       HRAA
                           1000
                                                                                                                       0.2
                           500
                                0                                                                                       0
                                        250        500       750   1000                                                           2         4         6      8     10

                                                   Rate (Kb)                                                                                        Tasks


                                         Figure 6.Rate Vs Throughput                                                        Figure 8. Number of Tasks Vs Success Ratio

    From Figure 3, it is clear that when the execution rate is
                                                                                                                                          No. Of Tasks Vs Energy
increased then the delay also increases. We can see that the
average execution delay of the proposed HRAA algorithm is
less when compared to the ARAM algorithm when the rate is                                                              12
increased.                                                                                                             10




                                                                                                   Energy(J)
                                                                                                                        8
   Figure 4 shows that when the execution rate is increased                                                                                                                      ARAM
then the success ratio gets decreased. From the figure we can                                                           6
                                                                                                                                                                                 HRAA
see that the HRAA achieves good success ratio, compared to                                                              4
ARAM.                                                                                                                   2
                                                                                                                        0
    Figure 5 shows that when the execution rate is increased
                                                                                                                                 2          4         6      8     10
then the energy consumption gets increased slightly. From the
results, we can see that HRAA consumes less energy than the                                                                                         Tasks
ARAM.
    Figure 6 shows that when the execution rate is increased                                                                     Figure 9. Number of Tasks Vs Energy
then the throughput is also increased. As we can see from the
figure, the throughput is more in the case of HRAA when                                                                               No. Of Tasks Vs Throughput
compared to ARAM.
                                                                                                                       3500
B. Based on Number of Tasks
                                                                                                                       3000
                                                                                                   Throughput(pkts.)




   In this experiment, we vary the number of tasks to be
                                                                                                                       2500
executed as 2, 4….10.
                                                                                                                       2000                                                      ARAM
                                                                                                                       1500                                                      HRAA
                                              No. Of Tasks Vs Delay
                                                                                                                       1000
                                                                                                                        500
                           14                                                                                                0
                           12                                                                                                         2         4      6     8     10
                           10                                                                                                                       Tasks
       Delay(s)




                           8                                              ARAM
                           6                                              HRAA                                               Figure 10. Number of Tasks Vs Throughput
                           4
                           2                                                                   From Figure 7, it is clear that when the number of tasks is
                           0                                                               increased then the delay also gets increased. We can see that
                                    2          4         6     8    10                     the average execution delay of the proposed HRAA algorithm
                                                   Tasks                                   is less when compared to the ARAM algorithm when the
                                                                                           number of tasks is increased.

                                    Figure 7. Number of Tasks Vs Delay                        Figure 8 shows that when the number of tasks is increased
                                                                                           then the success ratio gets decreased. From the figure we can
                                                                                           see that the HRAA achieves good success ratio, compared to
                                                                                           ARAM.
                                                                                              Figure 9 shows that when the number of tasks is increased
                                                                                           then the energy consumption gets increased slightly. From the




                                                                                      10                                                            http://sites.google.com/site/ijcsis/
                                                                                                                                                    ISSN 1947-5500
                                                               (IJCSIS) International Journal of Computer Science and Information Security,
                                                               Vol. 8, No. 5, August 2010



results, we can see that HRAA consumes less energy than the                     [7] Xiaozhi Wang, Junzhou Luo, “Architecture of Grid Resource Allocation
ARAM.                                                                                Management Based on QoS”, Springerlink – 2004.
                                                                                [8] Konstantinos Katsaros and George C. Polyzos , “Optimizing Operation
    Figure 10 shows that when the number of tasks is                                 Of A Hierarchical Campus-Wide Mobile Grid”, The 18th Annual IEEE
increased then the throughput is also increased. As we can see                       International Symposium on Personal, Indoor and Mobile Radio
from the figure, the throughput is more in the case of HRAA                          Communications (PIMRC’07)
when compared to ARAM.                                                          [9] Lei Zhang, Yuehui Chen, Runyuan Sun, Shan Jing and Bo Yang, “A
                                                                                     Task Scheduling Algorithm Based on PSO for Grid Computing ”,
                                                                                     International Journal of Computational Intelligence Research. (2008),
                            V. CONCLUSION
                                                                                [10] Abdullah M. Elewi, Medhat H. A. Awadalla and Mohamed I. Eladawy ,
    In this paper, we have designed a Hierarchical Resource                          “Energy Efficient Real Time Scheduling Of Dependent Tasks Sharing
Allocation Architecture (HRAA) which includes monitoring                             Resources” Proceedings of the High Performance Computing &
and scheduling operations for mobile grid. In this architecture,                     Simulation Conference, 2008.
a set of machines and a cluster head (CH) are included in each                  [11] Homam Reda El-Taj and Chan Huah Yong, “Applying Mobile Agents
                                                                                     on Mobile Grid Computing”, Computer Science Postgraduate
local cluster. A master server (MS) controls and groups many                         Colloquium (CSPC’07).
local clusters. The MS collects the information about the
                                                                                [12] Ming Wu and Xian-He Sun, “A General Self-adaptive Task Scheduling
resources in its local clusters and stores it in its own database.                   System for Non-dedicated Heterogeneous Computing”, Proceedings of
Each CH has a monitoring agent (MA) which will periodically                          IEEE International Conference on Cluster Computing, 2003.
predict the mobility of the cluster nodes and monitor the                       [13] Hesham A. Ali and Tamer Ahmed Farrag, “High Performance Mobile
resource availability and update their values. The MS forwards                       Computing Algorithm Based On Self-Ranking Algorithm (Sar)”,
the job request of a user to the ideal CH. If the CH finds the                       Department of Computers and Systems, Faculty of Engineering,
available processing power, workload and predicted time for                          Mansoura University, Egypt, Department of Computers and Systems,
                                                                                     Faculty of Engineering, Mansoura University, Egypt
resource availability on the local machines, then it distributes
                                                                                [14] Gurleen Kaur and Inderpreet Chopra, “Grid Computing – Challenges
the subtasks over these machines. Otherwise the job request is                       Confronted and Opportunities Offered”, Proceedings of Challenges and
resubmitted to the MS which again forwards the job request to                        Opportunities in Information Technology, COIT-2007.
another CH. The process is continued until the job is                           [15] Network Simulator: http:// www.isi.edu/nsnam/ns
successfully assigned. The completed job is returned back to
the MS through the corresponding CH. The MS then returns it
to the requested user. By simulation results, we have shown                                      S. Thenmozhi received MCA degree from Annamalai
that our proposed protocol achieves good success ratio and                                       University, India in 2000 and M.Phil degree in computer
throughput with the reduced delay and energy consumption.                                        science from Bhrathiar University, India in 2003. Currently
                                                                                                 she is pursuing the Ph.D from Anna University
                              REFERENCES                                                         Coimbatore, India. Her area of interest in research includes
                                                                                                 Grid Computing and Mobile Computing. She has published
[1] Litke, A., Skoutas D. and Varvarigou T., “Mobile Grid Computing:
                                                                                                 6 papers in National/International Conferences/Journals.
    Changes and Challenges of Resource Management in a Μobile Grid
                                                                                                 Presently, she is working as Assistant Professor in
    Environment”, presented in Workshop: “Access to Knowledge through
    Grid in a Mobile World”, PAKM 2004 Conference, Vienna                       Department of Computer Applications, Chettinad College of Engineering and
                                                                                Technology, Tamilnadu, India.
[2] Umar Farooq, Saeed Mahfooz and Wajeeha Khalil, “An Efficient
    Resource Prediction Model for Mobile Grid Environments”, 2006- PG
    Net.                                                                                      A.Tamilarasi post graduated from Bharathiar University,
                                                                                              India 1986. She obtained her Ph.D from University of
[3] Pengcheng Xiong, Yushun Fan, “Cost-aware Grid Workflow Resource                           Madras, Chennai in the year 1994. She was awarded JRF by
    Allocation”, IEEE Computer Society, 2007.                                                 UGC in the year 1986. She has published more than 40
[4] Yang-Suk Kee, Ken Yocum, Andrew A. Chien, Henri Casanova,                                 research papers in the reputed national/international
    “Improving Grid Resource Allocation via Integrated Selection and                          Journals. She is author of 10 books. Her area of interest
    Binding”, 2006 IEEE                                                                       includes Semigroup theory, Soft computing, Data Mining.
[5] S.S. Manvia, M.N. Birjeb, Bhanu Prasad, “An Agent-based Resource                          Presently she is working as a Professor Department of
    Allocation Model for computational grids”, IOS Press Amsterdam, The         MCA, Kongu Engineering College, Erode, Tamilnadu, India
    Netherlands, 2005.
[6] Aram Galstyan, Karl Czajkowski,Kristina Lerman, “Resource Allocation
    in the Grid Using Reinforcement Learning”, IEEE Computer Society
    2004.




                                                                           11                                    http://sites.google.com/site/ijcsis/
                                                                                                                 ISSN 1947-5500