Cost-Efficient Task Scheduling with Ant Colony Algorithm for Executing Large Programs In Cloud Computing

International Journal of Computer Applications Technology and Research
Volume 6–Issue 4, 194-198, 2017, ISSN:-2319–8656
www.ijcat.com 194
Cost-Efficient Task Scheduling with Ant Colony
Algorithm for Executing Large Programs
In Cloud Computing
Fatemeh Imani
Department of Computer
Engineering, Ardabil Branch,
Islamic Azad University,
Ardabil, Iran
Shiva Razzaghzadeh
Ardabil, Iran
Masoud Bekravi
Ardabil, Iran
Abstract: The aim of cloud computing is to share a large number of resources and pieces of equipment to compute and store
knowledge and information for great scientific sources. Therefore, the scheduling algorithm is regarded as one of the most important
challenges and problems in the cloud. To solve the task scheduling problem in this study, the ant colony optimization (ACO) algorithm
was adapted from social theories with a fair and accurate resource allocation approach based on machine performance and capacity.
This study was intended to decrease the runtime and executive costs. It was also meant to optimize the use of machines and reduce
their idle time. Finally, the proposed method was compared with Berger and greedy algorithms. The simulation results indicate that the
proposed algorithm reduced the makespan and executive cost when tasks were added. It also increased fairness and load balancing.
Moreover, it made the optimal use of machines possible and increased user satisfaction. According to evaluations, the proposed
algorithm improved the makespan by 80%.
Keywords: cloud computing; formatting; task scheduling; makespan; load balancing
1. INTRODUCTION
With the development of information technology, it is
necessary to do computing tasks everywhere at every time. It
is also essential that people be able to perform their heavy
computing tasks through some services without having
expensive hardware and software requirements. Cloud
computing is the latest solution provided by technology for
these needs. The National Institute of Standards and
Technology (NIST) defines cloud computing as a model to
provide easy access based on user demand through the
network for a group of modifiable and configurable
computing resources such as networks, servers, storage space,
applications and services. This access should be able to offer
quickly-provided or free services without needing resource
management or direct intervention. In this definition, the
cloud can be described with five essential features including
virtual computing resource sharing, WAN access, quick
flexibility, requested services (based on order or demand), and
measured services. The cloud environment is based on
requests, and users can increase or decrease the use of
resources. In other words, the results are related to usage in
the cloud environment. Sharing the usable computing power
among some tenants can improve the productivity rate
because servers are not idle for nothing in this method. One
reason is that computers are used more because cloud
computing customers do not need to calculate and determine
the maximum load [1]. Virtual machine scheduling problem
has been investigated in many studies conducted on cloud
computing environments. The main aim of scheduling in the
cloud is to shorten the makespan, increase the system
throughput, and establish load balancing in resource [2].
Running a program can be seen as the execution of different
tasks in it. Different tasks can be executed simultaneously
with several virtual machines. In this thesis, ACO was used
with a fair and accurate resource allocation approach based on
machine performance. As the number of natural resources
should be proportionate to the performances of ants, the tasks
performed by resources should be proportionate to the tasks
assigned to them. An appropriate path for resource allocation
is very important to perform tasks. Therefore, machines can
process tasks at the maximum power and on the shortest path.
As a result, resources are used optimally. In the cloud
environment, providers want to make the most of their
resources, and users want to minimize their costs [4].
However, they want to achieve their intended performance.
The appropriate and optimal use of resources such as memory,
processor and bandwidth is a challenge; therefore, the quality
of task scheduling is regarded an important problem which
has a great effect on the performance of cloud service
providers. All of the scheduling algorithms are intended to
minimize the makespan [4]. None of the previous studies dealt
with task distribution in a cloud environment by considering
the costs and optimal use of machines and increasing their idle
times. Considering task scheduling algorithms, the ACO was
adapted from social theories in this thesis to optimize the use
of machines, reduce machine idle time, and decrease the
makespan. It was also used to minimize executive costs by
modifying the idle time (vms) to perform insensitive tasks.
2. LITERATURE
In this section, previous research works on network
processing are reviewed. First, preliminary methods such as
Dynamic Level Scheduling are described, then the most
recent methods are reviewed.
2.1 Dynamic Level Scheduling
For this purpose, a specific model has been proposed. The
main aim of this method is to decrease the processing time. In
network processing environments, other scheduling
algorithms do not emphasize the subtasks of an application
which is run in the computing host or the virtual organization.
The main aim is to perform scheduling in a way that all the
input applications can use the available throughput. In the
paper presented by Cathody and Caratasa, the heuristic
technique was added to the abovementioned method to
increase the system efficiency [5].

www.ijcat.com 195
2.2 Allocating the Fast Process to the
Largest Task
The FPLTF scheduling algorithm (Xoa et al.) determine the
tasks based on the available resources in the system [6]. This
method depends on speed of processor and resources and the
size of tasks. In this method, the largest task is allocated to the
fastest resource. If there are many large tasks, this method will
not be efficient enough. The dynamic FPLTF (Chang et al.)
algorithm was developed with respect to the static FPLTF
algorithm. In this method, the highest priority is allocated to
the largest task. It is also necessary to estimate the data which
are required for processing [7].
2.3 WQR (Queue with Repeat)
This method is based on WQ. In this method, faster
processors are allocated to large tasks (Young et al.) by using
the FCFS and random scheduling methods. WQR iterates the
tasks to transfer them to available resources. The iteration of
tasks can be selected by the user. When one of these tasks is
finished, the scheduling algorithm stops the iteration of other
tasks. One of the problems of this method is that it spends too
much time allocating resources to the iteration operations.
2.4 Balanced Ant Colony Optimization
(BACO)
The main idea of this method is taken from ACO (Xoa et al.),
and it is mainly intended to decrease the processing time and
load balancing of each resource. This method changes the
density of pheromone based on the positions of resources,
something which can be possible by updating the pheromone
locally and globally. In this method, the makespans are
shortened at the same time as the system is kept in balance. In
the architecture of this network process scheduling, there are
four components: portal, information server, task scheduling
algorithm, and resources required for processing. The portal is
used as an interface for users [6].
Figure 1. Structure of the system
2.5 Load Balancing
Load evaluation is usually mechanized for the continuity of a
service when one or some components of the system fail. The
components are constantly monitored. When one component
does not respond, the load balancer comes up and prevents the
traffic from being sent to it. With an appropriate load
evaluation where resources are used, problems can usually be
mitigated. Not only does this decrease costs and creates green
computing, but it also keeps the pressure low on unique
circuits whose lifetime will be potentially elongated. In fact, it
can be stated that the load balancing is meant to find an
appropriate map of tasks on the available processors in the
system in a way that each processor runs an equal number of
tasks until the total makespan is minimized as much as
possible.
2.6 The Importance of Load Balancing
With load balancing, the load can be balanced through the
dynamic transfer of local tasks from one machine to another
one in a remote node or a machine which is used less often.
This solution maximizes user satisfaction, minimizes response
time, increases the exploitation of resources, increases the
failure times and improves the system efficiency. Load
balancing is also needed to achieve green computing in the
clouds [8].
3. OPTIMIZED ACO
Cloud computing is the extended version of network
computing which is done in a parallel and distributed way. It
is also a new model for business computing. Compared with
network computing, the new features of cloud computing
include heterogeneous resources distributed and dispersed in
large scale to include the datacenter. Moreover, the
virtualization technology creates latent heterogeneous
resources in cloud computing. Network computing is
generally used in scientific computing to solve the limited
domain problem. Cloud computing provides a user-oriented
plan which offers various services to meet the needs of users.
In cloud computing, resources are converted into virtual
resources by using the virtualization packaging technology.
This makes the resource allocation and interaction process be
different from user tasks and network computing [9,10].
3.1 Designing the Optimized ACO
To design the optimized ACO, user tasks are allocated to
resources which are the same as the machine output so that
the machine can do the processing with full power, and there
should not be any loads on resources. This increases the
efficiency of resources. Now tasks should be classified and
prioritized for the fair allocation of resources. This action was
not possible in the normal ACO. Users, resource providers
and the scheduler system are intended in cloud computing.
The main part of scheduling computations includes user tasks
and resources so that the fair distribution of resource
allocation can be possible in cloud computing by using the
optimized algorithm. The scheduling algorithm is optimized
includes two main steps. Figure 2 shows the architecture of
the optimized algorithm.
Figure 2. Architecture of the optimized algorithm

www.ijcat.com 196
3.2 The Strategy of the Proposed
Algorithm (Optimized Scheduling)
Figure 2 shows main mechanism of the algorithm in this
thesis. It uses the local optimal method to allocate resources to
a group of tasks and virtual machines. Now the algorithm is
briefly described.
1. The tasks sent by the user are classified and prioritized
with respect to the quality parameters in the computing
unit.
2. Given the type of tasks, two lists are created. One of
them is meant for processing tasks (runtime), and the
other one is meant for the tasks needing bandwidth.
3. A group of virtual machines named VMlist is given to
the system based on their normalized performances.
4. Given the number of virtual machine processors and the
expected waiting time of each task, the virtual machine is
selected.
5. Given the real bandwidth of virtual machine and the
expected bandwidth of tasks, the desired machine is
selected.
6. If tasks are equal to the performances of resources, they
are assigned to them. In other words, fairness is
considered in the allocation of tasks.
7. Finally, the virtual machine is freed after task processing
is done.
Figure 3. ACO algorithm
4. SIMULATION ENVIRONMENT
The experiments were implemented in the Cloud simulator
using library functions including CloudSim, SimJava, and
GridSim. This test is run in the CloudSim environment, and
the application is run at the user layer code [10,11].
4.1 Evaluation Criteria
In the next sections, three tests are conducted in the form of
different parameters such as makespan, the number of
processors, bandwidth, and user satisfaction. Then the
optimized algorithm is compared with different algorithms. In
this evaluation, the optimized ACO algorithm is compared
with other scheduling algorithms based on Berger’s model
and greedy model. In all the experiments, virtual machine
parameters including machine ID, the number of processors,
available memory, and bandwidth in Table (2-4) were used
along with task parameters such as task ID, class ID, length,
file size, output size, the expected time, and the expected
bandwidth according to Table (1-4). Then the proposed
algorithm is compared with other algorithms [12, 13].

www.ijcat.com 197
Table 1. Jobs Parameter
Expectation
BW
Expectation
time
Output_sizeFile_sizeLength
Class
type
Task
Id
-0440440044044414
-0440440444044411
-041044044044410
-04404440444044410
0444-044044044400
0444-0440444044400
1044-04404404406
0444-0441444004407
Table 2. Vm Parameter
4.2 Comparing the Results of Simulating
the Proposed ACO Method with Berger
and Greedy
4.2.1 Makespan in Different Algorithm
In this experiment, the optimized algorithm was compared
with the scheduling Berger algorithm and greedy base by
considering the makespan. The tests were conducted in a
highly heterogeneous environment. Figure 4 and Table 3
show the runtime of tasks in an analytic comparison.
Table 3. Vm Parameter
76543210Task Id
250006002001000200500400ACo-O
500140520250520200500400Berger
50000300200600200500400Greedy
Base
Figure 3. Comparing completing time
4.2.2 Comparing the Number of Processors
in the Machines Allocated to Tasks
In this section, the proposed algorithm is compared to
different scheduling algorithms by using improvement
strategies with respect to quality. The tests are run in this
range. The main research goal and necessity is to select
resources and to achieve the best time as well as the
appropriate cost. According to the datasets shown in Tables 2
and 3, the number of processors of machines used in the
Berger and Greedy Base algorithms can be compared with the
proposed algorithm.
Figure 4. Comparing the Number of Processors in the Machines
Allocated to Tasks
5. SUGGESTIONS
More studies can be conducted on resource scheduling.
Regarding future works, some suggestions are made as
follows:
• A fuzzy neural network of the service quality of tasks and
resources can be used with an ACO approach in the future.
• The methods based on the genetic algorithm can be used to
make significant improvements in the proposed algorithm
because they are very helpful in optimization problems.
6. REFERENCES
[1] Danielson K. 2008. Distinguishing Cloud Computing
from Utility Computing . Available from http:/ / www.
ebizq. net/ blogs/ saasweek/ 2008/ 03/
BandwidthMemoryCPUVM ID
1200204040
3000102421
1000102422
120051213

www.ijcat.com 198
distinguishing_cloud_computing/ . [Accessed 26 March
2008]
[2] Sran N. and Kaur N. 2013. Comparative Analysis of
Existing Load Balancing Techniques in Cloud
Computing, International Journal of Engineering Science
Invention, Vol.2, Issue 1.
[3] Hamo A. and Saeed A. 2013. Towards a Reference
Model for Surveying a Load Balancing, International
Journal of Computer Science and Network Security ,
Vol. 13.
[4] Peng L. 2009. the definition of cloud computing and
characteristics. http:// www.china cloud .cn/ [ Accessed
25 February 2009]
[5] Gkoutioudi, H. D. Karatza, “Task cluster scheduling in a
grid system”, Simulation Modeling Practice and Theory,
Vol. 18, pp. 1242-1252, 2010.
[6] Chang and et al, “An ant algorithm for balanced job
scheduling in grids”, Future Generation Computer
Systems, Vol. 25, pp. 20-27, 2009.
[7] Goa Y. and et al, “Adaptive grid job scheduling with
genetic algorithms”, Future Generation Computer
Systems, Vol. 21, pp. 151-161, 2005.
[8] Begum S and Prashanth C S R. 2013. Review of Load
Balancing in Cloud Computing. International Journal of
Computer Science Issues , Vol.10, Issue 1.
[9] Khetan A, Bhushan V and Chand Gupta S. 2013.A Novel
Survey on Load Balancing in Cloud Computing
International Journal of Engineering Research &
Technology
[10] Sundararajan S, Raj B. 2011. IBM Center for High
Performance On Demand Solutions. IBM Cloud
Computing White Paper
http://www.ibm.com/developerworks/Websphere
/zones/hipods/.[Accessed2011].
[11] Lia J, Fenga L, Fang S. 2014. An Greedy-Based Job
Scheduling Algorithm in Cloud Computing. JOURNAL
OF SOFTWARE, VOL. 9, NO. 4, APRIL.
[12] Buyya R, Ranjan R and Rodrigo N. 2009. Calheiros,
Modeling and simulation of scalable cloud c computing
environments and the CloudSim Toolkit: challenges and
opportunities. In: Proceedings of the seventh high
performance computing and simulation conference
(HPCS 2009, ISBN:978-1- 4244-49071), Leipzig,
Germany. New York, USA: IEEE Press; June 21-24.
[13] Baomin X U C Z . 2011. Job scheduling algorithm based
on berger model in cloud environment. Advances in
Engineering Software, vol. 42, pp. 419–425.

Cost-Efficient Task Scheduling with Ant Colony Algorithm for Executing Large Programs In Cloud Computing

More Related Content

What's hot (16)

Similar to Cost-Efficient Task Scheduling with Ant Colony Algorithm for Executing Large Programs In Cloud Computing (20)

More from Editor IJCATR (20)

Recently uploaded (20)

Cost-Efficient Task Scheduling with Ant Colony Algorithm for Executing Large Programs In Cloud Computing