A DIFFERENTIATED SERVICE BASED CONTENT
DISTRIBUTION METHOD IN CDN
Binjie Zhu, Ke Xu and Renjie Pi
Beijing University of Posts and Telecommunications, Beijing, China
Keywords: Content delivery, Overlay network, Differentiated service.
Abstract: With the development of Internet applications, there are more and more problems about delay in the
Internet. Content distribution network is an overlay network to distributing the content to the edge of
Internet to reducing the user perceived delay. This paper focuses on the content distribution problem in the
content delivery network. A differentiated service based content distribution algorithm was designed in this
paper. Clustering algorithm was used to get pedigree chart of the edge server, and then ant algorithm was
used to search for the optimal edge server based on the pedigree chart. The analysis shows that this
algorithm combines the clustering method and ant colony algorithm, and it has a good distribution of
results.
1 INTRODUCTION
CDN (Content Delivery Network) use distributed
caching and the load balancing techniques on the
Internet to build a distributed overlay network. This
overlay network pushing the content from the source
to the network edge equipment. On one hand, it
allows the user to access to near location server with
the desired content, reducing the end to end delay,
improved customer service quality; On the other
hand, the central server bottleneck was broken, and
reducing the backbone network traffic. It effectively
alleviating the throughput of backbone network, and
increasing of system capacity.
There are three layers in the CDN overlay
network. The first layers is source server where the
source content object come from. The second layer
is the copies server layer. This layer is used to
storage the copy of the content for distribution. The
third layer is the user layer, obtains the content from
the copy servers. The copy server is put in the edge
of internet, so the network traffics are pushed to the
edge of the Internet (George Palls, 2006). It is the
basic idea of content delivery network (CDN).
We designed a P2P-CDN with two layers, shown
as figure 1. The edge server in the CDN is divided
into two layers: the wide-area and the local level.
Wide-area node was distribution in different
locations, for the purpose of pushing the content into
the edge of Internet. Local level nodes were in the
same location, and the function is to increasing load
capacity.
Figure 1: P2P-CDN with two layers.
2 BACKGROUND AND
PREVIOUS WORK
After a contract was signed between the CDN pro-
576
Zhu B., Xu K. and Pi R..
A DIFFERENTIATED SERVICE BASED CONTENT DISTRIBUTION METHOD IN CDN.
DOI: 10.5220/0003589705760579
In Proceedings of the 13th International Conference on Enterprise Information Systems (SSE-2011), pages 576-579
ISBN: 978-989-8425-53-9
Copyright
c
2011 SCITEPRESS (Science and Technology Publications, Lda.)
vider and the CDN user, the contents were delivery
to a CDN provider. All the content files were copied
in the source server by the CDN providers. CDN
content were transmission to the edge cache server
by distribution module of CDN according to a
certain distribution means, shown as figure 2.
Figure 2: Content distribution.
The existing distribution method (Adrian J. 2006)
(S. S. H., 2003) does not support service
differentiation. All content objects as a whole, does
not distinguish the different server quality
requirements between different content. The
proposed method in this paper supports
differentiated Service and supports optimization for
each content object. A contract was signed between
the CDN provider and the CDN user in which
contain the service level requirements of each
content object. Different levels of service
corresponding to different number of copies. Every
content objects has the attributes of service level
requirements. How to distribute the contents of an
object with differentiated Service effectively is the
aim of this paper.
3 DIFFSERV-BASED CONTENT
DISTRIBUTION METHOD
CDN systems included source server, edge servers,
and the CDN management system. CDN
management system including the content
distribution related modules: service differentiation
module, content distribution control module, ant
search module, the delay measurement module,
shown as figure3. Content distribution module's
function is to distribute the content object to the edge
server node.
Delay measurement module is to measure the
network delay between the edge servers. If there are
N edge servers, n
(0 )nN<≤ points selected from
the edge servers randomly. The delay data were
statistics to get the delay date of other points to the n
points. The hierarchical diagram was obtained
through clustering these data. The pedigree chart was
transformed into a standard pedigree tree for a
heuristic algorithm of content distribution.
Figure 3: CDN system modules.
If there are n copies for one content, then n ants
climb out in ant colony search modules. For a set of
content objects, there may be a crowding
phenomenon, in which a lot of content was assigned
to one location, causing unbalance distribution.
However, this distribution unbalanced is necessary
from the view of all the point of positions. The
purpose content distribution network is to reducing
the delay, the best performance in a position to do
distribution copy is necessary. However, it is
necessary to balance distribution in the peer position
with the same effect. In another word, the balance is
necessary form the view of local area or the nearby
position. According to this principle, ants were
crawling, and sprinkle pheromone. Distribution
balance guided by the pheromone information.
The service level argument was obtained from
the service distinction module. The service level
information contains the number of copies for a
content object. The service level as an argument was
transmitted to ant search module. The ant searching
module run heuristic search in the space of edge
server based on the pedigree tree. Contents were
distribution to the optimal position based on the ant
search result. It is shown as figure 4.
A DIFFERENTIATED SERVICE BASED CONTENT DISTRIBUTION METHOD IN CDN
577
Figure 4: Distribution Flow.
4 EXAMPLE
For a detailed description of the algorithm process,
there will be an example. Assuming the six edge
points in the content distribution system, N = {A, B,
C, D, E, F}, every point is one edge server, shown as
the figure 5. Because there are fewer number of
point, so all the six points were selected for the delay
monitoring points. All the edge server point has a
distance to this 6 monitoring points, and it is forms a
matrix.
Figure 5: The edge servers.
Pedigree chart was obtained through the
hierarchical clustering of these points base on the
matrix. As shown in figure 6.
The chart was transformed into a pedigree tree,
as shown in Figure 7. Suppose there is a content
object, the content object needs to have three copies.
Because there are n copies of the content to be
distributed, so there are three ants crawling in the
pedigree tree. After the ants search in pedigree tree,
three positions got for the three copies.
Figure 6: Pedigree chart.
Figure 7: Standard pedigree tree.
Ants crawl in the pedigree tree with 3 deep. As
ants climb with a depth of 3, each ant is faced with
two choices. For example, there was an ant facing
the choice of points 2 and 6. According to the crawl
rules in front, it can randomly select the point, but
the pheromone was sprinkled in the choice path, in
order to facilitate the next ants to choose a different
path for load balancing purposes. Finally, the three
servers choose are shown in Table 1.
Table 1: The result of distribution.
Copies C1 C2 C3
Edge servers 2 1 4
5 CONCLUSIONS AND FUTURE
WORK
This paper studies the content delivery network
content distribution problem. Different quality of
service has a different number of copies, was
distributed with service differentiation. The contents
ICEIS 2011 - 13th International Conference on Enterprise Information Systems
578
with high QoS level were distributed to the optimal
position to achieve utility maximization. This paper
presents a hierarchical clustering based heuristic
method and ant colony algorithm to solve this
problem. Future work is to further optimize this
method.
ACKNOWLEDGEMENTS
This work is supported by the National Key project
of Scientific and Technical Supporting Programs of
China (Grant Nos. 2008BAH24B04,
2008BAH21B03, 2009BAH39B03); the National
Natural Science Foundation of China (Grant No.
61072060); the Program for New Century Excellent
Talents in University (No. NECET-08-0738);
Engineering Research Center of Information
Networks. Ministry of Education.
REFERENCES
Adrian J. Cahill, Cormac J. Sreenan, 2006, An efficient
resource management system for a streaming media
distribution network, Intreactive Technology and
Smart Education, Vo. 3, No. 1
George Palls, Athena, Vakali, 2006, Insight and
Perspectives for content delivery networks,
Communications of the ACM, January, Vol. 49, No. 1
S. S. H. Tse, 2003, Approximate Algorithms for Document
Placement in Distributed Web Servers, IEEE
Transactions on Parallel and Distributed Systems,
Vol. 16, No. 6, pp. 489-496
A DIFFERENTIATED SERVICE BASED CONTENT DISTRIBUTION METHOD IN CDN
579