Article can be accessed online at http://www.publishingindia.

com

A Comparative Study of Load Balancing

Algorithms in Cloud Computing Environment
Mayanka Katyal*, Atul Mishra**

Abstract Computation in cloud is done with the aim to achieve

maximum resource utilization with higher availability at
Cloud Computing is a new trend emerging in IT minimized cost.
environment with huge requirements of infrastructure
and resources. Load Balancing is an important
aspect of cloud computing environment. Efficient load 1.1. Cloud v/s Cluster and Grid
balancing scheme ensures efficient resource utilization
by provisioning of resources to cloud user’s on-demand Clusters [1] are parallel and distributed systems, governed
basis in pay-as-you-say-manner. Load Balancing may under the supervision of single administrative domain.
even support prioritizing users by applying appropriate
The node (stand-alone computers) in the cluster integrates
scheduling criteria. This paper presents various load
balancing schemes in different cloud environment to form a single computing resource.
based on requirements specified in Service Level
Agreement (SLA).
Grid [1] is aggregation of autonomous resources that
are geographically distributed. The nodes in grid permit
Keywords: Cloud Computing, Load Balancing, sharing and selection dynamically at runtime.
Resource Provisioning, Resource Scheduling, Service
Level Agreement (SLA)
Clouds [1], [2] are not the combination of clusters and
grid but are next generation to clusters and grid. Similar
to Cluster and Grid, Cloud is also a collection of parallel
and distributed systems. Cloud is not a single domain.
1. Introduction Unlike cluster and grid, cloud has multiple domains and
the nodes of cloud are “Virtualized” [9].
Cloud Computing is made up by aggregating two terms in
the field of technology. First term is Cloud and the second 1.2. Cloud Perspectives
term is computing. Cloud is a pool of heterogeneous
resources. It is a mesh of huge infrastructure and has Cloud has different meaning to different stakeholders.
no relevance with its name “Cloud”. Infrastructure There are three main stakeholders of cloud:
refers to both the applications delivered to end users as
services over the Internet and the hardware and system 1.2.1. End Users
software in datacenters that is responsible for providing
those services. In order to make efficient use of these These are the customers or consumers of cloud. They use
resources and ensure their availability to the end users the various services (Infrastructure/ Software/Platform)
“Computing” is done based on certain criteria specified provided by the cloud. Before using the cloud services, the
in SLA. Infrastructure in the Cloud is made available to users of cloud must agree to the Service Level Agreement
the user’s On-Demand basis in pay-as-you-say-manner. (SLA) specified by the Cloud Provider.

*
Student, Department of Computer Engineering, YMCA University of Science and Technology, Faridabad, Haryana, India.
E-mail: [email protected]
** Associate Professor, Department of Computer Engineering, YMCA University of Science and Technology, Faridabad,
Haryana, India. E-mail: [email protected]
6 International Journal of Distributed and Cloud Computing Volume 1 Issue 2 December 2013

They use the services on demand basis and have to pay of resources that make up cloud and providing these
for the services availed depending upon their usage. resources to the end users. Several provisioning related
Cloud provides its users flexibility in availing its services issues are mentioned in Table 1.
by incorporating utility computing.
Table 1:    Stakeholders of Cloud
Prior to signing of SLA, the users of cloud must verify that
SLA contains certain Quality-of-Service (QoS) parameters
Type of stakeholder Requirements/Issues
which are pre-requisites of the consumer, before using
cloud services. Some of the basic requirements or issues Security
of cloud users is listed in Table 1. Provenance
Privacy
End User
Hence, for end user, Cloud Computing is a scenario where High Availability
the user can have access to any kind of infrastructure, Reduced Cost
software or platform in a secure manner- at reduced cost- Ease-of-use
on demand basis-in an easy to use manner. Managing Resources
Outsourcing
1.2.2. Cloud Provider Resource Utilization
Energy Efficiency
Cloud provider can offer either public or private or hybrid Cloud Provider Metering
cloud. They are responsible for building of the cloud. Providing Resources
Cost Efficiency
Private clouds [4] are owned by enterprises or business
Meet end user requirements
for their internal use. They may use it to store and manage
Utility Computing
Big-Data of their organization or to provide enough
Elasticity/ Scalability
resources on demand basis to its team of employees or
Virtualization
clients. They offer greatest level of security.
Agility and Adaptability
OpenStack [5], VMware [6] and CloudStack [7] are Cloud Developer Availability
private clouds. Data Management
Reliability
Public clouds [4] may be used by individuals or Programmability
an organization based upon their requirements and
necessities. They offer greatest level of efficiency in 1.2.3. Cloud Developer
shared resources. Confidentiality is the major security
issue in using public cloud. They are more vulnerable This entity lies between end user and cloud provider.
than private clouds. Amazon web services [8], Google Cloud developer has the responsibility of taking into
Compute Engine [9], Microsoft Azure [10], HP cloud [11] consideration both the perspectives of the cloud (i.e. view
are some of the public clouds. of end user and cloud provider). The developer of cloud
must adhere to all the technical details of the cloud which
A hybrid cloud [4] is a combination of public and private
are essential to meet the requirements of both, the cloud
cloud. It allows businesses to manage some resources
user as well as the cloud provider. Some of the basic
internally within organization and some externally. The
issues that cloud developer must focus on are given in
downside is that the complexity of overall management
Table 1. Main motive of the developer is to bridge the gap
increases along with security concerns. To optimize the
between the end user of the cloud and the cloud provider.
use of one or more combination of private or public
clouds CliQr [11] allows the businesses to accommodate This paper discusses different Load Balancing schemes
changing needs of users. in cloud computing environment. The rest of the paper is
organized as follows: Section 2 discusses load balancing
Cloud provider must accomplish its job of “resource
in cloud computing environment. Section 3 discusses
provisioning”. Resource provisioning includes two
some related work to load balancing algorithms. We
main tasks. These include managing of huge bundle
conclude our study in Section 4.
 A Comparative Study of Load Balancing Algorithms in Cloud Computing Environment 7

2. Load Balancing in Cloud of Virtual Machine are mapped to same instance of a

Host based upon its availability. Application and System
Computing Environment software are executed on Virtual Machine on-demand.
Load balancing in cloud computing provides an efficient Class diagram of Cloud architecture illustrating
solution to various issues residing in cloud computing relationship between the four basic entities is shown in fig
environment set-up and usage. Load balancing must 1. Thus, the object oriented approach of CloudSim can be
take into account two major tasks, one is the resource used to simulate Cloud Computing environment.
provisioning or resource allocation and other is task
scheduling in distributed environment. Efficient Figure 1.    Class Diagram of Cloud
provisioning of resources and scheduling of resources as
well as tasks will ensure:
a. Resources are easily available on demand.
b. Resources are efficiently utilized under condition of
high/low load.
c. Energy is saved in case of low load (i.e. when usage
of cloud resources is below certain threshold).
d. Cost of using resources is reduced.

For measuring the efficiency and effectiveness of Load

Balancing algorithms simulation environment are
required. CloudSim [12] is the most efficient tool that can
2.1. Resource Allocation
be used for modeling of Cloud. During the lifecycle of a
Cloud, CloudSim allows VMs to be managed by hosts
Resource provisioning is the task of mapping of the
which in turn are managed by datacenters.
resources to different entities of cloud on demand basis.
Cloudsim provides architecture with four basic entities. Resources must be allocated in such a manner that no node
These entities allow user to set-up a basic cloud in the cloud is overloaded and all the available resources in
computing environment and measure the effectiveness the cloud do not undergo any kind of wastage (wastage of
of Load Balancing algorithms. A typical Cloud modeled bandwidth or processing core or memory etc.). Mapping
using CloudSim consists of following four entities of resources to cloud entities is done at two levels:
Datacenters, Hosts, Virtual Machines and Application
as well as System Software. Datacenters entity has the 2.1.1. VM Mapping onto the Host
responsibility of providing Infrastructure level Services
to the Cloud Users. They act as a home to several Host Virtual machines reside on the host (physical servers).
Entities or several instances hosts’ entities aggregate More than one instance of VM can be mapped onto a
to form a single Datacenter entity. Hosts in Cloud are single host subject to its availability and capabilities.
Physical Servers that have pre-configured processing Host is responsible for assigning processing cores to
capabilities. Host is responsible for providing Software VM. Provisioning policy define the basis of allocating
level service to the Cloud Users. Hosts have their own processing cores to VM on demand. Allocation policy or
storage and memory. Processing capabilities of hosts algorithm must ensure that critical characteristics of Host
is expressed in MIPS (million instructions per second). and VM do not mismatch.
They act as a home to Virtual Machines or several
instances of Virtual machine entity aggregate to form a 2.1.2 Application or Task Mapping onto VM
Host entity. Virtual Machine allows development as well
as deployment of custom application service models. Applications or tasks are actually executed on VM.
They are mapped to a host that matches their critical Each application requires certain amount of processing
characteristics like storage, processing, memory, software power for their completion. VM must provide required
and availability requirements. Thus, similar instances processing power to the tasks mapped onto it. Tasks
8 International Journal of Distributed and Cloud Computing Volume 1 Issue 2 December 2013

Table 2:    Comparison between Resource Allocation and Task Scheduling

Task Sub-Category Issues Resolved Provider Oriented

Customer
Oriented
Efficient Utilization
At host level
Resource Allocation Minimize Makespan Yes Yes
At VM level
Ensure Availability
Space-Sharing
Task Scheduling Minimize overall response time No Yes
Time-Sharing

must be mapped onto appropriate VM based upon its Case 2: Hosts and VMs, both are provisioned to VMs and
configuration and availability. tasks respectively in time sharing manner.

Case 3: Hosts are provisioned to VMs in space sharing

2.2. Task Scheduling manner and VMs are provisioned to tasks in time sharing
manner.
Task scheduling is done after the resources are allocated to
all cloud entities. Scheduling defines the manner in which Case 4: Hosts are provisioned to VMs in time sharing
different entities are provisioned. Resource provisioning manner and VMs are provisioned to tasks in space sharing
defines which resource will be available to meet user manner.
requirements whereas task scheduling defines the manner
in which the allocated resource is available to the end 3. Related Work to Load
user (i.e. whether the resource is fully available until Balancing Algorithms
task completion or is available on sharing basis). Task
scheduling provides “Multiprogramming Capabilities” in Cloud is made up of massive resources. Management of
cloud computing environment. these resources requires efficient planning and proper
layout. While designing an algorithm for resource
Task scheduling can be done in two modes:
provisioning on cloud the developer must take into
a. Space shared consideration different cloud scenarios and must be
b. Time shared aware of the issues that are to be resolved by the proposed
algorithm. Therefore, resource provisioning algorithm
Both hosts and VM can be provisioned to users either in can be categorized into different classes based upon the
space shared mode or time shared mode. environment, purpose and technique of proposed solution.
In space sharing mode resources are allocated until task
does not undergo complete execution (i.e. resources are not 3.1. Load Balancing on the basis of Cloud
preempted); whereas in time sharing mode resources are Environment
continuously preempted till task undergoes completion.
Cloud computing can have either static or dynamic
Table 2. gives the comparison of resource allocation and
environment based upon how developer configures the
task scheduling and specifies the issues resolved by each
cloud demanded by the cloud provider.
technique of load balancing.

Based on resource provisioning and scheduling, four cases 3.1.1. Static Environment
can be examined under different performance criteria so
as to get efficient load balancing scheme. In static environment the cloud provider installs
homogeneous resources. Also the resources in the cloud
Case 1: Hosts and VMs, both are provisioned in space are not flexible when environment is made static. In this
sharing manner. scenario, the cloud requires prior knowledge of nodes
capacity, processing power , memory, performance and
 A Comparative Study of Load Balancing Algorithms in Cloud Computing Environment 9

statistics of user requirements. These user requirements 3.2. Load Balancing based on Spatial
are not subjected to any change at run-time. Algorithms Distribution of Nodes
proposed to achieve load balancing in static environment
cannot adapt to the run time changes in load. Although Nodes in the cloud are highly distributed. Hence the node
static environment is easier to simulate but is not well that makes the provisioning decision also governs the
suited for heterogeneous cloud environment. category of algorithm to be used. There can be three types
Round Robin algorithm [13] provides load balancing in of algorithms that specify which node is responsible for
static environment. In this the resources are provisioned balancing of load in cloud computing environment.
to the task on first-cum-first-serve (FCFS- i.e. the task that
entered first will be first allocated the resource) basis and
3.2.1. Centralized Load Balancing
scheduled in time sharing manner. The resource which is
least loaded (the node with least number of connections) In centralized load balancing technique all the allocation
is allocated to the task. Eucalyptus uses greedy (first-fit) and scheduling decision are made by a single node. This
with round-robin for VM mapping. node is responsible for storing knowledge base of entire
cloud network and can apply static or dynamic approach
Radojevic proposed an improved algorithm over round for load balancing. This technique reduces the time
robin called CLBDM (Central Load Balancing Decision required to analyze different cloud resources but creates a
Model) [14]. It uses the basis of round robin but it also great overhead on the centralized node. Also the network
measures the duration of connection between client and is no longer fault tolerant in this scenario as failure
server by calculating overall execution time of task on intensity of the overloaded centralized node is high and
given cloud resource. recovery might not be easy in case of node failure.

3.1.2. Dynamic Environment 3.2.2. Distributed Load Balancing

In dynamic environment the cloud provider installs In distributed load balancing technique, no single node
heterogeneous resources. The resources are flexible in is responsible for making resource provisioning or task
dynamic environment. In this scenario cloud cannot rely scheduling decision. There is no single domain responsible
on the prior knowledge whereas it takes into account for monitoring the cloud network instead multiple
run-time statistics. The requirements of the users are domains monitor the network to make accurate load
granted flexibility (i.e. they may change at run-time). balancing decision. Every node in the network maintains
Algorithm proposed to achieve load balancing in dynamic local knowledge base to ensure efficient distribution of
environment can easily adapt to run time changes in load. tasks in static environment and re-distribution in dynamic
Dynamic environment is difficult to be simulated but is environment.
highly adaptable with cloud computing environment.
In distributed scenario, failure intensity of a node is
Based on WLC [15] (weighted least connection) not neglected. Hence, the system is fault tolerant and
algorithm, Ren proposed a load balancing technique in balanced as well as no single node is overloaded to make
dynamic environment called ESWLC. It allocates the load balancing decision.
resource with least weight to a task and takes into account
node capabilities. Based on the weight and capabilities Comparison of different static and dynamic load balancing
of the node, task is assigned to a node. LBMM (Load algorithms is given in Table 3. It also compares them on
Balancing Min-Min) algorithm proposed in paper [16] the basis of spatial distribution of nodes.
uses three level frameworks for resource allocation in A nature inspired solution is presented in paper [17] called
dynamic environment. It uses OLB (opportunistic load Honeybee Foraging for load balancing in distributed
balancing) algorithm as its basis. Since cloud is massively scenario. In Honeybee foraging the movement of ant
scalable and autonomous, dynamic scheduling is better in search of food forms the basis of distributed load
choice over static scheduling.
10 International Journal of Distributed and Cloud Computing Volume 1 Issue 2 December 2013

Table 3:    Comparison Table of Load Balancing Algorithms in Cloud Computing Environment

Dynamic Centralized Distributed Hierarchical

Algorithm Static Environment
Environment Balancing Balancing Balancing
Round-robin Yes No Yes No No
CLBDM[22] Yes No Yes No No
Ant Colony[20] No Yes No Yes No
Map Reduce[9] Yes No No Yes Yes
Particle Swarm Optimiza-
No Yes No Yes No
tion [21]
MaxMin[22] Yes No Yes No No
MinMin[22] Yes No Yes No No
Biased Random Sampling No Yes No Yes No
Active Clustering[18] No Yes No Yes No
LBMM No Yes No No Yes
OLB[23] Yes No Yes No No
WLC No Yes Yes No No
ESWLC No Yes Yes No No
Genetic Algorithm[24] No Yes Yes No No

balancing in cloud computing environment. This is a self 3.3. Load Balancing Based on Task
organizing algorithm and uses queue data structure for its Dependencies
implementation. Biased random sampling [18] is another
distributed load balancing technique which uses virtual Dependent tasks are those whose execution is dependent
graph as the knowledge base. on one or more sub-tasks. They can be executed only after
completion of the sub-tasks on which it is dependent.
3.2.3. Hierarchical Load Balancing Therefore, scheduling of such task prior to execution of
sub-tasks is in-efficient. Task dependency is modeled
Hierarchical load balancing involves different levels of using workflow based algorithms.
the cloud in load balancing decision. Such load balancing
techniques mostly operate in master slave mode. These Workflow basically uses DAG [25] as knowledge base
can be modeled using tree data structure wherein every to represent task dependency. Different workflow based
node in the tree is balanced under the supervision of its solution consider different parameters. Algorithm are
parent node. Master or manager can use light weight designed keeping in mind whether single or multiple
agent process to get statistics of slave nodes or child workflows are to be modeled or single or multiple QoS
nodes. Based upon the information gathered by the parent parameters are to be maintained in the system. Different
node provisioning or scheduling decision is made. workflows with or without completely different structure
are termed as multiple workflows. Workflows can also
Three-phase hierarchical scheduling proposed in paper be classified as Transaction Incentive (multiple instances
[19] has multiple phases of scheduling. Request monitor of one workflow that have same structure) and Data
acts as a head of the network and is responsible for Incentive workflows (size and quantity of data is large).
monitoring service manager which in turn monitor service
nodes. First phase uses BTO (Best Task Order) scheduling, Cost based scheduling algorithm in [26] is designed for
second phase uses EOLB (Enhanced Opportunistic single workflows. It partitions the workflows and assigns
Load Balancing) scheduling and third phase uses EMM each partition a deadline. Zhifeng Yu and Weisong Shi
(Enhanced Min-Min) scheduling. designed an algorithm for multiple workflows which
focus only on execution time. With an aim of maximizing
 Table 4:    Comparison of different Types of Load Balancing Scenarios in Cloud Computing Environment

Type of Algorithm Knowledge Base Issues to be addressed Usage Drawbacks

Response time
Not Flexible
Resource utilization
Not scalable
Prior knowledge base is required about Scalability Used in homoge-
Static
each node statistics and user requirements. Power consumption and Energy Utilization neous environment. Is not compatible with changing
user requirements as well as load
Makespan
Throughput/Performance
Location of processor to which load is trans-
ferred by an overloaded processor.
Run time statistics of each node are moni- Transfer of task to a remote machine.
Information Gathering. Used in heteroge- Complex
Dynamic tored to adapt to changing load require-
neous environment. Time Consuming
ments. Load estimation.
Limiting the number of migrations.
Throughput
Threshold policies
Throughput
Single node or server is responsible for Useful in small Not fault tolerant
Failure Intensity
Centralized maintaining the statistics of entire network networks with low Overloaded central decision mak-
and updating it from time to time. Communication between central server and load. ing node
processors in network.
Associated Overhead
Selection of processor that take part in load
balancing.
All the processors in the network respon- Migration time Useful in large and Algorithm complexity
sible for load balancing store their own lo-
Distributed Interprocessor communication heterogeneous en-
cal database (e.g. MIB) to make efficient Communication overhead
Information exchange criteria vironment.
balancing decisions.
Throughput
Fault tolerance
Threshold policies
Information exchange criteria Useful in medium
Nodes at different levels of hierarchy com- Selection of nodes at different levels of net- or large size net- Less fault tolerant
municate with the nodes below them to
Hierarchical work work with hetero-
get information about the network perfor-
Failure intensity geneous environ- Complex
mance.
Performance ment.
Migration time
A Comparative Study of Load Balancing Algorithms in Cloud Computing Environment
11
 12
Type of Algorithm Knowledge Base Issues to be addressed Usage Drawbacks
Type of workflow
Single workflow
Multiple workflow Used in modeling
Transaction incentive workflows of task dependen- Difficult to model
DAG is used to model dependencies of task Data incentive workflows
Workflow De- cies in any kind of Maintenance of knowledge base is
and can be used to make scheduling deci-
pendent environment (either complex.
sion.
Fault tolerance homogeneous or Higher Complexity
heterogeneous.)
Execution time
Makespan
Migration time
International Journal of Distributed and Cloud Computing
Volume 1 Issue 2 December 2013
 A Comparative Study of Load Balancing Algorithms in Cloud Computing Environment 13

throughput Ke Lie et al proposed scheduling strategy [3] Luo, S., Lin, Z. & Chenm, X. (2011). Virtualization
which is meant for transaction incentive workflows. security for cloud computing service. 2011
For clouds based on Hadoop CloudWF (computational International Conference on Cloud and Service
workflow system) encodes workflow blocks and block- Computing 978-1-4577-1637-9/11/$26.00 ©2011
to-block dependencies. Hadoop HBase sparse table is IEEE. Shenzhen, China: ZTE Corporation.
used to store information related to workflows. It is fault [4] Zhang, Q., Cheng, L. & Boutaba, R. (2010). Cloud
tolerant and uses map-reduce framework. computing: State-of-the-art and research challenges.
Journal of Internet Services and Applications, 1(1),
Table 4, compares different type of load balancing 7-18. DOI 10.1007/s13174-010-0007-6.
scenarios in cloud computing environment. It specifies [5] Open Stack: An Overview. Retrieved from www.
the knowledge base, usage and drawbacks of each type of openstack.org/downloads/openstack-overview-data-
algorithm and issues addressed by these algorithms. sheet.pdf.
[6] White Paper (2013). Measuring the Business Value
4. Conclusion of VMware Horizon View.
[7] Huang, A. Software Architect, Citrix Systems
Load Balancing is an essential task in Cloud Computing Apache Cloud-Stack Architecture.
environment to achieve maximum utilization of [8] Amazon Elastic Compute Cloud (EC2). Retrieved
resources. In this paper, we discussed various load from http://www.amazon.com/ec2/ (accessed on
balancing schemes, each having some pros and cons. On April 18, 2010).
one hand static load balancing scheme provide easiest [9] Google App Engine, http://appengine.google.com
simulation and monitoring of environment but fail to (April 18, 2010).
model heterogeneous nature of cloud. On the other hand, [10] Chappell, D. (2008). Introducing the Azure services
dynamic load balancing algorithm are difficult to simulate platform. White paper.
but are best suited in heterogeneous environment of cloud [11] CliQr Makes Cloud Benchmarking Available to All
computing. Also the level at node which implements this Users, Providing Critical Information Needed
static and dynamic algorithm plays a vital role in deciding to Optimize Application Deployment Cloud.
the effectiveness of algorithm. Unlike centralized Retrieved from www.cliqr.com/assets/downloads/
algorithm, distributed nature of algorithm provides better press/CliQrBenchmark.06.25.13.pdf.
fault tolerance but requires higher degree of replication [12] Calheiros, R. N., Ranjan, R., Beloglazov, A., Rose,
and on the other hand, hierarchical algorithm divide C. A. F. D. & Buyya, R. CloudSim: A Toolkit for
the load at different levels of hierarchy with upper Modeling and Simulation of Cloud Computing
level nodes requesting for services of lower level nodes Environments and Evaluation of Resource
in balanced manner. Hence, dynamic load balancing Provisioning Algorithms. extended version of a
techniques in distributed or hierarchical environment keynote paper: R. Buyya, R. Ranjan, and R. N.
provide better performance. However, performance of the Calheiros. Modeling and simulation of scalable
cloud computing environment can be further maximized Cloud computing environments and the CloudSim
toolkit: Challenges and opportunities. Proceedings
if dependencies between tasks are modeled using
of the Conference on High Performance Computing
workflows.
and Simulation (HPCS 2009) (pp. 21-24). IEEE
Press, New York, USA, Leipzig, Germany, June,
References 2009.
[1] Foster, I., Zhao, Y., Raicu, I. & Lu, S. (2008). [13] Sotomayor, B., Montero, R. S., Llorente, I. M. &
Cloud computing and grid computing 360-degree Foster, I. (2009). Virtual infrastructure manage-
compared. IEEE Grid Computing Environment ment in private and hybrid clouds. IEEE Internet
Workshops. Computing, 13(5), 14-22.
[2] Youseff, L. Toward a Unified Ontology of Cloud [14] Radojevic, B. & Zagar, M. (2011). Analysis of issues
Computing Retrieved from http://spsteve. wikispac- with load balancing algorithms in hosted (cloud)
es.asu.edu/file/view/unified+interface+cloud.pdf. environments. In proceedings of 34th International
Convention on MIPRO, IEEE.
14 International Journal of Distributed and Cloud Computing Volume 1 Issue 2 December 2013

[15] Lee, R. & Jeng, B. (2011). Load-balancing tactics in [21] Vesna Sesum-Cavic Institute of Computer Languages
cloud. In proc. International Conference on Cyber- Vienna University of Technology Wien, Austria, Eva
Enabled Distributed Computing and Knowledge Kühn. Applying swarm intelligence algorithms
Discovery (CyberC), IEEE, (pp. 447-454). for dynamic load balancing to a Cloud Based Call
[16] Wang, S. C., Yan, K. Q., Liao, W. P. & Wang, S. Center” 2010 Fourth IEEE International Conference
S. (2010). Towards a load balancing in a three- on Self-Adaptive and Self-Organizing Systems.
level cloud computing network. Proceedings of 3rd 978-0-7695-4232-4/10 $26.00 © 2010 IEEE DOI
International Conference on Computer Science and 10.1109/SASO.2010.19.
Information Technology (ICCSIT), IEEE, July, 1, [22] Naghibzadeh, M. (2007). A min-min max-min selec-
108-113. tive algorithm for grid task scheduling. 1-42440-
[17] Randles, M., Bendiab, A. T. & Lamb, D. (2008). 1007-X/07/$25.00, 2007 IEEE. Dept. of Computer
Cross layer dynamics in self-organising service Engineering Ferdowsi University of Mashad.
oriented architectures. IWSOS, Lecture Notes in [23] Al Nuaimi, K., Mohamed, N., Al Nuaimi, M. &
Computer Science, 5343, pp. 293-298, Springer. Al-Jaroodi, J. (2012). A survey of load balancing
[18] Randles, M., Lamb, D., Bendiab, A. T. (2010). A in cloud computing: challenges and algorithms.
Comparative Study into Distributed Load Balancing 2012 IEEE Second Symposium on Network Cloud
Algorithms for Cloud Computing. 2010 IEEE 24th Computing and Applications 978-0-7695-4943-9/12
International Conference on Advanced Information $26.00 © 2012 IEEE DOI 10.1109/NCCA.2012.29.
Networking and Applications Workshops 978-0- College of Information Technology, UAEU Al Ain,
7695-4019-1/10 $26.00 © 2010 IEEE. DOI 10.1109/ United Arab Emirates
WAINA.2010.85. [24] Zhao, C., Zhang, S., Liu, Q., Xie, J. & Hu, J. (2009).
[19] Shu-Ching, W., Yan, K. Q., Sheng, S. & Wei, W. C. Independent Tasks Scheduling Based on Genetic
(2011). A three-phases scheduling in a hierarchical Algorithm in Cloud Computing.
cloud computing network. 2011 Third International [25] Wu, Z., Liu, X., Ni, Z., Yuan, D. &Yang, Y. A.
Conference on Communications and Mobile Market-oriented hierarchical scheduling strat-
Computing 978-0-7695-4357-4/11 $26.00 © 2011 egy in cloud workflow systems. Journal of Super
IEEE DOI 10.1109/CMC.2011.2. Computing. DOI 10.1007/s11227-011-0578-4.
[20] Li, K., Xu, G., Zhao, G., Dong, Y. & Wang, D. (2011). [26] Xu, M., Cui, L., Wang, H. & Bi, Y. (2009). A mul-
Cloud task scheduling based on load balancing ant tiple QoS constrained scheduling strategy of mul-
colony optimization. 2011 Sixth Annual ChinaGrid tiple workflows for cloud computing. 2009 IEEE
Conference 978-0-7695-4472-4/11 $26.00 © 2011 International Symposium on Parallel and Distributed
IEEE. DOI 10.1109/ChinaGrid.2011.17. Processing with Applications 978-0-7695-3747-4/09
$25.00 © 2009 IEEE DOI 10.1109/ISPA.2009.95.