You are on page 1of 4

International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169

Volume: 5 Issue: 3 418 421

A Distributed Control Law for Load Balancing in Content Delivery Networks

Mr. Shendage Swapnil Sunil1 Prof. Dhainje Prakash B.3 Dr. Deshmukh Pradeep K2
PG Student, Department of CSE,SIETC, Vice-Principal, SIETC, Paniv, Solapur Principal, SIETC, Paniv,
Paniv, Solapur University,Maharashtra, University,Maharashtra, India3 SolapurUniversity,Maharashtra, India2

AbstractLarge Internet-scale distributed systems deploy hundreds of thousands of servers in thousands of data centers around the world.
Internet-scale distributed system to have emerged in the past decade is the content delivery network (CDN, for short) that delivers web content,
web and IP-based applications, downloads, and streaming media to end-users (i.e., clients) around the world. This paper focuses on the main
research areas in the field of CDN, pointing out the motivations, and analyzing the existing strategies for replica placement and management,
server measurement, best fit replica selection and request redirection. In this paper, I face the challenging issue of defining and implementing an
effective law for load balancing in Content Delivery Networks. A formal study of a CDN system, carried out through the exploitation of a fluid
flow model characterization of the network of servers. This result is then leveraged in order to devise a novel distributed and time-continuous
algorithm for load balancing.
Keywords: Content Delivery Network, Fluid flow model, Load balancing Algorithm.

The over-evolving nature of the Internet brings new
challenges in managing and delivering content to users, since
for example, popular Web service soften suffer congestion and
bottlenecks due to the large demands posed on their services.
A Content Delivery Network (CDN) is a
collaborative collection of network elements spanning the
Internet, where content is replicated over several mirrored
Web servers in order to perform transparent and effective
delivery of content to the end users. Collaboration among
distributed CDN components can occur over nodes in both
homogeneous and heterogeneous environments.
The typical functionalities of a CDN include:
Request redirection and content delivery services, to direct a
request to the closest suitable CDN cache server using
mechanisms to bypass congestion, thus overcoming ash
crowds or SlashDot effects.
Content outsourcing and distribution services, to replicate
and cache content from the origin server to distributed Web
Content negotiation services, to meet specic needs of each FIG.1.1 CONTENT DELIVERY NETWORK
individual user (or group of users).
Management services, to manage the network components, The three main entities in a CDN system are the
to handle accounting, and to monitor and report on content following:
usage. 1. content provider
Figure 1.1 shows the model of a CDN where the 2. CDN provider,
replicated Web server clusters spanning the globe are located at 3. end users.
the edge of the network to which end users are connected. A A content provider or customer is one who delegates
CDN distributes content to a set of Web servers, scattered over the Uniform Resource Locator (URL) name space of the Web
the globe, for delivering content to end users in a reliable and objects to be distributed. The origin server of the content
timely manner. The content is replicated either on-demand provider holds those objects. A CDN provider is a proprietary
when users request for it, or it can be replicated beforehand, by organization or company that provides infrastructure facilities
pushing the content to the distributed Web servers.A user is to content providers in order to deliver content in a timely and
served with the content from the nearby replicated Web server.
IJRITCC | March 2017, Available @
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 5 Issue: 3 418 421
reliable manner. End users or clients are the entities who network elements. Hence, a data exchange process among the
access content from the content providers Web site. servers is needed, which unavoidably incurs in a
CDN providers use caching and/or replica servers located in communication overhead.
different geographical locations to replicate content. CDN
cache servers are also called edge servers or surrogates. The
edge servers of a CDN are called Web cluster as a whole. Now introduce a continuous model of a CDN infrastructure,
CDNs distribute content to the edge servers in such a way that used to design a novel load balancing law. The CDN can be
all of them share the same content and URL. Client requests considered as a set of servers each with its own queue. I assume
are redirected to the nearby optimal edge server and it delivers a fluid Model approximation for the dynamic behavior of each
requested content to the end users. Thus, transparency for queue. I extend this model also to the overall CDN system.
users is achieved Such approximation of a stochastic system.
Let qi(t) be the queue occupancy of server i at time t.
II. LOAD-BALANCING STRATEGIES Iconsider the instant arrival rate i(t) and the instant service
The challenging issue of defining and implementing an rate i(t). The fluid model (Fig. 2) of CDN servers queues is
effective law for load balancing in Content Delivery given by
Networks.The content delivery networks technique is one of
the successful virtual networks rapidly developed over the last
decade with the specific advantage of optimizing the Internet.
Nowadays, the CDN has become one of the most important
For i = 1..N.
parts of the Internet architecture for content distribution. In
Equation (1) represents the queue dynamics over time. In
this article I highlight the innovative technologies in CDNs
particular, if the arrival rate is lower than the service rate, I
and present their evolution triggered by ever newer emerging
observe a decrease in queue length. On the other hand, the
applications. By resenting in-depth discussion about the
queue increases whenever the arrival rate is greater than the
architecture, challenges, and applications of CDNs, I
service rate. In the latter case, the difference in (1) represents
demonstrate their importance for the future Internet.
the amount of traffic exceeding the available systems serving
Akamais DNS-based load balancing system
continuously monitors the state of services and their servers
In DNS-based request-routing, a domain name has multiple
and networks. To monitor the entire systems health end to
IP addresses associated to it. When an end users content
end, Akamai uses agents that simulate the end user behaviour
request comes, the DNS server of the service provider returns
by downloading Web objects and measuring their failure rates
the IP addresses of servers holding the replica of the requested
and download times. Akamai uses this information to monitor
object. The clients DNS resolver chooses a server among
overall system performance and to automatically detect
these. To decide, the resolver may issue probes to the servers
and suspend problematic data centres or servers.
and choose based on response times to these probes. It may
Request-routing mechanisms inform the client about the
also collect historical information from the clients based on
selection of replica server generated by the request-routing
previous access to these servers. Both full and partial-site
algorithms. Several mechanisms have been proposed in the
CDN providers use DNS redirection. The performance and
literature. They can usually be classified as either static or
effectiveness of DNS-based request-routing has been
dynamic, dependingon the policy adopted for server selection
examined in a number of recent studies. The advantage of this
[20].Static algorithms select a server without relying on any
approach is the transparency as the services are referred to by
information about the status of the system at decision time.
means of their DNS names, and not their IP addresses. DNS-
Static algorithms do not need any data retrieval mechanism in
based approach is extremely popular because of its simplicity
the system, which means no communication overhead is
and independence from any actual replicated service. Since it
introduced. These algorithms definitely represent the fastest
is incorporated to the name resolution service it can be used by
solution since they do not adopt any sophisticated selection
any Internet application [89]. In addition, the ubiquity of DNS
process. However, they are not able to effectively face
as a directory service provides advantages during request-
anomalous events like flash crowds.
routing. The disadvantage of DNS-based request-routing is
Dynamic load-balancing strategies represent a valid
that, it increases network latency because of the increase in
alternative to static algorithms. Such approaches make use of
DNS lookup times. CDN administrators typically resolve this
information coming either from the network or from the
problem by splitting CDN DNS into two levels (low-level
servers in order to improve the request assignment process.
DNS and high-level DNS) for load distribution. Another
The election
limitation is that DNS provides the IP address of the clients
of the appropriate server is done through a collection and
Local DNS (LDNS), rather than the clients IP address
subsequent analysis of several parameters extracted from the
IJRITCC | March 2017, Available @
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 5 Issue: 3 418 421
IV. DISTRIBUTED LOAD-BALANCING ALGORITHM Then I have exchange of request among the server node,
In this section, I want to derive a new distributed algorithm which is given by following equation;
for request balancing that exploits the results. It is a hard task to g(Qi(t)) = aij(t) ... (1)
define a strategy in a real CDN environment that is completely jNei
compliant with the model proposed. As a first consideration, for i = 1n.
such a model deals with continuous-time systems, which is not Nei= {adjacent j of node i}, and aij(t) takes the portion of
exactly the equal to the traffic received at node from node if no request injected from node i into node j. The above equation
requests are lost during the redirection process. works on the principle of request redirecting by a high loaded
server i to a neighbouring less loaded server j. The
A. ANALYSIS OF ALGORITHM DESIGN neighbouring server j will handle the request behalf of server i.
In this project, I want to derive a new distributed algorithm for Equation (1) can be written in term associated with client
request balancing that exploits the results are presented. First incoming request at server i and term associated with request
of all, I observe that it is a hard task to define a strategy in a redirected from server to its neighbours
real CDN environment that is completely compliant with the .aij(t)= .aij(t) + .aij(t) ...(2)
model proposed. As a first consideration, such a model deals jNeijNe+ijNe-i
with continuous-time systems, which is not exactly the case in Ne+i(t) ={ adjacent j of server node i: Qj< Qi } and Ne-i(t) ={
a real packet network where the processing of arriving adjacent j of server node i: Qj> Qi } are set adjacent server
requests is not continuous over time. The objective is to derive whose queue is respectively less loaded and high loaded than
an algorithm that presents the main features of the proposed queue at server i.
load-balancing law and arrives at the same results in terms of V. BALANCING PERFORMANCE
system equilibrium through proper balancing of servers
I need to connect 10 server nodes to the interconnected
network, also 10 client nodes, each of the client node should
connect to a single server. I need to model each server nodes as
a Ma/Ma/1 queue with a rate of service of Si request rate of ai.
The implemented algorithm consists of two independent parts:
For every second t, the server exchange its status information
1. a procedure that is in charge of updating the status of the
data to its neighboring server at the same time it gets its
neighbours load,
information table updated. By this for every standard time
2. a mechanism representing the core of the algorithm, which interval server node will do the status update process. So each
is in charge of distributing requests to a nodes neighbours. sever in network have the knowledge of load in the network. So
this distributed network works fine in the simulator because
Even though the communication protocol used for status each individual server node have complete status of network.
information exchange is fundamental for the balancing At last the flash crowd scenario, simulations demonstrate that
process. our proposed well performs the analyzed existing algorithms in
I implemented a specific mechanism: I extended the terms of overall system performance that is availability,
HTTP protocol with a new message, called CDN, which is response time and queue length handling as shown in fig.5.1.
periodically exchanged among neighbouring peers to carry
information about the current load status of the sending node.
Naturally, a common update interval should be adopted to
guarantee synchronization among all interacting peers. For this
purpose, a number of alternative solutions can be put into
places, which are nonetheless out of the scope of the present
Though clients are involved in our proposed system network
they have no significant role other than requesting for service
to the closest server. All thesurrogate servers initialized to
handle the request raised by the client. Our proposed algorithm
will enhance the functionality of the surrogate servers and
mainly the overall system performance as whole.
Let the request queue maximum capacity be Qmax, and the
Qi(t) be the queue occupied by server i at time consider the
arrival rate be ai(t) service rate id gi(t)
Fig.5.1 system performance

IJRITCC | March 2017, Available @
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 5 Issue: 3 418 421
VIII. REFERENCES management, Comput.Netw., vol. 36, no. 23, pp.
[1] S. Manfredi, F. Oliviero, and S. P. Romano, 203235, Jul. 2001.
Distributed management for load balancing in [17] F. Blanchini, R. L. Cigno, and R. Tempo, Robust
content delivery networks, in Proc. IEEE rate control for integrated services packet networks,
GLOBECOM Workshop, Miami, FL, Dec. 2010, pp. IEEE/ACM Trans. Netw., vol. 10, no. 5, pp. 644652,
579583. Oct. 2002.
[2] H. Yin, X. Liu, G. Min, and C. Lin, Content delivery [18] V.Misra,W. Gong, W. bo Gong, and D. Towsley,
networks: A Bridge between emerging applications Fluid-based analysis of a network of AQM routers
and future IP networks, IEEE Netw., vol. 24, no. 4, supporting TCP flows with an application to red,
pp. 5256, Jul.Aug. 2010. Proc. ACM SIGCOMM, pp. 151160, 2000.
[3] J. D. Pineda and C. P. Salvador, On using content [19] D. Cavendish, M. Gerla, and S. Mascolo, A control
delivery networks to improve MOG performance, theoretical approach to congestion control in packet
Int. J. Adv. Media Commun., vol. 4, no. 2, pp. 182 networks, IEEE/ACM Trans. Netw., vol. 12, no. 5,
201, Mar. 2010. pp. 893906, Oct. 2004.
[4] D. D. Sorte,M. Femminella, A. Parisi, andG. Reali, [20] V. Cardellini, E. Casalicchio, M. Colajanni, and P. S.
Network delivery of live events in a digital cinema Yu, The state of the art in locally distributedWeb-
scenario, in Proc. ONDM,Mar. 2008, pp. 16. server systems, Comput. Surveys, vol. 34, no. 2, pp.
[5] Akamai, Akamai, 2011 Online. Available: 263311, Jun. 2002. [21] Z. Zeng and B. Veeravalli, Design and performance
[6] Limelight Networks, Limelight Networks, 2011 evaluation of queue-and-rate-adjustment dynamic
Online. . Available: load balancing policies for distributed networks,
[7] CDNetworks, CDNetworks, 2011 Online. . IEEE Trans. Comput., vol. 55, no. 11, pp. 1410
Available: 1422, Nov. 2006.
[8] Coral, The Coral Content Distribution [22] V. Cardellini, M. Colajanni, and P. S. Yu, Request
Network,2004Online Available: redirection algorithms for distributed Web systems, IEEE Trans. Parallel Distrib. Syst., vol. 14, no. 4, pp.
[9] Network Systems Group, Projects, Princeton 355368, Apr. 2003.
University, Princeton, NJ, 2008 Online. . [23] M. Dahlin, Interpreting stale load information,
Available: IEEE Trans. Parallel Distrib. Syst., vol. 11, no. 10,
[10] A. Barbir, B. Cain, and R. Nair, Known content pp. 10331047, Oct. 2000.
network (CN) request- routing mechanisms, IETF, [24] R. L. Carter and M. E. Crovella, Server selection
RFC 3568 Internet Draft, Jul. 2003 Online. . using dynamic path characterization in wide-area
Available: networks, in Proc. IEEE INFOCOM, Apr. 1997, vol.
[11] T. Brisco, DNS support for load balancing, IETF, 3, pp. 10141021.
RFC 1794 Internet Draft, Apr. 1995 Online. . [25] M. D. Mitzenmacher, The power of two choices in
Available: randomized load balancing, IEEE Trans. Parallel
[12] M. Colajanni, P. S. Yu, and D. M. Dias, Analysis of Distrib. Syst., vol. 12, no. 10, pp. 10941104, Oct.
task assignment policies in scalable distributedWeb- 2001.
server systems, IEEE Trans. Parallel Distrib. Syst., [26] C.-M. Chen, Y. Ling, M. Pang, W. Chen, S. Cai, Y.
vol. 9, no. 6, pp. 585600, Jun. 1998. Suwa, and O. Altintas, Scalable request routing with
[13] D. M. Dias, W. Kish, R. Mukherjee, and R. Tewari, next-neighbor load sharing in multi-server
A scalable and highly availableWeb server, in environments, in Proc. IEEE Int. Conf. Adv. Inf.
Proc. IEEE Comput. Conf., Feb. 1996, pp. 8592. Netw. Appl., Mar. 2005, vol. 1, pp. 441446.
[14] C. V. Hollot, V. Misra,D. Towsley, and W.Gong, [27] R. A. Horn and C. R. Johnson, Topics inMatrix
Analysis and design of controllers for AQM routers Analysis. Cambridge, U.K.: Cambridge Univ. Press,
supporting TCP flows, IEEE Trans. Autom. Control, 1995.
vol. 47, no. 6, pp. 945959, Jun. 2002. [28] F. Cece, V. Formicola, F. Oliviero, and S. P.
[15] C. V. Hollot, V. Misra, D. Towsley, and W. bo Gong, Romano, An extended ns-2 for validation of load
A control theoretic analysis of red, in Proc. IEEE balancing algorithms in content delivery networks,
INFOCOM, 2001, pp. 15101519. in Proc. 3rd ICST SIMUTools, Malaga, Spain, Mar.
[16] J. Aweya, M. Ouellette, and D. Y. Montuno, A 2010, pp. 32:132:6.
control theoretic approach to active queue

IJRITCC | March 2017, Available @