See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/334084499

Load balancing algorithms in software defined network

Article · April 2018

CITATIONS READS

0 1,093

3 authors:

Mustafa Hasan Albowarab Nurul Azma Zakaria

Technical University of Malaysia Malacca UniversitiTeknikal Malaysia Melaka
3 PUBLICATIONS   2 CITATIONS    38 PUBLICATIONS   34 CITATIONS   

SEE PROFILE SEE PROFILE

Zaheera Zainal Abidin

Technical University of Malaysia Malacca
45 PUBLICATIONS   126 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

MATHEMATICAL MODELLING OF OPTIMUM WAY POINT PATH PLANNING USING UAV DRONE NAVIGATION IN IMAGE ACQUISITION View project

Network Intrusion Detection System View project

All content following this page was uploaded by Mustafa Hasan Albowarab on 28 June 2019.

The user has requested enhancement of the downloaded file.

 International Journal of Recent Technology and Engineering (IJRTE)
ISSN: 2277-3878, Volume-7, Issue-6S5, April 2019

Load Balancing Algorithms in Software

Defined Network
Mustafa Hasan Al Bowarab, Nurul Azma Zakaria, Z.Zainal Abidin

Abstract— Compared with the traditional networks, the SDN

networks have shown great advantages in many aspects, but also However, SDN controllers have a global view of the
exist the problem of the load imbalance. If the load distribution network and can produce more optimized load balances[6].
uneven in the SDN networks, it will greatly affect the
performance of network. Many SDN-based load balancing
strategies have been proposed to improve the performance of the In this paper we introduce and review the load balance
SDN networks. Therefore, in this paper a finding form algorithms. Load balancing is a significant component of
comprehensive review help to improve further understanding of
current network infrastructure and computer systems where
lead b balancing algorithms in SDN.
Index terms—Software Defined Networking; SDN; Load
resources are distributed over vast ranges of systems and
Balancing. require sharing from a populous end user-base. Load

I. INTRODUCTION
The expand of the global networks which has demand balancing seeks to utilize resources, obtain minimum
new specification of requirements to integrate the response time and reduce overloads as optimally as possible
internetworking systems and networking for current and by distributing the workload. Load balancing is also a basic
future networks[1]. Nowadays, the huge information and big problem in many practical systems in daily life. The
data bang crucial challenge of implementing the networks, supermarket model is a popular example, whereby a central
which leads to find an intelligent [2], efficient and reliable balancer or dispatcher is in charge of assigning queued
network. customers to a particular server in order to reduce the
response time [7]. These Load Balancing techniques are
The current networks have different hardware widely observed in the data-center and enterprise network
equipment such as switches, routers and load balancer which settings in order to bolster scaled-up services. Early works
quiet difficult to deal with them by traditional architecture [8] are based on Round Robin Domain Name System (RR-
networks [3]. To overcome these challenges, the term of DNS) to allocate inbound connections towards a group of
Software Defined Network (SDN) is involved in the general servers. Other well-known load-balancing approaches are
network systems [4]. based on Internet Protocol (IP) level according to flow tuple
[9], or according to the relative load on the different network
The SDN system architecture primarily comprises 3
instances [10].
layers (illustrated in Fig.1). These include the application
layer, control layer and infrastructure layer (may also be Layers 4 and 7 employ further load-balancers. These
termed the data layer). The application and control layers load balancers are employed for numerous other network
interact with one another via API, which is the northward services, including acting as network proxy servers, whereby
interface. The control and infrastructure layers interact via the load balancer is builds on the proxy-server’s cache
the control data surface interface, which is located in the content. The main aim of this utilization is to raise the cache
southward interface[5]. hit ratio instead of achieving an equally spread out server
load balance [11].
The construction of this paper contains three sections,
where the first section explained the evolution of load
balancing, while the second section discuss the types of the
load balancing, finally the review the algorithms of load
balancing has been discussed in the third part.
II. EVOLUTION OF LOADBALANCING
In this section, we discuss the background of load
balancing in the networks that is illustrated in Fig. 2.
Fig.1SDN architecture Domain Name System (DNS) is employed to achieve the
first load balancing technology. In this step, a name is
configured for numerous IP addresses, so as to enable clients
Revised Manuscript Received on December 22, 2018. who query the name to receive one of the addresses. This
Mustafa Hasan ,Center for Advanced Computing Technology, allows various clients to have access to different servers and
Faculty of Information Communication Technology
UniversitiTeknikal Malaysia Melaka UTeM
Melaka, Malaysia

Published By:
Blue Eyes Intelligence Engineering
Retrieval Number:F11200476S519/19©BEIESP 686 & Sciences Publication
 Load Balancing Algorithms in Software Defined Network

thus fulfill load balancing. DNS load balancing is a are utilized by respective network applications to enable
fundamental tool to this aim, but may suffer from not being advanced features for network operators, network providers
able to differentiate between servers and is not able to reflect and users. Network applications and their elements may be
the servers’ current running status [12]. deployed onto SDN controller (SC) directly. Although in
practical settings, network owners restrict deployment due to
Due to the restriction of the DNS load balance approach, security and reliability concerns. Thus, only applications
Hardware load balancer (HLD) was introduced by several which support network management and network service
manufacturers in the mid-1990s [13]. Decoupling load provisions are enabled.
balance function from application enables the DNS to use
network layer techniques such as Network Address Load balancing occupies an important position to solve
Translation (NAT) [14] or Direct Server Return (DSR) [15] over-load traffic problem in the network. It has been one of
to send inbound and outbound traffic to the servers. Such the first appealing applications in SDN networks. These are
techniques are used to process the requests and replies to the some commonly used load balancing approaches [21]:
client.
• Random: This approach randomly distributes the
Atypically, Software Load Balancing (SLB) can be traffic to the available paths. Generally, hash
implemented into Server Operating System (Server OS) such function is used to map requests to available paths.
as Windows Server 2016 [16] or Red Hat’s High
Availability Linux Server [17]. Most of the existing • Round Robin: This approach distributes the request
solutions are focused on distributed network traffic to the paths in sequence, starting from the first path
betweenclustered servers or server farms. SLB is flexible for to the last one in rotation continuously.
cloud visualization environment in which the servers have • Weighted Round Robin: This approach assigns
individual OSs, or share an operating system. SLB in a weight for each path, then distributes requests
cluster environment that allows scaling of the sequentially with respect to the assigned weights.
networkservices where additional servers can be added
dynamically to the cluster. SLB distributes the load between • Least Connections: This approach forward the
servers, while a server cluster provides fault tolerance in the request to the path that has the least number of
system. current connections.
The proliferation of dynamic content led to the delivery IV. TYPES LOAD BALANCING TECHNIQUES
of dynamic services, content-rich applications that need to
Load balancing methods may be classified as either
understand the application-specific traffic. The traditional
static, dynamic or both [1] as shown in Fig. 3. Prior
load balancer could not cope with these growing
knowledge of the system is crucial to static methods, where
requirements. Therefore, HLB has evolved into Application
the rule is directly programmed within the load balancer.
Delivery Controllers (ADCs) [18] over the past ten years.
The caveat is that user behavior cannot be forecasted and is
Typically, in the data center, ADC is a device that sits
thus not optimal for networks. On the other hand, dynamic
methods avoids this problem and is more efficient as load is
spread out in a dynamic manner following a pre-
programmed load balancing pattern [2].
A. Static Load Balancing
In a static algorithm, equal division of traffic is done
within the servers. Static algorithm is suitable for systems
with low load variation. This algorithm needs to have prior
between the firewall and a web farm to provide several tasks information of the system resources in order to be able to
[11] One of these tasks is loading the traffic between web make sure that decision of load shifting does not depend on
current system state.
Fig.2The evolution stages of load balance The master processor delegates the initial tasks to be
executed to individual processors. The same processor is
servers. ADC can inspect packet headers and distribute the always responsible for the tasks delegated. Thus, the work
traffic to a selected server based on this information. load performance is determined from the onset via the
Various load balancing algorithms and methods currently master processor [23]. The slave processors compute the
exist. These include the round robin, fasted node selection, designated work and the results are fed to the master
weighted round robin, IP-based hashing and multi-tier round processor. Tasks are always carried out on the
robin methods. SDN is very flexible, enabling the processorwhich receives the designated task. Static load
installation of company-defined software based on white- balancing methods are not pre-emptive, and is used to
box switch. It further allows the programming of current decrease overall execution time for concurrent programs
equipment to fit the network requirements and decrease while reducing the possible delays in communicating among
costs related to deployment and management [19]. In the processors [24].
context of data centers, the decision of customers to apply a The features of Static Load Balancing methods are:
wide variety of load balancing algorithms depend on the
server size, resource availability, flexibility and peak traffic
hours. [20].
III. LOAD BALANCING ALGORITHMS
The core SDN controller mechanisms lays the
foundation for the autonomous network operations. These
Published By:
Blue Eyes Intelligence Engineering
Retrieval Number:F11200476S519/19©BEIESP 687 & Sciences Publication
 International Journal of Recent Technology and Engineering (IJRTE)
ISSN: 2277-3878, Volume-7, Issue-6S5, April 2019

• It needs less communication in order to minimize • Dynamic methods are suitable for adaptive
the communication delays, where this reduces the applications where the workload is unpredictable, or
execution time. keeps changing during an execution [27].
• It needs less communication in order to minimize • Dynamic methods are also mostly suitable for
the communication delays, where this reduces the heterogeneous and distributed systems.
execution time.
• These methods require that each node must know
• Weighted algorithms achieve a better response time the states of other nodes.
and processing time.
• Even within nodes during execution, processes may
• Load balancing methods load the distribution freely migrate from one another in order to
depending on the load at the time of selecting the guarantee equal loads.
node before the execution starts.
1) Distributed Algorithms: Distributed dynamic
• Static methods are mostly suitable for the constant scheduling methods distribute load balancing over all slave
work application, and for homogeneous and stable nodes rather than only on the master node. Knowledge of
environments that can produce better results in the work load is retrieved depending in the level of demand
present. These methods possess average scalability over a
particular centralized scheme, but suffer from the
disadvantage of being expensive to retrieve and maintain
due to the dynamic nature of system information [28].
There are a number of distributed algorithms as described in
the following subsections.

QoS-Aware Algorithm: SDN systems are unique in that

they are programmable due to control and data plane
decoupling. In particular, they offer basic and user-friendly
programmable network devices instead of complicated
network devices, as those found in active network
Figure 3: The Classification of Load Balancing in SDN protocols. SDN also attempts to separate control and data
planes within the architectural design of the network, with
which the network could be controlled on a separate control
• within these environments. plane and not impact data flows. Therefore, network
intelligence could be removed from switching devices and
• It is easy to be implemented. implemented within controllers. Subsequently, switching
• It is relatively simple to forecast the static methods’ devices could be manipulated by software without built-in
behavior. intelligence. Control plane decoupling within the data plane
seeks to both simplify the programmable environment as
• It is very difficult to forecast the loads’ arrival and well as enable more flexibility in defining the network
processing times prerequisite for future loads [25]. behavior [29],[30].
B. Dynamic Load Balancing Much of the modern network applications, from media
streaming, cloud services and so on, necessitate the use of
Dynamic load balancing techniques are more effective predictable and steady network resources possessing
than the static counterparts due to the dynamic distribution intensive Quality of Service (QoS) criteria. On the other
of pre-programmed load balancer patterns [22], [26]. The hand, OpenFlow is a software-defined network which
proper load balancing is crucial to optimizing minimum enables flow level programmability which supports the
response time, maximal throughput, minimal resource network programming as per QoS criteria as well as network
consumption, scalability and not running into any resource traffic conditions, all in a dynamic manner. Consequently,
overloads. The Dynamic Load Balancing method can be QoS-aware network reprogramming is crucial to traffic
achieved based on three ways: non-distributed, distributed, steering within the multi service SDN-based networks. For
or semi-distribute methods. In the non-distributed method, the particular resource reallocation, network traffic flows
there is one node (centralized) that receives all requests and comprise links as per QoS criteria and traffic engineering
distributes them to the servers. In the distributed method, all requirements.
nodes are shared with the distribution of the requests. As for
the semi-distributed method, the nodes are divided up into a Thus, various problems arise hindering the role of flow
group of clusters, where each cluster works as a central node level resource allocation, such as big flows and network
in order to distribute the requests, and all clusters are resource partitions, burst and dynamic traffics, as well as
responsible for the load balancing distribution[25]. many traffic classes operating under different criteria.
Resource partitioning remains an essential side effect of the
The features of Dynamic Load Balancing methods are as level of dynamicity of network traffic and big flows. If it is
follow: assumed that two 10 Gb/s path from node A and B exist,
• It selects the suitable node that requires real time
communication with the networks, which will lead
to an extra traffic to be added to a system. Dynamic
methods provide better performance.
• It is difficult to be implemented.
Published By:
Blue Eyes Intelligence Engineering
Retrieval Number:F11200476S519/19©BEIESP 688 & Sciences Publication
 Load Balancing Algorithms in Software Defined Network

each with 500 Mb/s free bandwidth, then there is 800 Mb/s methods such as nonlinear optimization, until being
rate from the first to the second node. Because of resource outclassed by the other two aforementioned methods.
partitioning, flow cannot be routed appropriately, thus Cooperative algorithms utilize various decision makers
network bears 1 Gb/s free bandwidth capacity. To combat which agree on decisions cooperatively such that each
these negative effects, flow routing must be executed with a operates optimally individually. On the other hand, non-
holistic view of network while taking into consideration its cooperative algorithms utilize multiple decision makers
effect on all other flows. To elaborate, big flow routing optimizing individual response times regardless of the status
requires the rerouting of a number of other flows because of of others, and is thus not cooperating with others. In these
resource partitioning. Therefore, it is apparent that static cases, Nash equilibrium condition is achieved when no
network configurations are not sufficient. Furthermore, big decision makers are able to obtain further benefits by
flows and dynamic network traffic necessitates the altering its own decisions unanimously. To specify, the
reprogramming of the network on a timely basis, which may network stability in these cases are studied in terms of
lead to instability and undesirable impact on QoS of flows. achieving load distribution so that individual jobs are able to
Within SDN networks, network features comprise the main switch between nodes with lower number of jobs.
problem of network reconfiguration with minimum side-
Load-balancing algorithms may be classified as either
effect and overhead. It is important to note that network
reconfiguration overhead depends on the number of rerouted static or dynamic. The former relies on the currently
flows. Raising this number may lead to network instability possessed knowledge for application load, whereas the latter
as well as heightened packet loss and end-to-end delay [3]. relies on settings where information pertaining to load
distribution is unknown.
a) Heuristic Approaches: are utilized to achieve
Many studies have analyzed load balancing, with a
optimal hybrid routing configurations. Studies have found
notable studies by [43] studying centralized static
that hybrid routing achieves optimal load balancing in cooperative load balancing; [44] and [45] studying
comparison to pure explicit routing. On the other hand, centralized static non-cooperative load balancing; [46] and
latency, overhead and method utilization may not be taken [47] studying centralized dynamic load balancing; and [48]
into account [6] [32] [33]. These studies have compared addressing the issue of distributed dynamic load balancing
among various heuristic approaches in order to heighten the relying upon local cooperation among neighboring network
resilience of software-defined networks in order to oppose nodes.
connection failure among nodes and controllers. Attempts
have been made to maximize controller placement One study analyzed the non-cooperative dynamic load-
balancing method, which is popularly implemented in game
reliability [34]. A minimum number of controllers are
theory frameworks with problems consisting of dynamic
heuristically searched and assigned to individual nodes in
load-balancing game where users distribute loads in non-
conjunction with proper node placement, with the ultimate cooperatively and selfishly [49]. The study also considered
aim of achieving a particular threshold reliability. Many the renowned game theory traffic model proposed by
studies have focused on resilience opposing network Wardrop [50], which sought to describe road traffic with
failures but do not take into account the further metrics such infinitely numbered agents, each in charge of a very small
as πimbalance or πmax latency [34], [35]. More amount of traffic. The framework considers flow demand to
specifically, trade-offs among metrics and objectives, such be routed from a given source to a destination by utilizing
as πmax latency is also not discussed. As opposed to various paths. The agents are able to distribute its own flow
evaluating entire solution spaces, no guarantee is made for within admissible paths. The network bears the trait of non-
optimizing the results obtained in the study [36]. decreasing latency functions which are based on edge flows.
Heuristic-based mechanisms have illustrated novel Multiple flows ensure that latencies of employed paths are as
optimization methods. Studies have attempted to implement low as possible, which is called the Wardrop equilibrium.
such optimization methods for load balancing problems. Therefore, the Nash equilibrium is dubbed a Wardrop
These methods range from honey bee swarm algorithm [37], equilibrium when there exists an infinite number of decision
lion optimization algorithm [38], whale optimization makers [51].
algorithm [39], gray wolf optimization algorithm [40], bat 2) Centralized Algorithms: In this approach, the concern
optimization algorithm [41].
of the load-balancing decision remains with the master node
b) Wardrop Load Balancing: The Wardop Load- and the data used for the load balancing is gathered from the
Balancing algorithm is employed to converge arbitrarily remaining (slave’s) nodes on either on demand basis or after
small neighborhoods of particular equilibrium for loads a certain predefined time interval. The obvious point is since
within providers. Due to its features, the algorithm is the data is not sent arbitrarily; the unnecessary traffic over
feasible for various SDN scenarios, where service requests the network is minimized. But, the scalability remains
originate from network nodes and controlled by SDN limited with this technique [28] [29]. There are a number of
controllers. centralized algorithms as described in the following
Various load-balancing methods have been studied subsections.
previously by researchers [42]. It has been recommended to
a) Routing Control Platform (RCP): Roughly a
categorize load-balancing algorithms as either global-based,
decade earlier, numerous studies have focused on the
cooperative-based or non-cooperative based approaches.
Global algorithms entail that individual nodes transmit Routing Control Platforms (RCP) [52-56]. These studies
current status to centralized load balancers via an extensive were the first to refactor IP routing architecture and made a
and cohesive system network. Thus, jobs are delegated to
each resource which also optimizes a particular objective,
such as the response time of the whole system over all jobs.
This method has been popularly used in conjunction with
Published By:
Blue Eyes Intelligence Engineering
Retrieval Number:F11200476S519/19©BEIESP 689 & Sciences Publication
 International Journal of Recent Technology and Engineering (IJRTE)
ISSN: 2277-3878, Volume-7, Issue-6S5, April 2019

logically centralized control plane distinct from forwarding

elements and more towards the BGP decision processes and
route control requirements given by a large operator.
At its core, RCPs comprise three architectural principles:
path computations based on a consistent network state,
expressively specified routing policies and controlled
communication between routing protocol layers [52].
Compared to BGP configurations distributed complexly over
various routers, these methods allow individual AS to
rapidly utilize novel, customer-facing services [54].
In the RCP architecture, legacy network routing
comprises iBGP Route Reflector (RR), which obtains
network information and switches it with RCP. Network
information is retrieved consistently by retrieving eBGP
learned routes. Therefore, the retrieved information is
directly distributed over border routers possessing full route
access. Consequently, the optimal routes are calculated
depending on the route learned from the eBGP. The entire
routing configuration and states are contained within the
RCP control plane. For proper handling, RCP maintains
local registries with global views and information exchanges
with external RCP for routing between domains. Utilizing
the RCP appropriately within the SDN is referred to as
‘Intelligent Route Service’. Control Point (IRSCP) is an
architecture mainly employed prior to the advent of SDN
[57].
The current RouteFlow Control Platform (RFCP) Fig. 4 Route Flow Control Platform (Architecture design) [57].
architecture has advanced from prior prototypes [58] to more
sophisticatedly-layered, distributed system design and
possessing ample flexibility to be employed within diverse b) Server-Based Load Balancing Algorithm
virtualization cases. This method is based on a modular (SBLA):The SBLA load balancing algorithm was proposed
architecture comprising three major components, as show in in [60] and it was suitable for server - cluster in virtual
Fig. 4. These include the RF-Slave, RF-Server and RF- environment Firstly, the controller used the SNMP protocol
Controller. The RF-Slave collects routing and forwards to collect the state information of the servers, then calculated
information from the Linux host. Alternatively, in order to the load of the serve is according to the SBLA algorithm,
extract complementary routing information, it utilizes peers and finally selected the lightest load server to respond to the
such as iBGPs to hook using a routing engine. Next, the RF- users. The algorithm minimized the server’s response time,
Server is a standalone application in charge of the core but was unsuitable for unstructured networks and data
system logic, including event processing, VM-to-switch centers.
mapping, and resource management. RFCP Services are also
utilized as operator-tailored modules utilizing a c) DUTE Algorithm: Gandhi et al. proposed a DUTE
knowledgeable information base to carry out arbitrary high- scheme which combined the hardware with the software
level routing logic. Lastly, the RF-Controller is an [61]. The scheme used the existing switches to build a
application above the OpenFlow Controller serving RFCP hardware load balancer to effectively increase the capacity,
via switch interaction and the topology sate collection of the reduce the cost and delay, but the flexibility was poor.
network [59]. Especially, when the switch failed, the disadvantage was
especially obvious. The hardware load balancer dealt with a
large number of traffic, while the software load balancer
served as a backup to ensure high availability and Gandhi et
al proposed a DUTE scheme which combined flexibility, but
it was difficult to implement.
C. Hybrid Load Balancing
These methods are achieved and employed to get rid of
the disadvantages associated with Dynamic and Static Load
Balancing methods, and they are being used to aggregate the
benefits and merits of static and dynamic algorithms in order
to design a new one [22]. In fact, this implies that
combinations the benefits of two or more existed algorithms
either dynamic or static algorithms are able to present a new
one.

Published By:
Blue Eyes Intelligence Engineering
Retrieval Number:F11200476S519/19©BEIESP 690 & Sciences Publication
 Load Balancing Algorithms in Software Defined Network

The features of the Hybrid Load Balancing methods are: 5. X. You and Y. Wu, “Software Defined Network Architecture based
Research on Load Balancing Strategy,” in AIP Conference
• The word “data” is plural, not singular. Proceedings, 2018, vol. 1967.
6. A. A. Neghabi, N. J. Navimipour, M. Hosseinzadeh, and A. Rezaee,
• The main disadvantage lies in its incapacity to “Load Balancing Mechanisms in the Software Defined Networks: A
enable noncomplex methods. Systematic and Comprehensive Review of the Literature,” IEEE
Access, vol. 6, pp. 14159–14178, 2018.
• Hybrid methods take over the attributes of both 7. S. Islam, “Network Load Balancing Methods : Experimental
static and dynamic LB techniques, seeking to Comparisons and Improvement,” arXiv Prepr. arXiv1710.06957.,
overcome the drawbacks of both methods. They are 2017.
more scalable. 8. E. D. Katz, M. Butler, and R. McGrath, “A scalable HTTP server:
The NCSA prototype,” Comput. Networks ISDN Syst., vol. 27, no. 2,
V. EVALUATION OF LOAD BALANCING pp. 155–164, 1994.
ALGORITHMS 9. A. Bestavros, M. Crovella, J. Liu, and D. Martin, “Distributed Packet
Rewriting and its Application to Scalable Server Architectures,” in
This paper classifies the load balancing researches in Proceedings Sixth International Conference on Network Protocols
SDN networks, as show in Fig. 3. The SDN architectures (Cat. No.98TB100256), 1998, pp. 290–297.
can be divided into centralized single controller architectures 10. L. Aversa and A. Bestavros, “Load Balancing a Cluster of Web
and distributed multiple controller’s architectures according Servers Using Distributed Packet Rewriting Luis,” in Conference
Proceedings of the 2000 IEEE International Performance,
to the number and organization of the controllers in SDN Computing, and Communications Conference (Cat. No.00CH37086),
networks, in the centralized architectures load balancing 2000, pp. 24–29.
researches are divided into the data plane and the control 11. E. Doron and M. Sekiguchi, “Techniques for providing scalable
plane. The data plane mainly includes link load balancing application delivery controller services,” US9386085B2, Apr-2016.
and server load balancing. The distributed architectures are 12. W. Chen, H. Li, Q. Ma, and Z. Shang, “Design and implementation
divided into the flat architecture and the hierarchical of server cluster dynamic load balancing in virtualization
architecture. This paper introduces, analyzes and environment based on OpenFlow,” in International Joint Conference
summarizes types of load balancing researches, so that on Awareness Science and Technology & Ubi-Media Computing
(iCAST 2013 & UMEDIA 2013), 2014, pp. 691–697.
researchers can quickly understand the relevant knowledge
13. J. Ju, G. Xu, and K. Yang, “An Intelligent Dynamic Load Balancer
in this field. for Workstation Clusters,” ACM SIGOPS Oper. Syst. Rev., vol. 29,
no. 1, pp. 7–16, 1995.
VI. CONCLUSION 14. P. Srisuresh and K. Egevang, “Traditional IP Network Address
Software Defined Network has been developed to Translator (Traditional NAT),” RFC 1631, pp. 1–16, 2000.
manage large networks like data center big data. Due to 15. C. Kopparapu, Load Balancing Servers, Firewalls, and Caches. 2002.
huge expanding of internet, enormous number of request is 16. J. H. Huh and K. Seo, “Design and test bed experiments of server
arriving at server per second. In this paper a detailed survey operation system using virtualization technology,” Human-centric
of load balancing algorithms has been done. The Comput. Inf. Sci., pp. 1–21, 2016.
classifications of load balancing have been divided to three 17. S. Cash, A. Karve, T. Mathews, S. Mullen, and M. Mulsow,
“Managed infrastructure with IBM Cloud OpenStack Services,” vol.
types of static, dynamic and hybrid load balancing. There is 60, no. 2, pp. 1–12, 2016.
requirement of efficient algorithm to balance the load of 18. K. Salchow Jr, “Load Balancing 101 : The Evolution to Application
server to avoid network degradation. The centralized Delivery Controllers,” 2007.
controller of SDN has the global view of network which 19. S. Khan, M. Ali, N. Sher, Y. Asim, W. Naeem, and M. Kamran,
makes load balancing in SDN easy. The load balancing “Software-Defined Networks (SDNs) and Internet of Things (IoTs):
algorithm must consider the current load to reflect the real A Qualitative Prediction for 2020,” Int. J. Adv. Comput. Sci. Appl.,
time change. Using single centralized controller can lead to vol. 7, no. 11, pp. 385–404, 2016.
single point of failure. So load balancing algorithm should 20. D. Mithbavkar, H. Joshi, H. Kotak, D. Gajjar, and L. Perigo, “Round
Robin Load Balancer using Software Defined Networking ( SDN ),”
be mainly based on distributed decision. Researchers Capstone Team Research Project. pp. 1–9, 2016.
should do more detailed study of distributed architecture to
21. M. Alanyali and B. Hajek, “Analysis of simple algorithms for
develop better load balancing algorithms taking advantage dynamic load balancing,” 1997.
of SDN architecture. The algorithm should be designed in 22. A. S. Milani and N. Jafari, “Load balancing mechanisms and
such a way that it minimizes the latency and response time techniques in the cloud environments : Systematic literature review
and maximize the throughput. and future trends,” J. Netw. Comput. Appl., vol. 71, pp. 86–98, 2016.
23. H. Kameda, E. S. Fathyt, I. Ryut, and J. Lis, “A Performance
ACKNOWLEDGMENT Comparison of Dynamic vs. Static Load Balancing Policies in a
Mainframe - Personal Computer Network,” in Proceedings of the
The researchers would like to thank the university technical 39th IEEE Conference on Decision and Control (Cat.
Malaysia Melaka (UTeM) for sponsoring this work. No.00CH37187), 2000, pp. 1415–1420.
24. I. Journal and F. Technological, “Comparison of Static and Dynamic
Load Balancing in Grid Computing,” Int. J. Technol. Res. Eng., vol.
REFERENCES 2, no. 7, pp. 1337–1340, 2015.
1. P. Martinez-julia and A. F. Skarmeta, “Empowering the Internet of 25. S. Hamadah, “A Survey : A Comprehensive Study of Static ,
Things with Software Defined Networking,” White Pap. IoT6-FP7 Dynamic and Hybrid Load Balancing Algorithms,” Int. J. Comput.
Eur. Res. Proj., 2014. Sci. Inf. Technol. Secur. (IJCSITS), vol. 7, no. 2, pp. 27–32, 2017.
2. M. Nasser et al., “Cyber-Security Incidents: A Review Cases in 26. S. Chen, Y. Chen, and S. Kuo, “CLB: A novel load balancing
Cyber-Physical Systems,” Int. J. Adv. Comput. Sci. Appl., vol. 9, no. architecture and algorithm for cloud services,” Comput. Electr. Eng.,
1, 2018. vol. 58, pp. 154–160, 2017.
3. P. Kumari and D. Thakur, “Load Balancing in Software Defined
Network,” Int. J. Comput. Sci. Eng., vol. 5, no. 12, pp. 227–232,
2017.
4. J. Li, E. Altman, and C. Touati, “A General SDN-based IoT
Framework with NVF Implementation,” ZTE Commun., vol. 13, no.
3, pp. 42–45, 2015.
Published By:
Blue Eyes Intelligence Engineering
Retrieval Number:F11200476S519/19©BEIESP 691 & Sciences Publication
 International Journal of Recent Technology and Engineering (IJRTE)
ISSN: 2277-3878, Volume-7, Issue-6S5, April 2019

27. I. Publication, “A comparative study of static and dynamic 52. R. Ramjee et al., “Separating Control Software from Routers,”
Load Balancing Algorithms,” IJARCSMS, vol. 2, no. 12, pp. 2006 1st Int. Conf. Commun. Syst. Softw. Middlew., pp. 1–10,
386–392, 2014. 2006.
28. B. Kang and H. Choo, “An SDN-enhanced load-balancing 53. N. Feamster, H. Balakrishnan, J. Rexford, A. Shaikh, and J. van
technique in the cloud system,” J. Supercomput., pp. 1–24, der Merwe, “The case for separating routing from routers,”
2016. Proc. ACM SIGCOMM Work. Futur. Dir. Netw. Archit. -
29. A. A. Hassan, W. Shah, M. F. Iskandar, M. S. Talib, and A. FDNA ’04, pp. 5–12, 2004.
Abdul-jabbar, “K Nearest Neighbor Joins and MapReduce 54. J. Van der Merwe et al., “Dynamic connectivity management
process enforcement for the cluster of data sets in BigData '' with an intelligent route service control point,” Proc. 2006
Jour of Adv Research in Dynamical & Control Systems,vol. 10, SIGCOMM Work. Internet Netw. Manag. - INM ’06, pp. 29–
pp. 690-695, 2018 34, 2006.
30. W. Xia, Y. Wen, C. H. Foh, D. Niyato, and H. Xie, “Survey on 55. Y. Wang, I. Avramopoulos, and J. Rexford, “Design for
software-defined networking,” IEEE Commun. Surv. Tutorials, configurability: Rethinking interdomain routing policies from
vol. 17, no. 1, pp. 27–51, 2015. the ground up,” IEEE J. Sel. Areas Commun., vol. 27, no. 3, pp.
31. M. M. Tajiki, B. Akbari, and N. Mokari, “Optimal Qos-aware 336–348, 2009.
network reconfiguration in software defined cloud data 56. T. D. Mohammed AA, Burhanuddin MA, Basiron H, “Key
centers,” Comput. Networks, vol. 120, pp. 71–86, 2017. enablers of IoT strategies in the context of smart city
32. Y. N. Hu, W. D. Wang, X. Y. Gong, X. R. Que, and S. D. innovation.,” Adv. Res. Dyn. Control Syst, vol. 10, no. 4, 2018.
Cheng, “On the placement of controllers in software-defined 57. P. Verkaik, D. Pei, T. Scholl, A. Shaikh, A. C. Snoeren, and J.
networks,” J. China Univ. Posts Telecommun., vol. 19, pp. 92– E. van der Merwe, “Wresting control from BGP: scalable fine-
97, 2012. grained route control,” 2007 USENIX Annu. Tech. Conf. Proc.
33. H. Yannan, W. Wendong, G. Xiangyang, Q. Xirong, and C. USENIX Annu. Tech. Conf., pp. 295–308, 2007.
Shiduan, “Reliability-aware Controller Placement for Software- 58. M. R. Nascimento, C. E. Rothenberg, M. R. Salvador, C. N. A.
Defined Networks,” EEE Int. Symp. Integr. Netw. Manag., no. Corrêa, S. C. de Lucena, and M. F. Magalhães, “Virtual Routers
February, pp. 672–675, 2013. as a Service: The RouteFlow Approach Leveraging Software-
34. F. J. Ros and P. M. Ruiz, “Five nines of southbound reliability Defined Networks,” Proc. 6th Int. Conf. Futur. Internet
in software-defined networks,” Proc. third Work. Hot Top. Technol. - CFI ’11, vol. 1, pp. 34–37, 2011.
Softw. Defin. Netw. - HotSDN ’14, no. May, pp. 31–36, 2014. 59. C. Rothenberg et al., “Revisiting IP Routing Control Platforms
35. Y. Zhang, N. Beheshti, and M. Tatipamula, “On resilience of with OpenFlow-based Software-Defined Networks,” Futur.
split-architecture networks,” in IEEE Global Internet Exp. Res. Work., 2012.
Telecommunications Conference - GLOBECOM 2011, 2011. 60. W. Chen, Z. Shang, X. Tian, and H. Li, “Dynamic Server
36. S. Lange et al., “Heuristic Approaches to the Controller Cluster Load Balancing in Virtualization Environment with
Placement Problem in Large Scale SDN Networks,” IEEE OpenFlow,” Int. J. Distrib. Sens. Networks, vol. 11, no. 7, pp.
Trans. Netw. Serv. Manag., vol. 12, no. 1, pp. 4–17, 2015. 1–9, 2015.
37. D. Karaboga, “An idea based on honey bee swarm for 61. R. Gandhi et al., “Duet : cloud scale load balancing with
numerical optimization,” 2005. hardware and software,” ACM SIGCOMM Comput. Commun.
Rev., vol. 44, no. 4, pp. 27–38, 2014.
38. M. Yazdani and F. Jolai, “Lion Optimization Algorithm (LOA):
A nature-inspired metaheuristic algorithm,” J. Comput. Des.
Eng., vol. 3, no. 1, pp. 24–36, 2016.
39. S. Mirjalili and A. Lewis, “The Whale Optimization
Algorithm,” Adv. Eng. Softw., vol. 95, pp. 51–67, 2016.
40. S. Mirjalili, S. M. Mirjalili, and A. Lewis, “Grey Wolf
Optimizer,” Adv. Eng. Softw., vol. 69, pp. 46–61, 2014.
41. X. Yang and A. H. Gandomi, “Bat Algorithm : A Novel
Approach for Global Engineering Optimization,” Eng.
Comput., vol. 29, no. 5, pp. 464--483, 2012.
42. D. Grosu, A. . Chronopoulos, and M. . Leung, “Cooperative
load balancing in distributed systems,” Concurr. Comput. Pract.
Exp., vol. 20, no. 16, pp. 1953–1976, 2008.
43. S. U. Khan and I. Ahmad, “A Cooperative Game Theoretical
Technique for Joint Optimization of Energy Consumption and
Response Time in Computational Grids,” IEEE Trans. Parallel
Distrib. Syst., vol. 20, no. 3, pp. 346–360, 2009.
44. D. Grosu and A. T. Chronopoulos, “Noncooperative load
balancing in distributed systems,” J. Parallel Distrib. Comput.,
vol. 65, no. 9, pp. 1022–1034, 2005.
45. R. Subrata, A. Y. Zomaya, and B. Landfeldt, “Game-theoretic
approach for load balancing in computational grids,” IEEE
Trans. Parallel Distrib. Syst., vol. 19, no. 1, pp. 66–76, 2008.
46. E. Even-Dar, A. Kesselman, and Y. Mansour, “Convergence
Time to Nash Equilibria,” Int. Colloq. Autom. Lang. Program.,
vol. 2719, pp. 502–513,2003.
47. E. Even-Dar and Y. Mansour, “Fast Convergence of Selfish
Rerouting,” Proc. Sixt. Annu. ACM-SIAM Symp. Discret.
Algorithms, pp. 772–781, 2005.
48. S. Shah and R. Kothari, “Convergence of the dynamic load
balancing problem to Nash equilibrium using distributed local
interactions,” Inf. Sci. (Ny)., vol. 221, pp. 297–305, 2013.
49. H. Ackermann, S. Fischer, M. Hoefer, and M. Schöngens,
“Distributed algorithms for QoS load balancing,” Distrib.
Comput., vol. 23, no. 5–6, pp. 321–330, 2011.
50. J. Wardrop, “Some theoretical aspects of road traffic research,”
in trid.trb.org, 1952, vol. 1, no. 3, pp. 325–362.
51. H. Kameda, J. Li, C. Kim, and Y. Zhang, Optimal load
balancing in distributed computer systems. 1997.

Published By:
Blue Eyes Intelligence Engineering
Retrieval Number:F11200476S519/19©BEIESP 692 & Sciences Publication
 Load Balancing Algorithms in Software Defined Network

Published By:
Blue Eyes Intelligence Engineering
Retrieval Number:F11200476S519/19©BEIESP 693 & Sciences Publication

View publication stats