Load Balancing Heuristic for Tasks Scheduling in Cloud Environment

Kadda Beghdad Bey

, Farid Benhammadi

, Mohamed El Yazid Boudaren

and Salim Khamadja

Informatics Systems Laboratory, Ecole Militaire Polytechnique, Algiers, 16111, Algeria

Ecole Militaire Polytechnique, Algiers, 16111, Algeria

Keywords: Cloud Computing, Task Scheduling, Resource Allocation, Makespan Optimization, Min-Min Algorithm.

Abstract: Distributed systems, a priori intended for applications by connecting distributed entities, have evolved into

supercomputing to run a single application. Currently, Cloud Computing has arisen as a new trend in the

world of IT (Information Technology). Cloud computing is an architecture in full development and has

become a new computing model for running scientific applications. In this context, resource allocation is

one of the most challenging problems. Indeed, assigning optimally the available resources to the needed

cloud applications is known to be an NP complete problem. In this paper, we propose a new task scheduling

strategy for resource allocation that minimizes the completion time (makespan) in cloud computing

environment. To show the interest of the proposed solution, experiments are conducted on a simulated

dataset.

1 INTRODUCTION

Cloud computing consists of hardware and software

resources made available on the Internet as managed

third-party services. Cloud computing is a promising

paradigm given the increasing demand of high

performance computing and data storage. Today,

most companies have switched to cloud computing

solutions for data hosting and applications running

in order to take advantage of its three main services:

Software as a Service (SaaS), Infrastructure as a

service (IaaS) and Platform as a Service (PaaS).

Generally, distributed computing environments aim

at running different computational intensive services

that have diverse computational requirements by

ensuring the scalability, availability and resource

utilization. In cloud systems, the same application

can be used by multi-clients at the same time while

preserving security and private data of each client.

This is made possible by using virtualization

principle to share a resource between several users.

Resources management in Cloud computing

environment is one of the most important

challenges. Resource allocation problem is defined

as the process of assigning available resources to the

needed cloud application over the internet

(Vinothina et al., 2012). In literature, several

solutions (Jeyaram and Vidhya, 2013; Keshk, 2014)

for resource allocation problems in cloud computing

system have been proposed. The optimality criterion

in resource allocation is to minimize the global

completion time (Makespan) while maximizing the

resources utilization.

The scheduling problem can be either static or

dynamic. Task scheduling algorithms have been

designed as NP-complete problems that do not

converge to optimal solutions. The scheduling

strategy defines the instants when the scheduling

algorithm is called to produce a schedule based on

forecasting resources performances and independent

tasks to be executed. The task scheduler in cloud

computing environment is to determine an

appropriate assignment of resources to tasks in order

to meet up the needs of clients. A large number of

jobs scheduling heuristics are available for

maximizing profit via resources allocation in cloud

computing systems (Kuribayashi, 2011; Abirami and

Shalini, 2012; Gouda et al., 2013).

In this paper, we tackle the problem of resource

allocation and tasks scheduling problem for cloud

computing systems. The Resource allocation

problem in cloud computing environments has

become a very active research topic. Task

scheduling in cloud environment consists in

mapping the task onto the available resources and

the main objective is to maximize resources

utilization and to minimize the total completion

time. The task scheduling approach, proposed in this

Bey, K., Benhammadi, F., Boudaren, M. and Khamadja, S.

Load Balancing Heuristic for Tasks Scheduling in Cloud Environment.

DOI: 10.5220/0006240304890495

In Proceedings of the 19th International Conference on Enterprise Information Systems (ICEIS 2017) - Volume 1, pages 489-495

ISBN: 978-989-758-247-9

489

paper, relies on a load balanced makespan-based

heuristic to achieve a good task mapping on

available resources. Jobs are decomposed into

various tasks where each task should be assigned to

the best suited resources for its execution.

The remainder of the paper is organized as

follows. Section 2 presents related works about task

scheduling algorithms in cloud environment. The

problem of tasks scheduling for resource allocation

and the new meta-task scheduling algorithm in cloud

computing environments are proposed in Section 3.

Simulation results are presented in Section 4; and

Section 5 concludes the paper.

2 RELATED WORKS

Cloud computing is one of the emerging

technologies in the distributed computing

environment to meet the scientific applications

demands in terms of computing power and data

storage. Cloud computing systems have become an

active research topic in recent years. In literature,

plethora of heuristic algorithms for resource

allocation and scheduling (Goudarzi and Pedram,

2011; Delhi and Giridhar, 2014; Beghdad-Bey et al.,

2015)

 has been proposed. The aim of these

scheduling techniques is to determine an optimal

allocation of cloud resources, where resources are

scheduled based on user requests.

(Goudarzi and Pedram, 2011) presented a new

approach based on SLA (Service Level Agreements)

for resource allocation problem in cloud along with

a distributed solution to resolve this problem. More

explicitly, authors present a mathematical model

based on Generalized Processor Sharing and a

heuristic algorithm in order to maximize the profit

for resource allocation. An online scheduling for

resource allocation and task scheduling problem in

cloud is presented in (Keshk, 2014). The author

proposes a load balancing approach using ant colony

algorithm to improve the resource allocation.

Similarly, (Tawfeek et al., 2015) proposed a new

task scheduling algorithm in cloud based on

 Ant

Colony Optimization (ACO). Authors present a

comparison between the proposed approach and

existing scheduling algorithms as First Come First

Served (FCFS) and Round-Robin (RR). They

demonstrate through simulation results that ACO

algorithm outperforms both algorithms. (Abirami

and Shalini, 2012) design a scheduling algorithm

named Linear Scheduling for Tasks and Resources

(LSTR) in order to improve task scheduling and

resource utilization. To establish the IaaS cloud

environment, a combination of Nimbus and

Cumulus services are imported to a resource node.

The virtualization technique used with the

scheduling algorithm yields higher resource

utilization, and improves the cloud resources

performance.

On the other hand, an innovative admission

control and scheduling algorithms for efficient

resource allocation are introduced by (Irugurala and

Chatrapati, 2013) to maximize profit by minimizing

cost and improving customer level. (Gouda et al.,

2013) use a priority algorithm, based on some

optimum threshold decided by the cloud owner,

which decides the allocation sequence for different

jobs requested among the different users. The same

idea can be applied to allocate available resource

with minimum wastage and maximum profit. (Silva,

2008) optimize the machines number allocated for

processing an analytical task so that maximum

speedup can be achieved within a limited budget.

Since web applications traffic is dynamic and

random, forecasting optimal number of machines

allocated to client applications in real-time is not a

trivial task.

The PSO (particle swarm optimization)

algorithm has also been applied for task scheduling

in cloud computing. (Guo, 2013) introduced a PSO

algorithm based on small position value rule in order

to minimize the cost of the processing. (Gomathi and

Karthikeyan, 2013) propose Hybrid Particle Swarm

Optimization (HPSO) based scheduling heuristic to

balance the load across the entire system while

trying to minimize the makespan and maximize the

resource utilization. Conventional Min-min and

Max-min algorithms are used (Katyal and Mishra,

2014) for task scheduling. Authors apply several

heuristics to select the best suited among the two

algorithms, in terms of overall makespan, for on-

demand resources allocation. In the same context,

authors also consider the resource utilization

criterion. (Kumar and Verma, 2012) describe a new

scheduling algorithm that uses three scheduling

algorithms such as Min-Min, Max-Min and Genetic

Algorithm. The results obtained show that the

improved genetic algorithm can minimizes the

completion time. (Fang, 2010) develop a new task

scheduling algorithm based on load balancing. This

approach can ensure high resource utilization based

on simulation results of CloudSim. Moreover, other

techniques about task scheduling are detailed

(Selvarani and Sadhasivam, 2010; Kanrar, 2012).

ICEIS 2017 - 19th International Conference on Enterprise Information Systems

490

3 PROPOSED BALANCED TASK

SCHEDULING ALGORITHM

The development and deployment of distributed

applications in cloud computing environments is

being challenged by the need of scheduling

efficiently a large number of tasks and resources. In

this work, we present a new scheduling strategy

based load balancing (LBE) approach for

independent task in Cloud environment. This

heuristic aim to minimize the total completion time

(Makespan) and improve the utilization of resource.

3.1 Formulation and Modeling

Problem

Cloud environment is a promising distributed

platform to meet the increasing computational

requirements of scientific and technical applications.

Cloud computing is a new computing paradigm in

which these components consist of physical and

virtual resources. The dynamic resources allocation

is the most important problem in this area.

Before defining and modeling the task

scheduling problem, let us describe the cloud

architecture considered in this work. We define the

cloud computing architecture as a set of clusters

containing a different number and type of computing

resources each. Therefore, each resource is

constituted of a different number and types of virtual

machines (VMs). Figure 1 illustrates the functional

architecture of the Cloud computing system which is

composed of three main layers:

- Client Layer

: represents the request interface that

allows each client to execute their service request

and to display the final result. The client submits

service requests to the cloud via cloud manager for

processing.

- Cloud Manager Layer: has a very important role.

It is installed in the intermediary layer and is

responsible for coordination between the cloud

service provider and clients. Cloud Manager

receives the host resource data from the monitoring

module to elaborate its distribution solution. It

selects the servers in each cluster that participate in

calculation, and distributes the service requests for

processing according to the computing power

availability. Finally, it receives the results from

different servers which are sending to the user.

- Cloud Resource Layer: is composed of a set of

clusters of different types. Each cluster includes a

different number and types of physical computing

servers that provide hardware infrastructure. Client’s

applications will be deployed and executed in this

environment.

In this work, we consider that each job is

assigned to only one cluster and each cluster is

composed of different number of heterogeneous

computing resources or VMs (Virtual Machines).

After clients submit their jobs into the cloud

computing environment through the service request

manager, Cloud controller divides them into several

independent tasks. The task scheduler aims at

determining a suitable assignment of resources to

job tasks to complete all jobs received from clients.

In our model, the aim of task scheduling is to

execute N Jobs on M clusters efficiently, the result

of which should satisfy clients’ expectations. Each

Job is decomposed into several independent tasks

={T

, T

, …, T

} that are assigned to a set of

resources in the same cluster C

={R

, R

, …, R

We assume that each cluster can estimate how

Figure 1: Cloud computing Architecture considered in this problem.

Load Balancing Heuristic for Tasks Scheduling in Cloud Environment

491

much time is required to perform each job. Expected

Time to Compute (ETC) is used to estimate the

expected execution time of job J

on the cluster C

ETC is an n×m matrix; where n is the number of

jobs and m is the number of clusters. The number of

resources for each cluster is denoted R

. The

objective is to obtain an optimal task mapping on

resources by minimizing completion time of tasks

(makespan) while maximizing the use of resources.

The makespan is equal to maximum completion

time among all tasks and can be estimated using the

following equation:



BETCMakespan







max

,...,1

where J

is the set of jobs assigned to cluster C

and B

is a pseudo-Boolean integer used to

determine whether the job of i

client is assigned to

the j

cluster :









otherwise

clusterto the j assigned client isof i if the job

thth

In order to maximize the use of resources, we

have used a new parameter

that represents the

average number of resources used and which is

defined as the sum of all resource utilizations in each

cluster divided by the total number of resources





RRR



Subject to:

 i represents the client index



















: The tasks of the i

client assigned to the k

resource

: The processing capacity of the k

resource allocated

to the i

client

3.2 Proposed Scheduling Approach

The problem of task scheduling is a challenge in

cloud environment that aims at minimizing the

completion time of tasks and maximizing the use of

resource processing capabilities. Task scheduling

strategy takes into account the processing

requirements for each task and the mapping

mechanism of these tasks according to the

processing capabilities of available resources.

A heuristic-based load balancing solution is

presented for task scheduling problem in cloud

computing environment. The proposed algorithm

uses an initialization algorithm in its first step to

schedule all tasks on the available resources. The

principle of the proposed approach consists of

swapping the tasks between two machines showing

the maximum and minimum execution times in

order to minimize the makespan. Then, in the second

step, we generalize this exchanging technique over

all machines until the makespan value does not

change, as shown in Figure 2. The same algorithm is

used to assign jobs on clusters.

4 EXPERIMENTAL RESULTS

In this section, we present a set of experiments

realized by CloudSim simulator to assess the

proposed heuristic with respect to known task

scheduling algorithms, like Min-min, in cloud

computing environment. Two series of simulations

have been carried out: the first uses ETC matrices

randomly generated by the CloudSim Simulator,

whereas the second series uses reference data with

12 different types of ETC matrix up to 16 and 32

heterogeneous machines, and up to 512 and 1024

randomly generated heterogeneous tasks.

For each randomly generated situation, the first

step of the proposed algorithm consists in assigning

tasks to resources in order to find a good initial

scheduling. The aim of the second step is to improve

the initial solution. Hence, the aim of the load

balanced approach is twofold: (i) to decrease the

makespan; and (ii) to increase the resources

utilization. Table 1 shows the obtained comparison

results between the proposed approach and the Min-

Min algorithm. We used for these experiments, ETC

matrices of 10 machines and up to 100 and 900

tasks.

In the second series of simulations, we show the

interest of the proposed load balancing strategy with

respect to the Min-Min heuristic. All experimental

results represent the average makespan over 10

different ETC matrices of the same property. Figure

3 shows the makespan distribution for 12 different

types of ETC matrix up to 32 heterogeneous

machines, and up to 1024 randomly generated

heterogeneous tasks (all heterogeneity cases). Table

2 shows the comparison results between the

proposed heuristic and the Min-min algorithm on

makespan value for 12 heterogeneity cases.

ICEIS 2017 - 19th International Conference on Enterprise Information Systems

492

Figure 2: Proposed heuristic algorithm for Task scheduling.

Instances are labeled as Tt-Mm-Cc as follows: T

indicates the tasks heterogeneity (l: low and h: high

M represents the machine heterogeneity

and C

shows the type of consistency; c: consistency; i:

inconsistency and s: semi-consistency. In this work,

we consider that the most important criterion is

minimizing the makespan value to measure the

quality of schedule.

Table 1: Simulation results of makespan for min-min and

proposed algorithm.

Number of

Cloudlets

Min-Min

Algorithm

Proposed

algorithm

100

4879 4285

200 10485 5017

300 15102 7242

400 21476 10882

500 24950 12185

600 26489 13752

700 29225 16537

800 34896 23394

900 33648 27548

Figure 3 shows also the makespan distribution

for 12 different types of ETC matrix up to 32

heterogeneous machines, and up to 1024 randomly

generated heterogeneous tasks (all heterogeneity

cases). Each of the four heterogeneity cases has

means of {Min-Min (0.1311), LBE (0.004)}, {Min-

Min (0.0129), LBE (0.007)}, {Min-Min (0.0209),

LBE (0.0111)} and {Min-Min (0.1294), LBE

(0.0219)}. Their standard deviations are {0.1210,

0.005}, {0.1358, 0.014}, {0.1114, 0.0137} and

{0.0994, 0.0252} respectively.

Clearly for all heterogeneity cases, the points of

the makespan distribution obtained by the Min-Min

heuristic are completely dispersed over the points of

the makespan values obtained by our algorithm for

all machines. The completion times obtained by the

Min-min heuristic have a large standard deviation

which indicates that the completion times of the

machines are far from the mean. Our completion

time has a much smaller standard deviation than the

other two because its values are all close to the

mean. This indicates that the completion times are

clustered closely around the mean and the refined of

the completion time values are very close.

// Initialization step: Assign task to the resource

1) Determinate the available performance for all resource

2) Compute the Execution Time (ETC) for each task on all resource

3) Sort the tasks in ascending order of their ETC

4) Assign tasks to the resources with gives minimum execution time

5) Find the initial tasks scheduling matrix M

// Improvement heuristic

Select the resource R

which fives the maximum completion time C

max

from M

while C

max

decrease do

for all assigned tasks T

to the resource R

for all R

j’

≠ R

for all assigned tasks T

to the resource R

j’

swap task T

with T

update C

i j’

= r

j’

+ ETC

i j’

– ETC

k j’

update C

k j

= r

+ ETC

k j

– ETC

i j

if (C

i j’

< C

max

) and (C

k j

< C

max

) then

confirm the swapping of task T

with T

update M

update C

max

end

// C

i j

: expected completion time of T

on resource R

;

// r

: expected time of resource R

;

// ETC

i j

: compute execution time of task T

on resource R

;

Load Balancing Heuristic for Tasks Scheduling in Cloud Environment

493

a. Low task and low machine heterogeneity cases. b. Low task and high machine heterogeneity cases.

c.High task and low machine heterogeneity cases. d. High task and high machine heterogeneity cases.

Figure 3: Makespan values distribution for different heterogeneity cases.

Table 2: Comparaison of makespans results.

Instance

Min-Min

algorithm

Min-Min

+ AG

Proposed

algorithm

ThMhCc

9267258 9219257

9173984

ThMhCi

17457713

1463582

10645286

ThMhCs

16458962

1256743

10548555

ThMlCc

564889 250384

181867

ThMlCi

143908 129983

121159

ThMlCs

744679 572668

446259

TlMhCc

690435 398865

315852

TlMhCi

3097331 161032

130576

TlMhCs

1346356 521061

434871

TlMlCc

18738 6752

5963

TlMlCi

4306 3971

3430

TlMlCs

23868 18043

12590

Experimental results reported in Table 2 show that

under high task and machine heterogeneities, our

algorithm improves the makespan from 1% to 39%

in all consistency situations. However, in low task

and high machine heterogeneities, our algorithm

gives better quality scheduling solutions between

54.25% and 95.78% than Min-min algorithm in all

consistency cases. This indicates that the proposed

scheduling algorithm performs well compared with

Min-Min heuristic scheduling algorithm in terms of

performance and makespan (improvements between

15..80% and 95.78%) except in the consistent case

for high tasks and machines (1%).

5 CONCLUSIONS

In this paper, we have proposed a new task

scheduling heuristic based on makespan

improvement in cloud computing environment. The

proposed approach generalizes the process of tasks

exchange between both machines with the maximum

and minimum makespan over all machines until the

makespan value does not change. Our approach has

then been assessed with respect to the Min-min

algorithm. The results obtained have shown

significant improvements in terms of execution time,

makespan and resources utilization. An interesting

future direction would be to evaluate our proposed

solution against other State-of-the-Art task

scheduling algorithms.

ICEIS 2017 - 19th International Conference on Enterprise Information Systems

494

REFERENCES

Abirami, S.P. and Shalini, R. 2012. Linear Scheduling

Strategy for Resource Allocation in Cloud

Environment. International Journal on Cloud

Computing: Services and Architecture (IJCCSA),

Vol.2, No.1, February 2012.

Beghdad-Bey, K., Benhammadi, F., Sebbak F. and

Maataoui, M. 2015. New Tasks Schedeling Strategy

for resources Allocation in Cloud Computing

Environment. Sixth International Conference on

Modeling, Simulation and Applied Optimization,

ICMSAO’2015, May 27-29, 2015.

Delhi Babua, K. and Giridhar Kumar, D. 2014. Allocation

Strategies of Virtual Resources in Cloud-Computing

Networks. International Journal of Engineering

Research and Applications, Vol. 4, Issue 11,

November 2014, pp.51-55.

Fang, Y., Wang, F. and Ge, J.2010. A Task Scheduling

Algorithm Based on Load Balancing in Cloud

Computing. Web Information Systems and Mining,

Lecture Notes in Computer Science, Vol. 6318, 2010,

pages 271-277.

Gomathi, B. and Karthikeyan, K. 2013. Task scheduling

algorithm based on hybrid Particle swarm

optimization in Cloud Computing environment.

Journal of Theoretical and Applied Information

Technology, September 2013. Vol. 55 N° 1.

Gouda,K. C., Radhika, T. V. and Akshatha, M. 2013.

Priority based resource allocation model for cloud

computing. International Journal of Science,

Engineering and Technology Research (IJSETR),

Volume 2, Issue 1, January 2013.

Goudarzi, H. and Pedram, M. 2011. Maximizing Profit in

Cloud Computing System via Resource Allocation.

IEEE 31st International Conference on Distributed

Computing Systems Workshops 2011: pp, 1-6.

Guo, L. 2012. Task Scheduling Optimization in Cloud

Computing Based on Heuristic Algorithm”, Journal of

Networks, vol. 7, NO. 3 march 2012.

Irugurala, S. and Chatrapati, K.S. 2013. Various

Scheduling Algorithms for Resource Allocation. In

Cloud Computing. The International Journal of

Engineering And Science (IJES), Volume 2, Issue 5,

Pages 16-24, 2013.

Jeyaram, G., and Vidhya, V., 2013. Efficient resource

allocation strategies for cloud data centers.

International Journal of Advanced Research in

Computer and Communication Engineering Vol. 2,

Issue 2, February 2013.

Kanrar, S. 2012. Enhancement of job allocation in private

Cloud by distributed processing. Proceedings of the

Second International Conference on Computational

Science, Engineering and Information Technology

(CCSEIT'12), Pages 94-98, ACM NY 2012.

Katyal, M. and Mishra, A. 2014. Application of Selective

Algorithm for Effective Resource Provisioning In

Cloud Computing Environment. International Journal

on Cloud Computing: Services and Architecture

(IJCCSA), Vol. 4, No. 1, 2014.

Keshk, A. E, 2014. Cloud Computing Online Scheduling.

IOSR Journal of Engineering (IOSRJEN), Vol. 04,

Issue 03, March. 2014.

Kumar, P. and Verma, A. 2012. Scheduling Using

Improved Genetic Algorithm in Cloud Computing for

Independent Tasks. Proceedings of the International

Conference on Advances in Computing,

Communications and Informatics (ICACCI '12), Pages

137-142, ACM, NY 2012.

Kuribayashi, S. 2011. Optimal Joint Multiple Resource

Allocation Method for Cloud Computing

Environments. International Journal of Research and

Reviews in Computer Science (IJRRCS), Vol. 2, No.

1, March 2011.

Selvarani, S. and Sadhasivam, G.S. 2010. Improved cost-

based algorithm for task scheduling in Cloud

computing. Computational Intelligence and

Computing Research (ICCIC), IEEE, pp.1-5, 2010.

Silva, J.N., Veiga, L. and Ferreira, P. 2008. Heuristic for

resources allocation on utility computing

infrastructures. MGC’08 Proceedings of the 6th

International Workshop on Middleware for Grid

Computing, ACM, NY, USA, 2008, pp. 1–6.

Tawfeek, M., El-Sisi, A., Keshk, A. and Torkey, F.2015.

Cloud Task Scheduling Based on Ant Colony

Optimization. International Arab Journal of

Information Technology, Vol. 12, No. 2, 2015.

Vinothina, V., Sridaran, R. and Ganapathi, P. 2012. A

Survey on Resource Allocation Strategies in Cloud

Computing. International Journal of Advanced

Computer Science & Applications, 2012, vol. 3, n°6.

Load Balancing Heuristic for Tasks Scheduling in Cloud Environment

495