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

net/publication/321483922

Global Server Load Balancing with Networked Load Balancers for

Geographically Distributed Cloud Data-Centres

Article · December 2017

CITATION READS

1 603

2 authors:

Amit Gajbhiye Shailendra Singh

Durham University National Institute of Technical Teachers' Training and Research, Bhopal
17 PUBLICATIONS 64 CITATIONS 101 PUBLICATIONS 1,078 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Amit Gajbhiye on 03 December 2017.

The user has requested enhancement of the downloaded file.

IJCSN - International Journal of Computer Science and Network, Volume 6, Issue 6, December 2017
ISSN (Online) : 2277-5420
www.IJCSN.org
Impact Factor: 1.5

Global Server Load Balancing with Networked

Load Balancers for Geographically Distributed
Cloud Data-Centres
1
Amit Gajbhiye; 2 Dr. Shailendra Singh
1
Department of Computer Engineering and Applications
NITTTR, Bhopal, India
2
Department of Computer Engineering and Applications
NITTTR, Bhopal, India

Abstract - The characteristic of high scalability, elasticity and availability of infinite computing resources have obliged many
organizations to deploy mission critical and user centric applications in the cloud computing environment. To avoid any single point
of failure in the cloud system the cloud service provider provide redundancy at all levels in the datacenter but the recent natural
disasters have made the datacenter itself a single point of failure. Thus, to deploy applications in the datacenters dispersed across
geographic regions has become the need of the hour. Efficient load balancing among these datacenters is critical to increase user
responsiveness and seamless failover in case of natural disasters. In this research paper we critically analysed the state-of-the art
techniques used for Global Server Load Balancing and have proposed a novel model of networked load balancers for load balancing
across the datacenters in cloud computing environment. The proposed model overcomes the shortcomings of existing DNS based
load balancing by considering current load status of datacenter, by making time to live of DNS server cache redundant and by
finding the exact location by real IP of end user.

Keywords - Global Server Load Balancing, Domain Name System, Load Balancing, Disaster Recovery, Cloud Computing

1. Introduction  Improves the responsiveness of services to users.

 Tracking and controlling traffic.
loud computing (or cloud for short) is a disruptive  Prevents over provisioning of resources.


C technology. It is a model for delivering the Avoid overloading and under loading of
hardware and software services on demand in a resources.
scalable manner. The compelling need which has
given birth to this technology is the demand of Load balancing can be done at the datacenter (in the
instant infrastructure provisioning with no or less upfront datacenter) level or at the global level (across the
cost and rapid scalability of the applications. The main datacenter). As mentioned at the datacenter level the load
thrust for this technology has been virtualization, parallel balancers are concerned with distributing dynamic load
programming model with distributed file system and among the various severs evenly. On the other hand, the
multiple speed gigabit networks. The datacenter hardware global server load balancing (GSLB) is concerned with
including various servers like storage, application and distributing the load geographically across multiple
web and the software services deployed on them is called datacenters. In the literature of cloud computing specially
cloud [1]. The efficient utilization of datacenter resources for load balancing much emphasis is given on the load
and customer satisfaction for the responsiveness of their balancing strategies within the datacenter. However, the
application or service is a crucial factor for the success of load balancing among the datacenters dispersed across the
cloud computing. This fact strive us to think how to geography is also crucial for two important reasons.
evenly distribute service requests(load) from the end users
to the servers in order to maximize user responsiveness First, looking at the present demanding business needs for
and minimise exploitation of a single server being over example in e-commerce and banking where critical
loaded with service request other being under loaded. In business processes are deployed and run in the cloud
cloud computing parlance this is called load balancing. computing environment, the organizations cannot afford
Load balancing [2] in cloud computing is essential for the the service outages even for short period of time. This
following reasons as it: scenario has increased the significance of factors like
continuous availability and fast responsiveness of services
 Increases resource utilization. many folds. The organizations are in a great need of
 Improves resource availability. efficiently utilizing their resources today than ever before.

682
Copyright (c) 2017 International Journal of Computer Science and Network. All Rights Reserved.
IJCSN - International Journal of Computer Science and Network, Volume 6, Issue 6, December 2017
ISSN (Online) : 2277-5420
www.IJCSN.org
Impact Factor: 1.5

Second the recent spate of natural disaster, power failures technical and business goals of GSLB. The author also
or DNS DDOS attack and many other unforeseen discusses the crucial challenges and their solution in this
circumstances datacenters itself has become a single point paper.
of failure. Global server load balancing is used when an
organization wants multi site load balancing and disaster The [5] have proposed a server load balancing protocol
recovery. They work at the datacenter level to provide for efficient server selection for the client. In this protocol
resiliency to cloud hosted application. the server registers itself and coordinates its services with
a Server Load Balancer (SLB) manager. For registration
In this research paper we review the state-of-the-art to SLB manager the server uses the Transmission Control
Protocol. The important jobs of SLB manager in this
techniques for global server load balancing and propose a
protocol is to register virtual IP address of server with a
novel framework for global sever load balancing by
DNS based global server load balancer, support VIP
networking load balancers across geographically
address route injection and provide health status of
distributes load balancers. Our theoretical analysis services to applications which helps in selection of
suggests that the proposed model not only overcome the optimal site. A DNS based HTTP redirection and route
shortcomings of existing DNS based load balancing by health injection method for global server load balancing is
considering current load status of the datacenter, by outlined in [6].
making time to live of DNS server cache redundant and
by finding the exact location by real IP of end user but The authors of [7], proposed a model in which the servers
also provide seamless failover and maximum uptime in handles the task of balancing the loads by themselves.
case of natural disasters or a complete datacenter failure. When the server receives a request it either processes the
request or passes the request to its peer. The decision to
The primary contributions of this paper are: process the request depends on the current load of the
system and/or on the request contents. As there in no
 A distributed and dynamic load balancing model for dedicated hardware resources for load balancing the
global server load balancing with networking among disadvantages associated with them such as single point
load balancers in datacenters distributed across of failure are not found in this model of global server load
geography. balancing.
 A novel load calculation strategy to calculate the load
of entire datacenters based on the summation of load The peers communicate with each other by passing
of virtual machines on individual server in messages either by following request/response semantics
datacenter. or by circulating unsolicited multicast messages about
 A request migration strategy from one datacenter to their current load status. The messages may be passed
another to reduce the response time for requests. periodically or in response to some events. In [8] the
authors have proposed a global server load balancing
switch which acts as a proxy to authoritative name
2. Related Work servers. The GSLB switch gathers load information of the
servers from switches attached to servers directly and
returns the IP address of the least loaded server to the
In this section we will review the existing research hosts.
literature in the field of global server load balancing. Most
of the academic research work in the field of load In [9] the authors have presented a system and a method
balancing in cloud computing is concentrated on for load balancing multiple geographically separated
balancing the loads on the servers of the same datacenters. servers. The system consists of load balancing domain
But there is some pioneering research in industry for name servers (DNS-LBS) which are located close to the
global server load balancing. Prominent among the servers of Internet Service Provider and hence are better
organization involved in GSLB research are Akamai, F5 able to judge the performance of servers. The DNS-LBS
Networks and Citrix. returns the IP address of the server that provides better
performance according to the physical location of the
The authors in [3] have proposed a site selection method server.
for selecting the best server among the set of application
servers located across geographically distributed The paper discusses two GSLB methods called Round
datacenters. The proposed DR3 method combines the Trip Time (RTT) and Static Proximity. RTT method is
DNS reply race with DNS reflection for selecting a site dynamic probing method in which the round trip time is
with faster response time and lower cost. The DR3 used to find the IP address of closest site and the IP
method selects site with higher accuracy and lower cost as address is returned to the user. In contrast to RTT method
compared to DNS reply race and ping method of site static proximity method uses pre-configured IP address
selection. The research work in[4] discusses the various tables to find the closest site.

683
Copyright (c) 2017 International Journal of Computer Science and Network. All Rights Reserved.
IJCSN - International Journal of Computer Science and Network, Volume 6, Issue 6, December 2017
ISSN (Online) : 2277-5420
www.IJCSN.org
Impact Factor: 1.5

3.Analysis of State-of-the-art GSLB authoritative servers. The TTL signifies the validity
Techniques of cache entry. For a period specified by TTL the
DNS server will send the cached address of services
delivering server without querying the authoritative
In this section, we critically review the state-of-the-art servers for the updated address if in case the old IP
techniques in the field of global server load balancing and address is changed. For typical DNS implementation
discusses their suitability in the field of cloud computing. the value of TTL is 86400 seconds, which is 24
hours. In case of disaster or complete failure of
3.1 DNS based Global Server Load Balancing datacenter the DNS will keep sending the client
request for a period of 24 hrs to the defunct
The most common method of global load balancing datacenter which is highly undesirable in dynamic
followed nowadays is with the help of Domain Name cloud environment.
System (DNS) [10]. DNS works in client/server  The DNS distributes the load based on geographic
architecture to map host names in the application layer to location of the client. DNS follow either recursive or
an IP address in the network layer of TCP/IP protocol iterative resolution mechanism in case the resolution
suit. Whenever a client requests for a service over the cannot be done at local DNS. Due to this resolution
internet the request is first sent to the DNS for translating mechanisms the DNS may not know the correct
domain names to IP addresses. The DNS client sends a geographic location of the end client and may direct
GetHostByName query to request the nearest DNS server the end clients to avail a service from distant server,
to map a name to an IP address. The local DNS server which is not only time consuming but also incurs cost
searches its cache to check if it already contains an entry in cloud environment.
for the host name if it finds an entry it replies with the
corresponding IP address and if it does not have an entry
it refers to the other DNS server to provide the 3.2 A Variation of the DNS based Global Server
information. DNS can be configured to respond with Load Balancing
different IPs based on the location of the client or in a
“round-robin” manner. Thus the DNS based Global To overcome the disadvantages of not considering current
Server Load Balancing effectively divides load among the load of servers a new model is employed by many
servers across the geography by responding with a organizations like Yahoo by deploying dedicated global
different IP address of the server for the client request. server load balancer. Fig. 2 shows the steps followed by
Fig 1. shows the DNS based GSLB mechanism. this model to resolve a Fully Qualified Domain Name.
Datacenter (Site) A for example, has a virtual IP address
The disadvantages of DNS based GSLB Model method is (VIP) of 11.11.11.11 and Datacenter (Site) B has a VIP of
as follows [11]: 22.22.22.22. In this model a Global Server Load Balancer
acts as an authoritative name server which responds to the
 It does not consider the current load on the server but client request for appropriate server for establishing
simply directs the requests to the servers in a round connection for service or communication. Figure 2,
robin fashion which may create load imbalance in the explains the whole communication steps for an example
system. site www.ExampleModel2.com.

Fig. 1 GSLB Model with DNS

 To increase the efficiency by reducing the time spent

in searching the DNS database, DNS servers uses the
cache to store the most frequent name-address
resolutions. The entries are cached in cache for a time Fig. 2 GSLB model with dedicated servers
(called TTL short for time to live) specified by

684
Copyright (c) 2017 International Journal of Computer Science and Network. All Rights Reserved.
IJCSN - International Journal of Computer Science and Network, Volume 6, Issue 6, December 2017
ISSN (Online) : 2277-5420
www.IJCSN.org
Impact Factor: 1.5

a) The DNS client (resolver) requests the local cost effective delivery to end users and provide seamless
DNS server, which in our example is the name failover in case of complete datacenter failure. We
server of Internet Service Provider (ISP). balance the load at two levels first on the basis of current
load status and second on the basis of real IP address of
b) The client’s DNS server queries root name end user. In essence we are distributing load among
servers iteratively and finally ends up at the central load balancers by decentralising the authority load
authoritative name server for balancing.
www.ExampleModel2.com. In this model GSLB
is the authoritative name server. 4. Proposed Approach

c) The GSLB is connected to the load balancer of 4.1 Networked Load Balancers System:
each of the datacenter and it has a record for the Architecture, Design and Working
current load of the datacenter i.e. the health of
datacenter. In this section, we present the system model of the
networked load balancers for distributed load balancing.
d) Based on the load at each datacenter the GSLB This model focuses on organizations that have many
now makes a decision as to which datacenter to datacenter distributed across the globe with their
connect to and returns the answer to client’s respective centralised load balancer. We propose the idea
DNS server. The value is stored at the client’s of connecting the load balancers of these organizations
DNS server for the TTL time period as specified datacenter so that they can communicate and coordinate
by GSLB. with each other. The load balancers in the datacenters can
communicate with each other to share their current load
e) The choice of datacenter is returned to the client status and coordinate accordingly distribute service
resolver, which now makes a TCP connection to request (load) among them. Fig. 3 shows the proposed
the preferred datacenter. model.

This model works exactly like normal DNS but it applies In the proposed model the centralised load balancers of
additional intelligence to decide the preferred datacenter the datacenters maintain a load table depicted in Fig. 4(a).
to connect to as opposed to normal DNS. However this The load table contains the current load information
model also has the following shortcomings. against the IP address of datacenter and their health status
which essentially is the further load accepting capacity of
 First, although this model have the advantage of the datacenter. The three value of health status is under
considering load at the time of deciding preferred loaded, normally loaded and over loaded. Each load
datacenter it still is infected by the delays caused by balancer is connected to every other load balancer like a
high TTL and even if the authoritative server mesh topology with the shortest link, calculated with
(GSLB) reduces the TTL the resolving name servers Dijkstra’s algorithm. These shortest links are calculated
(ISP DNS server) may override the TTL value. once as the location of datacenter are fixed and do not
Therefore, even if TTL specified by GSLB is 60 change frequently. The connected load balancers share
seconds the resolving name servers may cache the their load status with each other following these shortest
response for 3600 second to avoid overloading its links. Each load balancer sends its load status to every
resolving name server [12]. other load balancer it is connected to via the shortest path
 Second, most organization [13] [14] claims to available.
provide load balancing based on end client location
but they considers the IP address of resolving name
server for calculating the proximity of request serving
datacenter and end client, which may not be true for
all the cases because the end client may use any DNS
Server for resolving names to IP address and even if
the client uses the DNS server of ISP, the ISP may
host its DNS services far from end client.

Motivated by the need to provide an effective and

efficient global server load balancing and to fill the gap of
current practices, we propose novel global server load
balancing model for centralised load balancer in the
datacenter. In our dynamic load balancing model load
balancers across the geography communicate and
coordinate with each other to provide high responsive and Fig. 3 Proposed GSLB Model

685
Copyright (c) 2017 International Journal of Computer Science and Network. All Rights Reserved.
IJCSN - International Journal of Computer Science and Network, Volume 6, Issue 6, December 2017
ISSN (Online) : 2277-5420
www.IJCSN.org
Impact Factor: 1.5

The load balancer consolidates the load information it has  Communication with IPM: LAM also communicates
received from other load balancers and prepares a with IPM to exchange both types load tables.
complete load table shown in Fig. 4(b) sorted on
ascending load value. Thus each load balancer has the
current health status of all the load balancers in the form
of complete load table.

The complete load table is now used to direct the requests

across the datacenter according to the geographic location
of end client. As all the datacenter now have a list of least
to most loaded datacenters with them i.e. a list with
ascending load values of datacenter with their IPs. The
updated values of IPs of these least loaded datacenters are
then updated to the authoritative name server for the
domain of the service and to the DNS servers of the ISP
of the end client so that they always resolve to the IP
address of least loaded datacenter.
Fig. 5 Modules of proposed model

Datacenter IP Load value Health Status

B. IP updater Module (IPM)
1.1.1.1 25 Under loaded
Every load balancer in the system has an IPM module.
This module takes care of inter and intra communication
(a)
of load balancers. It communicates with the end client to
accept or migrate the request to other datacenter and with
Datacenter IP Load value Health Status
1.1.1.1 25 Under Loaded the load balance module if the decision is taken to accept
2.2.2.2 48 Normal the request. Apart from this the following are its other
. . . important responsibilities:
. . .
. . .
 Accepting Load Tables: The module communicates
3.3.3.3 96 Over Loaded with LAM for the transaction of load tables. It
interacts with IPM of other load balancers to
(b) exchange the load tables.
 Sending Periodic IP Updates: This is the most crucial
Fig. 4 Data Structures of the proposed model.
(a) Load Table (b) Complete Load Table task of the whole system. It sends the authoritative
name severs and ISP DNSs the least loaded server
IPs. It may send multiple IPs to DNS or whole
4.2 Working of Networked Load Balancers System complete load table.
 Seamless transfer of request: When a service request
The working of the proposed model is implemented with comes to a datacenter the IPM examines the real IP
the help of two sub module named Load Assessor Module address of end client and finally routes it to best
(LAM) and IP updater Module (IPM) as shown in Fig. 5. datacenter based on the factors like health status and
The two modules communicate with each other to pass geo location of end client.
the information. The following two sections describe their  Identifying the defunct datacenter: In case of natural
responsibility in detail. disaster or power outage the whole datacenter may be
down. IPM identifies such datacenter and does not
A. Load Assessor Module (LAM)
route the request to such datacenters.
LAM manages the load calculation and load table
preparation. It maintains a periodic interaction with IP C. Working of LAM
updater Module (IPM) and updates load table entries. The most important job of LAM module is to calculate
Thus it has the following responsibility. the load of datacenter. For this we propose a novel
dynamic load calculation algorithm. The algorithm
 Load Assessment and preparation of complete load proposed is depicted in Algorithm 1. Based on the
table: LAM calculates the load of datacenter and calculated load we categorise the health status of
builds it load table in every 1800 seconds (30 datacenter. A physical server hosts many virtual machines
minutes). It also consolidates the load information (VMs) in the virtualised cloud environment. The load of a
received from all other load balancers to prepare server in principle can be calculated as the sum of loads
complete load table with ascending load value. of virtual machines it hosts [5].

686
Copyright (c) 2017 International Journal of Computer Science and Network. All Rights Reserved.
IJCSN - International Journal of Computer Science and Network, Volume 6, Issue 6, December 2017
ISSN (Online) : 2277-5420
www.IJCSN.org
Impact Factor: 1.5

The set of all physical machines (servers) in the D. Working of IP Updater Module
datacenter is S = {S , S , S , … , SN } where N is the total Our model does load balancing at two levels for the first
number of servers in the datacenter. The set of all the level it directs the requests to the least loaded datacenters
hosted VMs on the server Si , Vi is given by and at the second level it further consider the real IP
{Vi , Vi , … , Vim } where m is the number of virtual address for geo sensitive load balancing if the first level
machines on server Si . of load balancing is not cost effective.

The LAM module completes the load table shown Fig. As soon as the IPM receives a request it finds out the IP
4(a) by executing the Algorithm 1 and delivers it to IPM. address of resolving nameserver of end client by running
It also consolidates the complete load table depicted in an applet at the client side. Now to know the authoritative
Fig. 4(b) from the load tables of other load balancers nameservers for the requested domain it sends NS (name
routed to it by other IPMs in the network. server) type query with the fully qualified domain name
in query name field of query message to the DNS of end
The load tables which are exchanged between the load client. Once it receives the response for this query it
balancers are very small in size as they just contain the knows the authoritative nameservers and updates them
record for itself and every load balancer prepares its own with IP address of least loaded datacenter.
complete load table. The process of load table delivery
from one IPM to another IPM is done with the help of At the first level every IPM learns and caches the list of
user datagram protocol (UDP). UDP ensures the best frequently used domain names and their corresponding
effort delivery of load tables. authoritative name servers and sends them the unsolicited
DNS responses consisting of IP address of least loaded
datacenters. For every first request for a particular domain
ALGORITHM 1 which the datacenter receives the IPM module sends a NS
DCHEALTHCHECK type (name server) query with the fully qualified domain
Input: name in query name field of query message to its
1) : Processing power of virtual machine m of resolving DNS server to know the list of all authoritative
server i. nameservers for the domain. It then updates its cache with
2) : Memory resource utilization of virtual all the authoritative servers and then sends them an
machine m of server i. unsolicited A (address) resource record containing the IP
3) : Bandwidth utilization of virtual machine m of address of least loaded datacenter. As a result of this the
server i. authoritative name servers always contains IP address
4) 𝑙 𝑎 𝑎𝑥 , 𝑙 𝑎 𝑟 , 𝑙 𝑎 least loaded datacenter making TTL practically
ineffective as they are forced to update their cache with
Output: unsolicited DNS response.
1) Health status of datacenter.
But we merely do not update the authoritative nameserver
Comments: we also update the resolving nameservers (DNS of ISP)
1) , , are the weights assigned to , of end client. To update the DNS of ISP the model
and < + + < . employs the secure and portable client-server architecture
of java applets. The model will deploy an applet at client
1: 𝑎 ℎ side (end user) to find out its default DNS server and send
them A resource record for the least loaded datacenter.
2: 𝑎 ℎ This ensures that all the clients who are using this
3: ( )← ∗ + ∗ + ∗ nameserver as their default nameserver will always get
the address of least loaded datacenter.
4: ( )+= ( )
5: In cloud environment the least loaded server may not
6: always be the optimal datacenter to deliver the services as
7: ← × ∑𝑁= ( ) response time to mission critical application deployed in
𝑁
8: >= 𝑙 𝑎 clouds is also a critical factor at the global level of load
𝑎𝑥
9: balancing. To cater to the need of less response time, we
10: 𝒍 < 𝑙 𝑎 have employed a second level of load balancing in our
11: model. This balancing is done on the basis of geographic
12: 𝒍 location. Once the request comes to least loaded
13: datacenter the IPM module judges the feasibility of
serving the request on the basis of response time (round
trip time) and distance of the end client. This is done with

687
Copyright (c) 2017 International Journal of Computer Science and Network. All Rights Reserved.
 IJCSN - International Journal of Computer Science and Network, Volume 6, Issue 6, December 2017
ISSN (Online) : 2277-5420
www.IJCSN.org
Impact Factor: 1.5

the help of real IP address of end client and not on the IP practices in order to evaluate its overall performance
address of authoritative or resolving name server which improvements. We will continue to refine our model to
helps in finding the exact location of client and the nearest move towards achieving greater efficiency specially by
least response time datacenter for the request. The IPM efficiently calculating the load of the datacenter in the
can forward the requests from one datacenter to another dynamic cloud environment. We will also deploy and
datacenter on the basis of response time and health status evaluate our model in real time environment.
of datacenters. The request migration algorithm is
depicted in Algorithm 2 for IPM module. References
In case of disaster the IPM will not receive the load tables [1] Fox, Armando, Rean Griffith, Anthony Joseph, Randy
from the defunct datacenter and it will not consolidate the Katz, Andrew Konwinski, Gunho Lee, David Patterson,
load information of such datacenters in the complete load Ariel Rabkin, and Ion Stoica. "Above the clouds: A
Berkeley view of cloud computing." Dept. Electrical
tables. Such datacenters will automatically be left out of Eng. and Comput. Sciences, University of California,
request forwarding business as they do not have an entry Berkeley, Rep. UCB/EECS 28 (2009): 13.
in the complete load table. [2] Rahman, Mosaddequr, Sajid Iqbal, and Jerry Gao. "Load
balancer as a service in cloud computing." In Service
Oriented System Engineering (SOSE), 2014 IEEE 8th
ALGORITHM 2 International Symposium on, pp. 204-211. IEEE, 2014.
MIGRATEREQUEST [3] Lin, Qunyang, Junqing Xie, Zhiyong Shen, and Xunteng
Xu. "DR3: Optimizing Site Selection for Global Load
Balance in Application Delivery Controller." In Open
Input: Cirrus Summit (OCS), 2012 Seventh, pp. 11-15. IEEE,
1) 𝐿 : Health Status of least loaded datacenter 2012.
2) 𝑁 : Health Status of nearest datacenter [4] MacVittie, Lori. "Cloud Balancing: The Evolution of
3) 𝐿 : Response time from least loaded datacenter
Global Server Load Balancing." F5 Networks, Inc
(2010).
4) 𝑁 : Response time from nearest datacenter
[5] Wollman, William V., Harry Jegers, Maureen Loftus,
5) 𝐿 : IP address of least loaded datacenter and Caleb Wan. "Plug and play server load balancing
6) 𝑁 : IP address of nearest datacenter and global server load balancing for tactical networks."
In Military Communications Conference, 2003.
Output: MILCOM'03. 2003 IEEE, vol. 2, pp. 933-937. IEEE,
2003.
1) IP address of datacenter to fulfil request
[6] http://support.citrix.com/article/CTX123976
[7] O'Neil, Kevin, Robert Nerz, and Robert Aubin. "System
1: for balancing loads among network servers." U.S. Patent
= Application 10/162,419, filed June 4, 2002.
[8] Hsu, Ivy Pei-Shan, David Chun Ying Cheung, and
2: 𝑁 <= 𝐿
Rajkumar Ramniranjan Jalan. "Global server load
balancing." U.S. Patent 7,454,500, issued November 18,
3: 𝑁 2008.
4: else if [9] Bahl, Pradeep, Feng Sun, Bernard D. Aboda, and Arnold
5: 𝐿 S. Miller. "System and method for performing client-
4: centric load balancing of multiple globally-dispersed
servers." U.S. Patent 7,653,700, issued January 26, 2010.
5: 𝒍 𝑁 =
[10] Bourke, Tony. Server load balancing. " O'Reilly Media,
6: 𝐿 Inc.", 2001.
7: [11] Tenereillo, Pete. "Why DNS Based Global Server Load
Balancing (GSLB) Doesn't Work." Aquired at:
http://tenereillo. com/GSLBPageOfShame. htm 17
(2004).
5. Conclusion and Future Work [12] Cardellini, Valeria, Michele Colajanni, and S. Yu Philip.
"Dynamic load balancing on web-server systems." IEEE
Internet computing 3 (1999): 28-39.
In this paper, we have proposed a new model for global
[13] http://www.communitydns.eu/global-load-balancing-
server load balancing which gives resilient and service.html
intelligently balanced datacenters. Balancing load at two [14] Krapf, E. "Alteon's Global Server Load Balancing."
levels this framework considers the current health status Business Communications Review (1999): 60.
of datacenter to deliver services from the least loaded
datacenter to reduce the response time. It calculates the
geographic location of end user with the help of its IP
address to deliver services from the least loaded
datacenter which is nearest to the user. Our future
research will be directed towards comparing the
implementation of proposed model with the existing

688
Copyright (c) 2017 International Journal of Computer Science and Network. All Rights Reserved.
View publication stats