IRJET- Optimization of Completion Time through Efficient Resource Allocation of Task in Cloud Computing by Enhancing the Genetic Algorithm using Differential Evolutionary Algorithms

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 06 Issue: 08 | Aug 2019 www.irjet.net p-ISSN: 2395-0072
© 2019, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal | Page 881
OPTIMIZATION OF COMPLETION TIME THROUGH EFFICIENT
RESOURCE ALLOCATION OF TASK IN CLOUD COMPUTING BY
ENHANCING THE GENETIC ALGORITHM USING DIFFERENTIAL
EVOLUTIONARY ALGORITHMS
E. YASHASRI1, K.M UMA MAHESWARI2
1E. YASHASRI & Department of Computer Engineering, SRM Institute of Science and Technology, Kancheepuram,
Tamil Nadu, 603203, India.
2K.M UMA MAHESWARI, Assistant Professor, & Department of Computer Engineering, SRM Institute of Science and
Technology, Kancheepuram, Tamil Nadu, 603203,
---------------------------------------------------------------------***----------------------------------------------------------------------
Abstract - The primary purpose of our project is to develop
technique for efficient resource allocation of a task in cloud
computing that has a mono-objective aim which is to reduce
the make span or completion-time. We are going to achieve
this by a soft novel algorithm that is the “genetic algorithm”.
To enhance this objective parameter – minimization of
completion time, we will combine genetic algorithm with the
differential evolutionary algorithms. The laid down technical
assumptions are that the directed acyclic graph which is
represented by G (V, E), where the V (Vertices) is a set of nodes
in the graph and E(Edges) denotes the precedence relations
between the tasks. The nodes are weighted by the value of its
computation time and the edges are weighted with the cost of
communications between two tasks. If two tasks are assigned
on the same processor the communication cost is zero. A
genetic algorithm takes selection, mutation and crossover as
its operations in the respective order, who’s output acts as an
input to the DE algorithm that has the same operations but in
a reverse order of workflow. These double operational tasks
refine the output and gives us the best task allocation that is
required for the respective resource. This algorithm also
handles real time parameters that eventually show an
improvement by just working on the completion time.
Key Words: Resource Allocation, Genetic Algorithm,
Differential Evolutionary algorithms, Cloud Computing,
IaaS, Completion time, Load Balancing, DAG (Directed
Acyclic Graph)
1. INTRODUCTION
Cloud computing is a service which enables a delivery of
computing resources over the Internet. Cloud computing is
the latest trends in the world ofInformation Technology,The
key word of cloud computing is the “cloud” itself, the exact
meaning of the word cloud in this reference is that clouded
cover computing is a group of web based computing
resources that are delivered on demand by information
services to users from any location in the world. The main
idea of cloud computing is that it is a large group of
dependent computers.
These computers can be personal computers or
network servers. This service extends beyond a single
company or enterprise. Cloud storage can be viewed as a
model in which data is stored, managed and backed up
remotely and made available to users over a network asweb
or typically the Internet. The service and the data served by
the cloud are available to wide group of users. Access to
these data are via the Internet. Only theauthorizeduserscan
access these data’s and the applications that are at any
computer over any Internet connection.
In cloud computing, the allocationofresourceisone
of the processes used to assign the available resourcestothe
needed application users, these are called the cloud
resources they can be provisioned by either on-demand or
a fine-grained or even in a multiplexer manner. In the cloud
the resource allocation is mostly based on the following
factors software, platform and the infrastructure
respectively known as software as a Service(SaaS),Platform
as a Service (PaaS) and Infrastructure as a service (IaaS).
In platforms of cloud, resource allocation mostly
takes place at two major levels: Thefirstlevel involves,when
the cloud user wishes to be uploaded to the cloud, then the
load balancer (which can by optimized) assigns the
requested object to the remote physical computers, in order
to balance the computational load of several applications
across physical computers. The second level involves the
application receiving multiple incoming requests, these
requests should be assignedtoaninstanceofapplicationthis
is done to balance the computational load across a set of
instances for the exact same application.
Our focus in this paper is to ensure the apt resource
is assigned to the request both in the first and second level.
Since computational cost will eventually be balanced
through this method, we will be focusing on reducing the
makespan. When a request is issued the load balancer also
called as the resource monitor will use the proposed
algorithm with the improved mathematical model and
allocate resource that is fit for the job.
The second main constrain that we focus in this
paper is to ensure that the resource in this case the
processor is never idle. Thus, we will be reducing the idle

time and increase the efficiency of the entirecloud.Decrease
in the makespan will make both the computational cost and
load balancing efficient. As mentioned before the allocation
of the resources can be either on-demand orfine-grained. As
we will be dealing with dynamic requests, the on-demand
technique is implemented in the algorithm.
The enhanced Algorithm model will improvise the
Improved genetic algorithm and provide with better
completion time than the existing genetic algorithm.Genetic
algorithm is a biological concept that is used for generating
population. It is inferred by Darwin’s theory of evolution. In
his theory Darwin the principle of “Survival of the fittest” is
used as the method of scheduling in which the tasks are
assigned to resources according to the value of fitness
function for each parameter of the task scheduling process.
The three main principles which are used for in this
algorithm are Initial population, Fitness function, Selection.
The main techniques used would be Crossover, Mutation,
Selection. The techniques are used in the traditional genetic
algorithm to produce the best out of the lot, which can be
mapped to the fittest value in the N number of iterations.
Initial population points to the number of individuals or the
resources that are used in the GA to find the optimal
solution. The fittest value will be the maximum completion
time of the resources that is apt for the provided request,
this is selected criteria from the initial population.
Therefore, in this paper we will be changing the
mathematical criteria required for the crossover and
mutation and selection of the fittest value. The Differential
evolutionary algorithm is like the GA, the steps are vice
versa. Hence when we implement both these algorithms
together, we will get the optimized combination of task to
resource. The corresponding goal tooptimizethecompletion
time will be achieved. In the proposed system we will be
running both the Improved and the Enhanced genetic
algorithm with Differential evolutionary algorithm and
tabulate and represent a graph for both the make span and
cost for each of the algorithmstoprovethatoptimization has
occurred.
2. ARCHITECTURE
Cloud is the solution with dynamic way to scale up and scale
down the services when required. It is based on the concept
of implementing virtualization that includes making a layer
between the resources (OS or VM) and host machine. It
improves the concept of decoupling the dependency of OS
and hardware.
A client sends a request to the cloud that can be
broken into individual tasks.Everytask isaninterdependent
sub-task that has its output to be the input for another task.
That’s why, it is very important in the perspective of the
cloud to assign the right virtual machine to the unique
request that has been raised by the client.
A cloud consists of processors or VM layer, under
which we have the Virtualization layer, that is followed by
hardware or servers’ layer remotely located. A cloud also
consists of a resource monitor that interacts between the
enhanced genetic algorithm and gives the apt resource
allocation to the client as per their requirement.
Figure 1: Architecture
3. RELATED WORK
When job scheduling itself is viewed it is vital to understand
the conservation. Job scheduling model for cloud computing
1 describes on detail about energy consumption, profit of
service, to recognize decision to analyze the applications.
These parameters resulted in a fruitful output which
benefited in many ways. The use of scheduling algorithms
with the above criteria eventually helped in an optimized
dynamic selection of jobs alias the tasks.
The basic study of Genetic algorithm based on
resource allocation for cloud computing describes aboutthe
scheduling procedure that is done using the Virtual machine
that uses the information of tasks to make the task
scheduling, in this the model included the type of request as
a main parameter for task scheduling. The main aim of the
algorithm was in creating a criterion for task scheduling
using quality evaluation.
Research of cloud computing task scheduling
algorithm based on improved genetic algorithm 3 uses the
algorithm energy aware improvedgeneticalgorithmfortask
scheduling in the cloud computing focuses onthe makespan
to optimize the cost, also to provide a quality of service with
conserved energy. These three principles were the
parameters which differed from the traditional genetic
algorithm. The process implemented was using the voltage
dynamic scheduling. Thisoptimized energyconservationina
new level.
Eventually mono objectivesystemsifnotfocusedon
the right systems will tend to create drawback in other
aspects. Hence Genetic algorithm based bi-objective task
scheduling in hybrid cloud computing 4 aimedonoptimizing

both the execution time and the execution cost for a hybrid
system. This simultaneous optimization of the execution
time and the cost of the scheduling in hybrid cloud provided
a new perspective on task scheduling in a cross cloud.
Understanding the various types of resources that
are in a virtualized form in a cloud, that are required for the
dynamic scheduling is projected in Various priority-based
task scheduling algorithms incloudcomputing5.Analysisthe
types of resource provides the priority of scheduling this
was captured and optimized. In Resource allocation incloud
computing 6 the take on the difference in the various
software and hardware resources for multiple allocation
algorithms was performed which gave a result of choosing
the best duo for ensuring efficient resource allocation.
In order to achieve higher performance and better
utilization of resources with minimum response time, Cloud
pad computing should be able to execute maximum number
of tasks for any type of complicated tasks in any
environment. The paper Survey on scheduling algorithmsin
cloud computing and the paper A survey: to harness an
efficient energy in cloud computing both suggest this
technique.
Hybrid algorithms provided easier workflow; this
was presented in workflow scheduling using hybridGA-PSO
algorithm in cloud computing 9. In this paper the main aim
was to increase make span and ensure load balance and
optimize the fittest value calculation.Thiswasachieved with
integration of Genetic algorithm and Particle Swarm
Optimization algorithms for improving the cost reduction
and make span along with load balance on virtual machine.
This however could be extended for dynamic resource
allocation too.
Cloud task scheduling algorithm based on improved
Genetic algorithm paper aimed to optimize both the
execution and the computational cost of resource for
efficient allocation. This work involved changed
mathematical model than to that of a traditional Genetic
algorithm. The usage of the three-state selection method
helped converting the simple genetic algorithm toimproved
genetic algorithm, for improving the completiontime.Hence
the main goal of this work was to improve the completion
time. This work is considered as the base for our proposed
system. Similar to this work but implementing different
mathematical model with different set of parameters and
dataset we will achieve the objective of efficient flowtime
and reducing the idle time.
Cloud computing, the longest hope for using a utility
for computing has helped potentially convert a great part of
the IT industry, it has enabled the software more portable
and services more affordable. The major differentiation of
cloud computing from other distributed computing
techniques is the on-demand resourceallocation.Something
like the allocating of Electric power.
After reviewing several research papers, we’vegained
a fair amount of knowledge on the various algorithms and
techniques implemented to increase the efficiency of the
dynamic allocation of resources. With the knowledgegained
by the data provided we hope to implement these ideas in
the process of achieving our goal in project.
4. MODULES
 Datacentre configuration
 Production of Initialization
 Improved Genetic Algorithm
 GA-Differential Evolution Method (Proposed)
 Result and Analysis
5. MODULES DESCRIPTION
5.1 DATACENTER CONFIGURATION
 Input: Hosts, Vms, Tasks - configurations
and parameter initialization.
 Output: Datacentre Environments.
In this module, the data center environments are configured
by giving the values of MIPS, memory, storage, bandwidth,
number of cores on an individual host and virtual machines
and the number of servers (hosts) and virtual machines.And
the workflow(DAG) tasks are submitted to schedule and
execute on the datacenters. Also, the parameters for
initialization and task scheduling values are entered.
5.2 PRODUCTION OF INITIAL POPULATION
In this module, in order to improve the quality of the initial
population, Min-MinandMax-Minalgorithmsareintroduced
to initialize the population.
Min-Min Algorithm: Min-Min scheduling is based on the
concept of assigninga task havingminimumcompletiontime
(MCT) first for execution on the resource, which has the
minimum completion time(fastestresource).Thisalgorithm
is divided into two steps. In first step, expected completion
time of each task in the metatask is calculated on each
resource. In second step, the task with minimum expected
completion time is selected and assigned to the
corresponding resource, and then the selected task is
removed from the metatask. This process repeats until all
the tasks in metatask get mapped. Min-min algorithm is a
simple algorithm but it gives the fast result when the size of
task in metatask is small as compared to large size task. On
the other hand, if large size tasks overlay the number of
smaller tasks, it gives a poor resource utilization and large
make span because large size tasks have to wait for the
completion of smaller tasks.
Max-Min Algorithm: Max-Min algorithm performs the same
steps as the Min-Min algorithm but the main difference
comes in the second phase, where a task ti is selected which
has the maximum completion time instead of minimum

completion time as in min-min and assigned to resource Rj,
which gives the minimum completion time. Hence it named
as the Max-Min algorithm. This process is repeated until the
metatask get empty or all the tasks are mapped.
The main aim of max-min scheduling algorithm is to reduce
the waiting time of large size jobs. In this algorithm, small
size tasks are concurrently executed with large size tasks,
hence reducing the make spanandprovidingbetter resource
utilization.
5.3 IMPROVED GENETIC ALGORITHM:
There are three main operations of simple genetic algorithm
for population: Selection, Crossover, and Mutation.
Selection operation: in each generation, individuals who
adapt to the environment are selected by the fitness
function, and the next generation is produced through these
individuals.
Selection of Fitness Function
In genetic algorithm, the selection of fitness function is very
important, and it is directly related to the convergenceofthe
algorithm and the search of the optimal solution.
f = max (TVi) i∈{0,1,2,…,M-1} --- (1)
Where, TVi represents the time that the virtual machine i
execute all the cloud tasks assigned to it, and M represents
the number of virtual machines.
Selection operation(mathematically)
The selection operation is to select the good individuals and
eliminate the bad individuals. The average fitness fa and the
random fitness fr are introduced into the selection
operation.
sum = Σ f(i) i ∈ {1,2, 3,...,n} --- (2)
fa = sum /n --- (3)
fr = fa. Math. Random () --- (4)
In these equations, sum represents the total finishing time,
including all individuals in this population, f(i) is the fitness
of individual i, and n is population size, by formula (3) and
(4) we can see fa>fr.
The three-stage selection method is adopted in this
paper; each selection retains the best individual of the
contemporary population and makes different grades for
other individuals. If the individual fitness is f<fr, the
individual is defined as a excellent individual and retained;
Individuals with fitness fr<f<fa are defined as sub excellent
individuals and replaced by variants of the optimal
individual; Individuals with fitness f>fa arepoorindividuals,
these individuals will be eliminated and replaced by new
individuals. This not only ensures the stability and diversity
of the population, but also helps to find the global optimal
solution.
Crossover operation
Two different individuals in a population exchange gene in
the same position to produce new individuals. Suppose the
individual A crosses with the individual B, and the crossover
region length is L, in this region, the number of same genes
of two individuals is K, and the crossoverregionsimilarity is:
S = K/L
If the crossover region length is 9, the genes
LA=022130111, LB=012112031,thenK=4,crossover region
similarity S=0.44, the crossover region similarity threshold
μ=0.5 is specified in this paper. Two individuals perform
crossover operation, and when the crossover region
similarity is greater than the threshold, we decide that the
crossover operation is meaningless and the two individuals
will reproduce in asexual form: individual recombine genes
randomly in crossover region then add it to the next
iteration.
Mutation operation
The genes of certain individuals in the population
may change, which may make individuals more adaptable
and, of course, weaker.
5.4 GA-DIFFERENTIAL EVOLUTION METHOD
(PROPOSED)
In the proposed algorithm, the operators of Genetic
Algorithm and Differential Evolution Method are combined
to find the optimal solutions to minimize the make span of
the tasks execution and find the best task schedule in the
cloud environments. The directed acyclic graph which is
represented by G(V, E), where the V(Vertices) is a set of
nodes in the graph and E(Edges) denotes the precedence
relations between the tasks. The nodes are weighted by the
value of its computation time and the edges are weighted
with the cost of communications between two tasks. If two
tasks are assignedonthesameprocessorthecommunication
cost is zero. The directed acyclic graph containsthetasksare
T0, T1, T2, . . . , T10 , and T0, is entry task and T10 is exit
task. In this proposed algorithm the Genetic Algorithm (GA)
is applied, theirselection,crossoverandmutationoperations
and produce a set of solutions, these solutions are
considered as an initial population to Differential
Evolution(DE) method and the DE operations applied on it
and produce a solution as next population to GA. This
operation processes are repeated for a number of times and
when the DE terminate the operations all the solutions are
updated. To avoid the priority violence, it must be sorted
from left to right, but when the new solutions priority is
violated after the production of children. In this algorithm
the premature convergence is avoided by comparing the
produced children with their parents, if the fitness value is

better than the parents then the parents are replaced by
children else it is aborted. Once the initial population is
generated, the fitness values of the solutions are evaluated.
In each iteration of the algorithm, the termination condition
is checked. If the termination condition is satisfied, then the
algorithm is produced optimal solutions. Otherwise, GA and
DE algorithms operators applied to the individuals
respectively. The GA-DE algorithm implements the
operations as:
1) Production of initial population
The initial population contains individuals and the
sizes of the chromosomes arefixed.Inthismethod,theinitial
population is considered for having a better solution from
MIN-MIN and MAX-MIN method. The prioritiesfoundfor the
DAG tasks are considered as the initial population. The
remaining of the individuals are generated randomly. The
first and last places in the chromosomewhichtheyrelated to
the start and exit nodes are established in the chromosome,
respectively and the remaining tasks are randomly
generated and they are sorted selected from left to right in a
way that priorities are not violated the precedence
constraints.
2) Selection Operator
The roulette wheel selection function is used to select
an individual for a good fitness values and the genetic
operations are implemented on the selectedindividuals.The
individual which is having high fitness value gets more
chances to be selected.
3) Crossover Operator
Figure 2: Crossover Operator
Crossover replaces same genes of one individual with
genes of another to produce two valid offspring. The
powerful single-point crossover operator is used. A random
point between l and n is generated and the crossover is
applied if the genes of both parents from entry node to
crossover point are not identical. The crossover singlepoint
which is equal to 5 produces the two new offspring. The left
side of the children inherit from their ’ genes of parents in
the same genes location then selected genes are removed
from the parent and the remaininggenesareimportedto the
child from left to right. And the produced children will be
valid. After producing the children, the fitness of them
obtained by fitness function. Then the fitness values of
children are compared with parents and if fitness values of
children are better than parents, the children will replace
parents.
4) Mutation Operator
Mutation operator obtain a new chromosome by
changing two genes in a way that precedence constraint is
not violated. In this operator, at first, a gene is selected
randomly. Then, first successor of the selected task (tj) from
the mutation point to the end is found. If there is m th gene
which it is a member of [ i+1, j-1 ] and the predecessorsoftm
are not in front of ti, ti and tj can be changed with each other.
If in mutation function these conditions do not happen, so,
the mutation operator algorithm will be run from the
beginning. Finally, the fitness value of a childismeasuredfor
the child and if the fitness value of the child betters the
parent’s fitness then the child will replace the parent.
5) Termination Conditions
The genetic and DE algorithms are considered as
evolutionary algorithms, which can be executed for
unlimited time. But for achieving solutions in the algorithm
we need to terminate processes. Hence, termination
condition must be considered. Some of the policies to
terminate algorithms are considering thefitnessevaluations
or the running times of the system or by exploring the
population diversity. In this paper, the algorithm is
terminated when the algorithm is iterated for 30 times.
Table 1: Parameters
Number of Hosts 100
Number of Virtual
Machines
40
Selection Technique Roulette wheel
Crossover Rate Single point
Mutation Factor m gene  {i+1, j-1}

5.5 RESULT AND ANALYSIS
[1] Input: Number of Tasks on X-axis and Completion Time
(Make span) on Y-axis
Output: Comparisonof MakespanVersustheNumberof
Tasks
A graph is drawn between the number of tasks on X-axisand
the corresponding make span of the applications on Y-axis.
The Genetic Algorithm-Differential Evolution algorithm is
compared against the existing IGA scheduling algorithms in
terms of the make span by 100, 200, 300, 400 and 500 tasks.
The number of tasks increases in the graph, the proposed
algorithm shows the better optimal make span values
compared to existing improved Genetic Algorithms.
Table 2: Optimization of Execution time
ALGORITHMS
NUMBER
OF TASK
30
NUMBER
OF TASK
50
NUMBER
OF TASK
100
SIMPLE
GENETIC
ALGORITHM
675.25
seconds
798.67
seconds
827.36
seconds
IMPROVED
GENETIC
ALGORITHM
413.677
seconds
520.12
seconds
627.067
seconds
Figure 3: Graph for Optimization of Execution time (Table
2)
[2] Input: Number of Tasks on X-axis and cost of expenses
on Y-axis.
Output: Illustrated a chart of comparison of cost and
number of tasks.
Another graph is plotted between the number of tasks and
the cost of expenses to execute the tasksintheallocateddata
Centre based on the utilization of resources. It shows the
minimization of the cost of expenses compare to the existing
algorithm.
Table 3: Comparison of Execution Cost
Figure 4: Graph for Comparison of Cost (Table 3)
6. CONCLUSION
The completion cost has reduced to a great extent when
compared to simple genetic algorithm andimprovedgenetic
algorithm. The execution cost has reduced to a large extent
when compared to the SGA and IGA. By achieving these two
goals we can conclude by stating, the resource allocation of
the tasks in a cloud environment has been optimized.
ALGORITHMS NO OF
TASKS
(30)
COST
NO OF
TASKS
(50)
COST
NO OF
TASKS
(100)
COST
SIMPLE
GENETIC
ALGORITHM
$ 1020.567 $ 1700.53 $ 1940.632
IMPROVED
GENETIC
ALGORITHM
$ 784.98 $ 540.315 $ 971.954
ENHANCED
GENETIC
ALGORITHM
WITH DE
ALGORITHM
$ 418.886 $ 501.448 $ 651.659

7. REFERENCES
1) Mizan, M. and Latip. “Modified bees life algorithm
for job scheduling in hybrid cloud.” International
Journal of Engineering and Technology, 2, 974–9.
2) Leena V. A., A. B. A. S. and S., R. M. “Genetic
algorithm based bi-objective task scheduling in
hybrid cloud platform.” International Journal of
Computer Theory and Engineering.
3) Junwei G, Y. Y. “Research of cloud computing task
scheduling algorithm based on improved genetic
algorithm.” In Proceedings of 2nd International
Conference on Computer Science and Electronics
Engineering, 2134–2137.
4) Here Comes Hybrid Cloud, Gartner.
http://blogs.gartner.com/thomas_
bittman/2012/09/24/mind-the-gap-here-comes-
hybrid-cloud/. Accessed: 2016-03-29.
5) Ms. Pooja, P.Vasani, Mr. Nishant And S. Sanghani,
(2013) "Literature Review: Various Priority Based
Task Scheduling Algorithms In Cloud Computing,"
Journal Of Information, Knowledge And Research
In Computer Engineering, (2) (2), 298-302.
6) Shivani Sharma and Dhanshri Parihar, (2014) "A
Review on Resource Allocation in Cloud
Computing," International Journal of Advance
research, Ideas and Innovations in Technology.
7) M.Padmavathi, Shaik. MahabbobBasha andSrinivas
Pothapragada, (2014) "A Survey on Scheduling
Algorithms in Cloud Computing," Journal of
Computer Engineering (IOSR-JCE), (16)(4), 27-32.
8) Malathi.P And Arumugam.S, (2015) "A Survey: To
Harness An Efficient Energy In Cloud Computing,"
International Journal Of Ubicomp (Iju), (6)(3), 1-7.
9) Workflow Scheduling Using Hybrid GA-PSO
Algorithm in Cloud Computing Ahmad
M. Manasrah1 and Hanan Ba Ali2 Wireless
Communications and Mobile Computing
Volume 2018, Article ID 1934784.
10)Cloud Task Scheduling Algorithm based on
Improved Genetic Algorithm Hu Yao, XueliangFu*,
Honghui Li, Gaifang Dong, Jianrong Li College of
Computer and Information Engineering, Inner
Mongolia Agricultural University, Hohhot,010018,
China. Available onlineatwww.ijpe-online.comvol.
13, no. 7, November 2017.

IRJET- Optimization of Completion Time through Efficient Resource Allocation of Task in Cloud Computing by Enhancing the Genetic Algorithm using Differential Evolutionary Algorithms

Recommended

More Related Content

What's hot (19)

Similar to IRJET- Optimization of Completion Time through Efficient Resource Allocation of Task in Cloud Computing by Enhancing the Genetic Algorithm using Differential Evolutionary Algorithms (20)

More from IRJET Journal (20)

Recently uploaded (20)

IRJET- Optimization of Completion Time through Efficient Resource Allocation of Task in Cloud Computing by Enhancing the Genetic Algorithm using Differential Evolutionary Algorithms