Use of genetic algorithm for

International Journal on Information Theory (IJIT), Vol.4, No.3, July 2015
DOI: 10.5121/ijit.2015.4302 1
USE OF GENETIC ALGORITHM FOR
BALANCING THE GRID LOAD
Saad Masood Butt1
, Moaz Masood Butt2
and Azura Onn3
12
Computer and Software Engineering Department, Bahria University Islamabad,
Pakistan
saadmasoodbutt668@yahoo.com
moazbutt786@hotmail.com
3
Department of Management and Human Resource, Universiti Tenaga National,
Malaysia
azura@uniten.edu.my
ABSTRACT
Grid computing can involve lot of computational tasks which requires trustworthy computational nodes.
Load balancing in grid computing is a technique which overall optimizes the whole process of assigning
computational tasks to processing nodes. Grid computing is a form of distributed computing but different
from conventional distributed computing in a manner that it tends to be heterogeneous, more loosely
coupled and dispersed geographically. Optimization of this process must contains the overall
maximization of resources utilization with balance load on each processing unit and also by decreasing
the overall time or output. Evolutionary algorithms like genetic algorithms have studied so far for the
implementation of load balancing across the grid networks. But problem with these genetic algorithm is
that they are quite slow in cases where large number of tasks needs to be processed. In this paper we give
a novel approach of parallel genetic algorithms for enhancing the overall performance and optimization
of managing the whole process of load balancing across the grid nodes.
KEYWORDS
Network Protocols, Grid Computing, Load Balancing, Distributed Computing, Bubble Sort, 2-element
Insertion Sort
1. INTRODUCTION
With the advent of distributed computing we encounter large number issues regarding
processing powers and load balancing across the distributed networks specially like grid
computing. In order to get an optimized results without any time delay and with maximum
throughput we need enhanced algorithms across the networks. Hence this faster processing
results can be obtained by introducing a single processing unit or it can be inflated to diversity
of processing types like multiple processing nodes, geographically distributed processing units
and parallel processing. Grid computing is a modern and diverse form of distributed computing.
It focused on the ability to support computation across multiple administrative domains that sets
it apart from traditional distributed computing. Grids presents a method of using the
computational resources optimally within an association involving different computing
resources. Its supports multiple administrative domains and security authentication and other
organizational mechanisms that capable it to be distributed locally or dispersed on the multiple
geographic location in form of wide area network. Within the network connected nodes can
share their resources over a network domain. These resources can be utilized by any other node
that is connects to the grid network. I we ponder over this fact then we consider resource sharing
across the grid computing network is an important factor over the internet networks (Neil Y,
2012).
Multiple sharing resources are accumulated across the grid network. These resources involves
number of dimensions including computational powers, processing capabilities. In grid

2
computing environment all computational nodes are connected in such a manner that one node
across the network is used for the admittance point for the whole set of resources which are
physically dispersed across the network [1]. So on the whole it seems for every user that their
computer has powers analogous to super computers.
This assorted nature of gird has numerous challenges like resource management of multiple
resources like storage capacity and processing powers, hardware and software based
heterogeneity, multiple security issues while connecting to different administrative domain.
When connecting to the grid. So on the whole the multi folded nature of grid has generated
some stern concerns needed to be considered. Load balancing across the network manages the
concept of load across the network in a manner that no node is over loaded or under loaded
while the resources are used. These load balancing algorithms deal with the overall optimization
of grid network in terms of resource sharing.
Evolutionary algorithms like genetic algorithms are from the family of heuristic techniques
which are in used from long time for finding optimized solutions. These algorithms incorporates
the mechanism inspired from biological evolution [2]. There are many variants of genetic
algorithm including different mutation and cross over procedure. Parallel genetic algorithms are
quite new in this perspective but different from conventional genetic algorithm methodologies
as they are faster and efficient in processing and gave more optimized results. There are many
variants of parallel genetic algorithm including master slave model, coarse grain genetic
algorithm, and dual species genetic algorithm
2. RELATED WORK
There is lot of examples in literature which deals with parallel processing in distributed
environment. In order to design well managed two-way applications writer has proposed the
infrastructure named as middleware. This paper gives idea on a middleware concept and figure
out the problems of service discovery, organization of communication between devices,
harmonization, data-security and minimization of communication between the devices within
distributed and collaborative environments [3].
Distributed systems along with the concept of P2P systems is also a form of computer networks.
By considering the strategies of mutual understanding and trust between the peers in P2P
systems writer has proposed that by using different protocols set of reliable peers can be
generated [4]. Schedulers based on genetic algorithms are quite efficient for jobs allocation to
different computing resources in a grid environment. Author proposes a widespread study on the
utilization of genetic algorithm for designing resourceful grid based schedulers for overall
minimization of total completion time. Two schemes based on encoding has been focused and
the majority of genetic algorithm operators for each of these schemes are implemented and
logically studied. Dealing with grid infrastructure at its service level generates two important
issues which needs to be managed, these issues includes management of workloads and resource
management in an optimized manner. Using neighbourhood property and focusing on the tree
based techniques generated optimized results in for managing load balancing across the grid
networks [5]. Genetic algorithms to resolve the predicament of load balancing resourcefully.
Proposed algorithm considers multi purposes in its resolution of assessment and resolves the
scheduling problem in a way that concomitantly minimizes maxspan and overall decreases
communication cost, and capitalize the processors efficiency and utilization [6]. Proposed
genetic algorithm make use of dynamic load-balancing methodology for solving issue of
resource management across the networks Algorithm generated high-quality results when size
of tasks to be scheduled become bigger. Implementation of central scheduler in grid
environment overall optimized the scheduling process by considering the concept of least
communication for making load balancing decisions along the grid networks [7].

3
[8] proposed approach to solves the task scheduling for grid local search algorithm based on the
concept of gravitational attraction search is incorporated with genetic algorithm in order to
enhance its capacity to search more sharply in problem based search space and attain more
precise response in optimized time. Comparison between the results o HYGAGA and genetic
algorithm emphasizes major improvement in the recital of search algorithm[9][10].
2.1. Proposed Methodology
Grid Model:
We have taken in account the grid information system model for solving the issue o load
balancing across the grid nodes. GIA is a grid model provides the information regarding the
number of resources, number of tasks and state information of resources along with the
information that which task is being served by which resource across the grid network. This
model is good enough to use in the place where some kind of heuristics is being implemented.
We enhance this model by adding the concept of parallel genetic algorithm. We consider that
there is a scheduler present in the grid. This scheduler according to GIS model has all the
information related to the number o task. Let us consider that set of task come to the scheduler
in GIS model of grid. These tasks are selected on the basis o criteria we define based on the total
completion time of tasks. We select those tasks first whose completion time is approaching to
zero
Tct 0
Suppose a new array of tasks based on their completion time reaches to scheduler for
scheduling. The central scheduler has all the information about the current state of the system.
At this point of time we mainly focuses on the selection of optimized set of tasks for assigning
to multiple resources for processing. In order to find the optimized set of tasks we use the
concept of parallel genetic algorithm. Parallel genetic algorithm is quite efficient as compare to
conventional genetic algorithm techniques for finding optimized solutions. As the number of
tasks are growing continuously so the concept of processing by only single processor is not
optimized solution. The main scheduler selects three more schedulers for finding the set of
optimized tasks. The main scheduler is called master scheduler while the rest of the three
schedulers are known as slave schedulers. These slave schedulers are selected on the basis of
state information present at the master scheduler. Let us consider the figure 1 for the illustration
of this kind o grid model.

4
Figure 1. Master and slave nodes in scheduler
Figure 2: Group of nodes connected to grid
Suppose we have currently five nodes connecting to the grid. Then the grid node N1 is
consider to be the master node for assigning the tasks to the slave node and let’s say we
have N2, N3 and N5 are slave nodes depending upon the state of nodes and the current
load on the node. These three nodes are selected for the processing of genetic algorithm
and they return the optimized set of tasks that can be assigned to respective nodes for
processing. The master node selects the node N2, N3 and N5 for the processing and
implementation of genetic algorithm as these are the nodes with less loads. Now we see
that how genetic algorithm is being practically implemented on these grid nodes.
2.2. SELECTION OF OPTIMIZED SET O JOBS FOR ASSIGNING TO
RESOURCES
Genetic Algorithm
The master node N1 implements the first two steps of genetic algorithm. Let’s assume that node
N1 gets the set of tasks or in other words an array of tasks which are selected on the basis of
heuristic names as first come first serve. On the basis of FCFS basis the node N1perofrms the
next step named as fitness function over the first set of population which are actually the tasks
and we present these tasks in the form of numbers. We consider the tasks completion time as the
population which passes through the fitness function. We consider that at first on the basis of
first come first serve we have the following set of tasks in the form of an array.
Initial Population
A Initial population based on the completion time of each task:
This is actually the completion time of each task in milliseconds. Genetic algorithm has initial
population Most of the genetic algorithm selects the initial population without any define
criteria. But we introduce the heuristic that is first come first serve for the selection of initial
population. Now at next step the population passes through the fitness function.

5
Fitness Function
Fitness unction actually approves that which set of population is capable of reaching to the next
step of cross over and mutations in genetic algorithm. Hence fitness unction is responsible for
doing most of the task in genetic algorithm, hence we tried that we consider all those factors
which can overall increases the efficiency of load balancing procedure and can minimize the
makes pan. We have taken in account the following factors for fitness function. Fitness function
based on our algorithm comprise of important factors. The following equation gives our fitness
function.
• WT = Waiting time for each task before it is selected for processing
• TPT =Total processing time needed by a task to complete
• CLP=Current load on processor
• LS=Length of population string (which is selected on the basis of FCFS)
• DF=Delay factor (if some kind of delay is occurred from any processing node)
Current load on the processing node can be measure by taking in consideration the following
factors. is the tasks that are currently being executed by processor as well as size of all tasks that
are waiting in processors queue. This load on overall processor can find by using following
equation.
• CL= Current load present on particular processor
• RT= Time left for task that is currently processing by the processing node
• TS=Total size of task in terms of time (time taken by each task to complete)
• j=Number of processing nodes which are capable of processing
Our input string that passes through the fitness function are based on the total completion
needed by each task. The result of fitness function is categorized in two ways.
F(pop) > 0
F(pop)<=0
The matrix defined below shows the fitness function of each population and its probability of
selection. The highest probability strings are selected for mutation and cross over and for
generating new child population to be used in next generation. Each string has to pass through
this fitness function in order to select for the future population. The following table shows the
selection of strings.

6
Table 1: Fitness values of selected strings
2.4. Selection
For selection the population we have used the proportionate selection method/This kind of
selection methodology is similar to the roulette wheel selection methodology in which selection
is done on the basis of slice of wheel. The proportionate selection methodology uses the fitness
value for the selection of individuals.
Considering the equation based on the table 1
pi = Fi / SFi
Where pi is the probability of each string to be selected
F is the fitness value of each chromosome within the string
Where SFi is the sum of all the fitness values of each individual within the string. So all out of
population strings, the ones with lesser fitness value that is approaching to zero are selected as
the parents.
Crossover and Mutation
Crossover and mutation are the two important steps to be followed after the completion of
fitness function and selection procedures based on genetic algorithm. Parallel genetic algorithm
breaks down the procedure of GA into two parts and hence the crossover and mutation
procedure is performed by slave nodes selected by master node. These slave nodes are selected
on the basis of following conditions.
Total work load on each node → 0
Hence all those nodes whose total work load approaches to zero can be selected as a slave node.
Suppose in our case there are three nodes N2, N3 and N5 that are selected for performing the
crossover and mutation procedures. We have used the single arithmetic crossover for
implementing the rest of PGA. It works like this
Suppose we have two parents: and <y1,…,yn> and <z1,…,zn>
We will select gene (k) at random after that some selected point and mix values. Then the child
become
(y1,…. , yk,α.zk +(1-α).yk,…., yn )

7
Suppose we have these set of parents generated from master node and sent to slave node for
cross over.
Suppose we have α=4
After crossover the above mentioned results come out. Now the mutation that is also performed
by the slave nodes is done after crossover operation
3. CONCLUSIONS
Hence our algorithm that is based on the PGA based strategy is resourceful in manner that it
optimizes the overall procedure by decreasing the total execution time, total completion time
and total waiting time for all tasks. Hence as the time decreases the overall efficiency of the
parallel genetic algorithm increases. So in case of grid computing the implementation of parallel
genetic algorithm gives optimal results in selection of tasks to be assigned to decreases the
overall execution time for tasks in case where the numbers of tasks are continuously increasing.
The selection of minimum loaded node in grid computing overall increases the efficiency of
proposed algorithm. So in case of grid computing where the resources have to exploit and
nodes are quite heterogeneous this strategy gives the optimal results and of great use. Both
priorities and sizes of tasks are of uneven length. This algorithm is implemented on the local
grid created during this research work.java based software gridsim is use to implement on the
grid. Infrastructure for checking the load balancing issues and results in terms of time. As a
future work some new techniques for load balancing, like Ant colony optimization and particle
swarm optimization will be taken into consideration.
REFERENCES
[1] Yen, N. Y., & Kuo, S. Y. (2012). An intergrated approach for internet resources mining and
searching. J. Converg, 3(2), 37-44.
[2] Pyshkin, E., & Kuznetsov, A. (2010). Approaches for web search user interfaces. FTRA Journal of
Convergence, 1(1).
[3] Ling, A. P. A., & Masao, M. (2011, May). Selection of model in developing information security
criteria on smart grid security system. In Parallel and Distributed Processing with Applications
Workshops (ISPAW), 2011 Ninth IEEE International Symposium on (pp. 91-98). IEEE.
[4] Schmid, O., & Hirsbrunner, B. (2012, March). Middleware for distributed collaborative ad-hoc
environments. In Pervasive Computing and Communications Workshops (PERCOM Workshops),
2012 IEEE International Conference on (pp. 435-438). IEEE.
[5] Aikebaier, A., Enokido, T., & Takizawa, M. (2011). Trustworthy group making algorithm in
distributed systems. Human-centric computing and information sciences, 1(1), 1-15.
[6] Maheshwari, P. (1996, January). A dynamic load balancing algorithm for a heterogeneous
computing environment. In System Sciences, 1996, Proceedings of the Twenty-Ninth Hawaii
International Conference on, (Vol. 1, pp. 338-346). IEEE.

8
[7] Li, Y., Yang, Y., Ma, M., & Zhou, L. (2009). A hybrid load balancing strategy of sequential tasks
For grid computing environments. Future Generation Computer Systems, 25(8), 819-828.
[8] Zomaya, A. Y., & Teh, Y. H. (2001). Observations on using genetic algorithms for dynamic load-
balancing. Parallel and Distributed Systems, IEEE Transactions on, 12(9), 899-911.
[9] Nikravan, M., & Kashani, M. H. (2007, June). A genetic algorithm for process scheduling in
distributed operating systems considering load balancing. InProceedings 21st European Conference
on Modelling and Simulation Ivan Zelinka, Zuzana Oplatkova, Alessandra Orsoni, ECMS.
[10] Naseri, N. K., & Jula, A. (2012, August). A hybrid genetic algorithm-gravitational attraction
search algorithm (HYGAGA) to solve grid task scheduling problem. In International Conference
on Soft Computing and its Applications (ICSCA'2012).
Authors
Engineer. Saad Masood Butt received his BS (Software Engineering) degree
from Bahria University Islamabad, Pakistan in 2008. He completed his MS
(Software Engineering) degree in 2010 from Bahria University Islamabad,
Pakistan. He is the recognized Engineer of Pakistan approved by Higher
Education Commission and Pakistan Engineering Council (PEC). He has got
more than 7 years international work experience and was associated with
various organizations and universities in Pakistan and Malaysia.

Use of genetic algorithm for

More Related Content

What's hot (19)

Similar to Use of genetic algorithm for (20)

Recently uploaded (20)

Use of genetic algorithm for