An efficient scheduling policy for load balancing model for computational grid system

Computer Engineering and Intelligent Systems www.iiste.org
ISSN 2222-1719 (Paper) ISSN 2222-2863 (Online)
Vol 3, No.7, 2012

An Efficient Scheduling Policy for Load Balancing Model for
Computational Grid System
Mukul Pathak1 Ajeet Kumar Bhartee2 Vinay Tandon3
1, 2.Department of Computer Science & Engineering, Galgotias College of Engineering & Technology, Greater
Noida (U.P.), India
1mukul.pathak2501@gmail.com
2ajeetkbharti@gmail.com
3.Department of Master of Computer Application, Aligarh College of Engineering & Technology, Greater Noida
(U.P.), India
3vnay.tandon@gmail.com
Abstract
Workload and resource management are two essential functions provided at the service level of the Grid system. To
improvement in global throughput need, effective and efficient load balancing are fundamentally important. We also
check that what type of scheduling policy is used by that algorithm, because an efficient scheduling policy can utilize
the computational resources efficiently by allowing multiple independent jobs to run over a network of
heterogeneous clusters. In this paper, a dynamic grid model, as a collection of clusters has been proposed. An
efficient scheduling policy is used, and its comparison with the other scheduling policy has been presented.

1. INTRODUCTION. In order to fulfil the user expectations in terms of performance and efficiency, the Grid
system needs efficient load balancing algorithms for the distribution of tasks [1]. A load balancing algorithm
attempts to improve the response time of user’s submitted applications by ensuring maximal utilization of available
resources. The main goal is to prevent, if possible, the condition where some processors are overloaded with a set of
tasks while others are lightly loaded or even idle [2]. Although load balancing problem in conventional distributed
systems has been intensively studied, new challenges in Grid computing still make it an interesting topic and many
research projects are under way. This is due to the characteristics of Grid computing and the complex nature of the
problem itself. Load balancing algorithms in classical distributed systems, which usually run on homogeneous and
dedicated resources, cannot work well in the Grid architectures. In this chapter we define the motivation of this
research and then identify the research questions. This chapter also discusses overall organization of thesis.

2. CHARACTERISTICS OF GRID

There are these main issues that characterize computational Grids [3, 4]:

• Heterogeneity: A Grid involves a multiplicity of resources that are heterogeneous in nature and might span
numerous administrative domains across wide geographical distances.

Resources are heterogeneous
Resources are administratively disparate
Resources are geographically disparate
Users do not have to worry about system details (e.g., location, operating system, accounts).
Resources are numerous.
Resources have different resource management policies.
Resources are owned and managed by different, potentially mutually distrustful organizations and
individuals that likely have different security policies and practices.

51

Vol 3, No.7, 2012

• Scalability: A Grid might grow from few resources to millions. This raises the problem of potential performance
degradation as a Grids size increases. Consequently, applications that require a large number of geographically
located resources must be designed to be extremely latency tolerant.

• Dynamicity or Adaptability: In a Grid, a resource failure is the rule, not the exception [5]. In fact, with so many
resources in a Grid, the probability of some resource failing is naturally high. The resource managers or applications
must tailor their behaviour dynamically so as to extract the maximum performance from the available resources and
services.

• Parallel CPU execution: One of most important feature of Grid is its scope for massive parallel CPU capacity. The
common attribute among such uses is that the applications have been written to use algorithms that can be partitioned
into independently running parts. A CPU intensive Grid application can be thought of as many smaller “sub jobs,”
each executing on a different machine in the Grid.[5] To the extent that these sub jobs do not need to communicate
with each other, the more “scalable” the application becomes. A perfectly scalable application will, for example,
finish 10 times faster if it uses 10 times the number of processors.

• Virtual organizations: The users of the Grid can be organized dynamically into a number of virtual organizations,
each with different policy requirements [6, 7]. These virtual organizations can share their resources collectively as a
larger Grid.

• Resource balancing: A Grid contains a large number of resources contributed by individual machines into a greater
total virtual resource. For applications that are Grid enabled, the Grid can offer a resource balancing effect by
scheduling Grid jobs on machines with low utilization [8].

• Reliability and Management: High-end conventional computing systems use expensive hardware to increase
reliability. A Grid is an alternate approach to reliability that relies more on software technology than expensive
hardware [51]. The goal to virtualized the resources on the Grid and more uniformly handle heterogeneous systems
will create new
Opportunities to better manage a larger more disperse IT infrastructure.

3. GRID ARCHITECTURE

Architecture identifies the fundamental system components, specifies purpose and function of these components, and
indicates how these components interact with each other. Grid architecture is protocol architecture, with protocols
defining the basic mechanisms by which VO [9, 10, and 11] users and resources negotiate, establish, manage and
exploit sharing relationships. Grid architecture is also a services standards based open architecture that facilitates
extensibility, interoperability, portability and code sharing. The components that is necessary to form a Grid.

Grid Fabric: It comprises all the resources geographically distributed (across the globe) and accessible from
anywhere on the Internet. They could be computers (such as PCs or Workstations running operating systems such as
UNIX or NT), clusters (running cluster operating systems or resource management systems such as LSF, Condor or
PBS), storage devices, databases, and special scientific instruments such as a radio telescope.

Grid Middleware: It offers core services such as remote process management, collocation of resources, storage
access, information (registry), security, authentication, and Quality of Service (QoS) such as resource reservation and
trading.

Grid Development Environments and Tools: These offer high-level services that allow programmers to develop
applications and brokers that act as user agents that can manage or schedule computations across global resources.

52

Vol 3, No.7, 2012

Grid Applications and Portals: They are developed using Grid-enabled languages such as HPC++, and
message-passing systems such as MPI. Applications, such as parameter simulations and grand-Challenge problems
often require considerable computational power, require access to remote data sets, and may need to interact with
scientific instruments. Grid portals offer web-enabled application services i.e., users can submit and collect results
for their jobs on remote resources through a web interface.

4 LOAD BALANCING APPROACHES

Load balancing problem has been discussed in traditional distributed systems literature for more than two decades.
Various algorithms, strategies and policies have been proposed, implemented and classified [12].

4.1 STATIC LOAD BALANCING ALGORITHM

Static load balancing algorithms allocate the tasks of a parallel program to workstations based on either the load at
the time nodes are allocated to some task, or based on an average load of our workstation cluster. The decisions
related to load balance are made at compile time when resource requirements are estimated. The advantage in this
sort of algorithm is the simplicity in terms of both implementation as well as overhead, since there is no need to
constantly monitor the workstations for performance statistics.

4.2 DYNAMIC LOAD BALANCING ALGORITHM

Dynamic load balancing algorithms make changes to the distribution of work among workstations at run-time; they
use current or recent load information when making distribution decisions. Multicomputers with dynamic load
balancing allocate/ reallocate resources at runtime based on no a priori task information, which may determine when
and whose tasks can be migrated.

5. LOAD BALANCING STRATEGIES

There are two major strategies which usually used in load balancing algorithm will employ [17].

SENDER-INITIATED VS RECEIVER-INITIATED STRATEGIES

The question of who makes the load balancing decision is answered based on whether a sender-initiated or
receiver-initiated policy is employed [13]. In sender- initiated policies, congested nodes attempt to move work to
lightly-loaded nodes. In receiver-initiated policies, lightly-loaded nodes look for heavily-loaded nodes from which
work may be received.

Figure shows the relative performance of a sender-initiated and receiver-initiated load balancing algorithm. As can
be seen, both the sender-initiated and receiver-initiated policies perform substantially better than a system which has
no load sharing.

The sender-initiated policy performing better than the receiver-initiated policy at low to moderate system loads.
Reasons are that at these loads, the probability of finding a lightly loaded node is higher than that of finding a
heavily-loaded node. Similarly, at high system loads, the receiver initiated policy performs better since it is much
easier to find a heavily loaded node. As a result, adaptive policies have been proposed which behave like sender
initiated policies at low to moderate system loads, while at high system loads they behave like receiver-initiated
policies.

53

Vol 3, No.7, 2012

6. LOAD BALANCING POLICIES

Load balancing algorithms can be defined by their implementation of the following policies [14-15]:
• Information policy: specifies what workload information to be collected, when it is to be collected and
from where.
• Triggering policy: determines the appropriate period to start a load balancing operation.
• Resource type policy: classifies a resource as server or receiver of tasks according to its availability status.
• Location policy: uses the results of the resource type policy to find a suitable partner for a server or
receiver.
• Selection policy: defines the tasks that should be migrated from overloaded resources (source) to most idle
resources (receiver).

7. PROBLEM STATEMENT

In grid environments, the shared resources are dynamic in nature, which in turn affects application performance.
Workload and resource management are two essential functions provided at the service level of the Grid software
infrastructure. To improve the global throughput of these environments, effective and efficient load balancing
algorithms are fundamentally important. The focus of our study is to consider factors which can be used as
characteristics for decision making to initiate Load Balancing. Load Balancing is one of the most important factors
which can affect the performance of the grid application.
This thesis work analyzes the existing Load Balancing modules and tries to find out performance bottlenecks in it.
All Load Balancing algorithms implement five policies [16]. The efficient implementation of these policies decides
overall performance of Load Balancing algorithm.

The main objective of this paper is to propose an efficient Load Balancing Algorithm for Grid environment. Main
difference between existing Load Balancing algorithm and proposed Load Balancing is in implementation of
Scheduling policies: selected by Selection Policy. In my thesis we make a scheduling policy, which can be used
more reliably to make decision about selection of job for migration from heavily loaded node to lightly loaded node.

8. PROPOSED METHODOLOGY
Load balancing is defined as the allocation of the work of a single application to processors at run-time so that the
execution time of the application is minimized. This chapter is going to discuss the design of proposed Load
Balancing algorithm.

9. PROPOSED GRID MODEL COMPONENTS

Cluster-Level consists of a collection of computing nodes. Cluster-Level manager (CM) can fully control the
computing nodes within it, but cannot operate the computing nodes of other clusters directly. The computing nodes
within the cluster are referred as friends. CM maintains the load information along with registration information of
its computing nodes. In Cluster-Level, each friend runs a CM. CM role is to balance the intra-cluster workload. A
designated friend with highest CPU speed in each cluster is treated as the cluster server or master. Cluster System
Monitor (CS) determines the load index of computing nodes and provides this information to CM.

Grid-Level consists of a collection of interconnected clusters. Grid-Level manager (GM) is responsible for load
control among its clusters as shown in. GM maintains the load information along with registration information of
neighbouring masters in the grid. Neighbours for each cluster are formed in terms of communication costs. GM
calculates the minimum communication cost of sending or receiving jobs to/from remote clusters based on the
information collected in the last exchange interval. Master of each cluster also runs the GM. K denotes the number of
clusters in Grid-Level. Grid System Monitor (GS) determines the load index of masters and provides this information
to GM.

54

Vol 3, No.7, 2012

Client Interface provides a graphical user interface to the user for the submission of jobs. Scheduler is responsible for
scheduling of submitted jobs. Dispatcher performs the dispatching of jobs to other masters. Collector is in charge of
capturing jobs from other masters. Each neighbour of a cluster is responsible for completing the jobs assigned to
them by their master.
A decentralized job scheduling approach is used since the jobs generated by users are directly submitted to master.

Scheduler runs as a sub-component of CM

9. DESIGN OF LOAD BALANCING MODEL

Load balancing should take place when the load situation has changed. There are some particular activities which
change the load configuration in Grid environment. The activities can be categorized as following Arrival of any new
job and queuing of that job to any particular node.

• Completion of execution of any job.
• Arrival of any new resource
• Withdrawal of any existing resource.

Whenever any of these four activities happens activity is communicated to master node then load information is
collected and load balancing condition is checked. If load balancing condition is fulfilled then actual load balancing
activity is performed.

SI-LB algorithm gets the load information of cluster and communicating this information to GM via mutual
information feedback. Based on the load information SI-LB chooses most suitable node for each job, thereby
minimizing job execution time and maximizing system throughput. Load information, generally defined in term of
load index is necessary condition of SI-LB algorithm. The load at each computing node contributes to the overall
load of the cluster and can be determined as: CPU utilization, CPU speed and queue length. The load index is
determined dynamically and the weighted sum of squares method is used to calculate the load at each computing
node. Figure represents the logical structure of a processing cluster. Any cluster in the grid can be a processing
cluster.
In Cluster-Level load balancing, depending on the current workload of its associated cluster, estimated from its own
friends in NLIST, each Cluster-Level manager (CM) decides whether to start or not a load balancing operation. If it
decides to start a load balancing operation, then it tries to load balance the workload among its under-loaded friends
in NLIST. If any friend in the cluster at any instant of time is under loaded or over-loaded, it requests or allots jobs
to/from other friends in (NLIST) with minimal load using the symmetric initiated approach to load balancing. If the
master is unable to load balance the workload among its friends, then the jobs are transferred to the under-loaded
masters in (CLIST) with minimum load and minimum communication delay.
In Grid-Level load balancing, the load balancing is performed only if CM fails to load balance their workload among
their associated friends. In this case, jobs of overloaded clusters are transferred to under loaded ones in (CLIST)
regarding the minimum communication delay and load. The chosen under loaded clusters are those which need
minimal communication delay for transferring jobs from overloaded clusters.

10. IMPLEMENTATION DETAILS AND EXPERIMENTAL RESULTS
To implement the proposed Load Balancing Algorithm, an application has been developed, which is executed in
simulated grid environment. The application has been developed using J2EE and Alea 3 simulator.

EXPERIMENTAL RESULTS
In this section we show the performance of the Alea 3 simulator through several experiments. All experiments were
performed on Intel Core I3 2.27GHz Laptop with 3GB of RAM. Unless otherwise indicated, the JVM (Java Virtual
Machine) was limited by 1GB of available RAM.

55

Vol 3, No.7, 2012

The experiment involved 103,656 jobs 14 clusters having 806 CPUs.
Through the experiment, we have compared three different scheduling algorithms: FCFS, EASY-BF and
Random+CONS scheduling policy. Figure presents graphs depicting the average machine usage per cluster (left) and
the number of waiting and running jobs per day (right) as were generated by the Alea 3 during the experiment. These
graphs nicely demonstrate major differences among the algorithms. Concerning the machine usage as expected FCFS
generates very poor results. FCFS is not able to utilize available resources when the first job in the queue requires
some specific and currently unavailable machine(s). At this point, other more flexible" jobs in the queue can be
executed increasing the machine utilization. This is the main goal of the EASYBF algorithm. As we can see,
EASY-BF is able to increase the machine usage by using the backfilling approach. Still, EASY-BF does not allow
delaying the execution start of the first job in the queue, which restricts it from making more aggressive decisions
that would increase the utilization even more. Random+CONS algorithm does not apply such restrictions and thanks
to the application of a more efficient schedule-based approach it produces the best results.

In case of the second criteria, similar reasons as in the previous example caused that FCFS is not able to schedule
jobs fluently, generating huge peak of waiting jobs during the time. For the same reason, the resulting make span is
also much higher than in the remaining algorithms. EASY-BF is capable of a higher resource utilization and
reduction of the number of waiting jobs through the time. ESG again produces the best results. As can be seen, these
and several other graphical outputs such as those presented in Figure help the user to understand and compare the
scheduling process of different scheduling algorithms.

11. CONCLUSION AND SCOPE OF FUTURE WORK
Every Load Balancing algorithm implements five policies. The efficient implementation of these policies decides
overall performance of Load Balancing algorithm. In this work we analyzed existing Load Balancing algorithm and
proposed an enhanced algorithm which more efficiently implements three out of five policies implemented in
existing Load Balancing algorithm. These three policies are: Information Policy, Triggering Policy and Selection
Policy. Proposed algorithm is executed in simulated Grid environment.

12. FUTURE DIRECTIONS

• More complex models such as nesting of clusters need to be investigated.
• Additional factors like network bandwidth that may affect the performance of the algorithm need to be
studied.
• Experiments could be tried in a real environment.

REFERENCE

[1] Krishnaram Kenthapadi, Stanford University , kngk@cs.stanford.edu and
Gurmeet Singh Mankuy , Google Inc., manku@google.com,Decentralized Algorithms using both Local and Random
Probes for P2P Load Balancing.

[2] B. Yagoubi , Department of Computer Science, Faculty of Sciences, University of Oran and Y. Slimani ,
Department of Computer Science, Faculty of Sciences of Tunis, Task Load Balancing Strategy for Grid Computing .

[3] Hans-Ulrich Heiss and Michael Schmitz, Decentralized Dynamic Load Balancing: The Particles Approach.

[4]Junwei Cao1, Daniel P. Spooner, Stephen A. Jarvis, and Graham R. Nudd, Grid Load Balancing Using Intelligent
Agents.

[5]Ian Foster, Argonne National Laboratory & University of Chicago, What is the Grid? A Three Point Checklist.

56

Vol 3, No.7, 2012

[6]Jennifer M. Schopf, Mathematics and Computer Science Division, Argonne National Lab, Department of
Computer Science, Northwestern University, Grids: The Top Ten Questions.

[7]Karl Czajkowski, Ian Foster and Carl Kesselman, Resource Co-Allocation in Computational Grids.

[8]Ann Chervenak, Ian Foster, Carl Kesselman, Charles Salisbury and Steven Tuecke, The Data Grid:Towards an
Architecture for the Distributed Management and Analysis of Large Scientific Datasets

[9]Foster, I., C. Kesselman, and S. Tuecke, “The Anatomy of the Grid: Enabling Scalable Virtual Organizations”.
International Journal of Supercomputer Applications, 2001.

[10]Jean-Christophe Durand, “Grid Computing a Conceptual and Practical Study”, November 8, 2004

[11]Clovis Chapman1, Paul Wilson2, “Condor services for the Global Grid: Interoperability between Condor and
OGSA”, Proceedings of the 2004 UK e-Science All Hands Meeting, ISBN 1-904425-21-6, pages 870-877,
Nottingham, UK, August 2004 http://www.cs.wisc.edu/condor/doc/condor-ogsa-2004.pdf

[12]Javier Bustos Jimenez, Robin Hood: An Active Objects Load Balancing Mechanism for Intranet.

[13]Shahzad Malik, Dynamic Load Balancing in a Network of Workstations, 95.515F Research Report, November
29, 2000.

[14]Menno Dobber, Ger Koole, and Rob van der Mei, Dynamic Load Balancing for a Grid Application,
http://www.cs.vu.nl/~amdobber

[15]Guy Bernard, A Decentralized and Efficient Algorithm for Load Sharing in Networks of Workstations. [51] D.
Klus_a_cek, L. Matyska, and H. Rudov_a. Alea { Grid scheduling simulation environment. In 7th International
Conference on Parallel Processing and Applied Mathematics (PPAM 2007), volume 4967 of LNCS, pages
1029{1038. Springer, 2008}}.

[16] Francois Grey, Matti Heikkurinen, Rosy Mondardini, Robindra Prabhu, “Brief History of Grid”,
http://Gridcafe.web.cern.ch/Gridcafe/Gridhistory/history.html.

[17] Gregor von laszewaski, Ian Foster, Argonne National Laboratory, Designing Grid Based Problem solving
Environments www-unix.mcs.anl.gov/~laszewsk/papers/cogpse- final.pdf.
Mukul Pathak

The author is pursuing Post Graduation in engineering from Galgotias College of
Engineering & Technology, Greater Noida (U.P.). He had completed engineering
from College of Engineering & Technology; Moradabad (U. P.) affiliated to
Gautam Buddh Technical University in 2010.

57

Vol 3, No.7, 2012

Ajeet Kumar Bhartee

The author currently working in Galgotias College of Engineering & Technology,
Technology, Greater Noida (U.P.), and have 9+ year experience. He had
completed engineering from Madan Mohan Malvia Engineering College;
Gorakhpur (U. P.) affiliated to Uttar Pradesh Technical University in 2001 and
completed his Masters from CDAC Noida (U.P) in 2007.

Vnay Tandon

The author currently working in Aligarh college of engineering and technology,
Aligarh (U.P.), and have 13 year experience. He had completed Master in
Computer Application; and Pursuing his M-Tech.

Figure-1: Grid Architecture

58

Vol 3, No.7, 2012

Figure-2: Static Load Balancing [12]

Figure 3: Dynamic Load Balancing [12]

Figure 4: System Utilization of Sender initiated, receiver initiated
and No load

59

Vol 3, No.7, 2012

Figure 5: Design of Load Balancing Model

Figure 6: Logical Structure of Processing Cluster

60

Vol 3, No.7, 2012

Figure 7: FCFS Scheduling Result

Figure 8: EASY-BF Scheduling Result

Figure 9: CONS+RANDOM Scheduling Result

61

This academic article was published by The International Institute for Science,
Technology and Education (IISTE). The IISTE is a pioneer in the Open Access
Publishing service based in the U.S. and Europe. The aim of the institute is
Accelerating Global Knowledge Sharing.

More information about the publisher can be found in the IISTE’s homepage:
http://www.iiste.org

The IISTE is currently hosting more than 30 peer-reviewed academic journals and
collaborating with academic institutions around the world. Prospective authors of
IISTE journals can find the submission instruction on the following page:
http://www.iiste.org/Journals/

The IISTE editorial team promises to the review and publish all the qualified
submissions in a fast manner. All the journals articles are available online to the
readers all over the world without financial, legal, or technical barriers other than
those inseparable from gaining access to the internet itself. Printed version of the
journals is also available upon request of readers and authors.

IISTE Knowledge Sharing Partners

EBSCO, Index Copernicus, Ulrich's Periodicals Directory, JournalTOCS, PKP Open
Archives Harvester, Bielefeld Academic Search Engine, Elektronische
Zeitschriftenbibliothek EZB, Open J-Gate, OCLC WorldCat, Universe Digtial
Library , NewJour, Google Scholar

An efficient scheduling policy for load balancing model for computational grid system

Recommended

More Related Content

What's hot (20)

Viewers also liked (13)

Similar to An efficient scheduling policy for load balancing model for computational grid system (20)

More from Alexander Decker (20)

Recently uploaded (20)

An efficient scheduling policy for load balancing model for computational grid system