Chapter 2

Traffic analysis and bandwidth requirement for the traditional

VOD architectures

2.1 Chapter Introduction

This chapter provides a planning methodology to analyze the traditional VOD architectures,
to determine their bandwidth requirements and traffic flow. It examines all the key parameters
that have influence on VOD communication networks. The initial rough estimation of the
requirement bandwidth for inbound and out bound traffic in the different VOD architectures
is described subsequently in this chapter. The VOD system allow users to access video
services, such as entertainment movies, advertisements, electronic encyclopedia, interactive
games, and educational video documentaries from the video servers through a broadband
network. Potential new applications based on VOD system include video information
retrieval services, collaboration and conferencing systems, and distance learning [20]. To
meet and design objectives and to fulfill the requirements of a VOD service, VOD system
require continuous data transfer over relatively long periods of time, media synchronization,
very large storage, and special indexing and retrieval techniques to adapt multimedia data
types. Therefore, a cost effective design for a VOD system needs to evaluate a collection of
various VOD system components [21]. In terms of the VOD transmission network, the
system design must guarantee requires amount of bandwidth for video traffic, and this
bandwidth must support the VOD service quality which needs to overcome the packet loss.
Video content delivery consumes large amounts of bandwidth in these networks due to its
scalability, i.e. it must be able to support a large number of clients, thus imposing a heavy
burden on the network and the system resources. A high definition (HD) stream, for instance,
may require 10Mbps or more bandwidth under MPEG-2 encoding. Therefore, any network
link that handles many subscribers, each capable of demanding one or more VOD streams,
must have enough bandwidth to meet the user’s demands. In addition, VOD system client
must comply with the necessary buffer size and video request rate for the VOD delivery
policy [22]. Video on demand over broadband networks has been a prolific area of research

9
[21-28]. The particular problem of provisioning VOD bandwidth, system deployment and
architecture design have also received extensive attention in the literature [29-39]. Victor o.
k. Li et. al., develops a performance evaluation tool for the system design and a user activity
model to describe the utilization of network bandwidth and video server usage [25]. Frederic
Thouin et. al., did an extensive survey in video-on-demand networks, their design approach
covered for future research challenges [26]. Smith D. E. presented a mathematical model to
determine the require bandwidth for surfing in multicast topology [30]. Scalable architectures
in VOD system for broadband operators based on the P2P streaming concept together with
scheme equipment and allocation, are discussed in the references [31-32] and [29, 33]
respectively.
This chapter is organized as follows. In Section 2.2, the architectures and topologies that
considered in traditional VOD networks, as well as the access mechanisms and out bound
traffic pattern is presented. Section 2.3 presents the planning methodology to develop an
estimate outbound channel requirements in VOD system. Section 2.4 presents simulation and
results of traffic analysis based on the relevant parameters of VOD network. The last section
is summarizes the key issues and the conclusions to this chapter.

2.2 VOD Network Components

This section describes the basic component required for traditional video on demand system.

2.2.1 Basic Components of the VOD System

A typical VOD broadband network consists of a number of remote client clusters that
communicate their video requests via the network outbound links (client-to-server), and their
video broadcasts via the network inbound links (server-to-clients). The components of the
VOD network comprises the service control point(s), intelligent peripherals (such as
multimedia storage servers, set-top boxes, cluster switches), and primary multimedia routers;
these determine the system performance and communication costs. The IP-based video
delivery encodes all video, whether broadcast or on demand video, into IP data packets and
transmits them to subscribers over IP networks [174]. The VOD system architecture designs
range from the simplest centralized system to a complex distributed system. The architecture
of VOD system is based on the incorporation of continuous media into a large array of

10
extremely high-capacity storage devices, such as optical or magnetic disks. Those are
randomly accessible, with a short seek time, and permanently on-line [23, 27]. In general,
video object delivery from the server to the client may be composed of multiple media
streams, such as audio and video. The retrieve video objects must process not only maintain
continuity playback for each of the constituent media streams, but also preserve the temporal
relationships among the video frames [24]. Numbers of network architectures exist for
producing VOD designs that minimize network costs and fulfill the service quality
constraints. Initially, this chapter focuses on two main widely used architectures: Centralized
Network and Distributed Proxies. More on this is described in what follows:

2.2.1.1 Centralized Network Architecture

In centralized architecture, all remote clusters communicate with the network’s centralized
primary servers through public network, which represent the backbone of the network; there
are no local servers. All client requests are received by the primary server’s router, which acts
as a gateway where data decoding, de-multiplexing, regeneration, multiplexing, encoding,
and carrier switching takes place. The primary servers then retransmit the video data
information to the destination clients via the inbound links. The basic characteristic of the
centralized video-on-demand system, as depicted in Figure 2.1, is that the multimedia
information is always transported on demand from the central multimedia server to the
subscribers through the network. If the server fails or becomes incapable of supporting
existing connections, these connections will be blocked [22]. This solution suffers from very
significant scalability problems, especially when scaled up for millions of potential users.
Providing access to a large library of pre-encoded content using this approach requires
enormous servers with enormous network connections [31].

11
 Media
Archive

Switch
Server

Switch Media
Server Archive

Server

Public Networks

Client Client

Client Client

Client

Figure 2.1: Centralized video-on-demand network

2.2.1.2 Distributed Proxy in VOD Architecture

In distributed proxy architecture, proxy servers are installed at strategic locations in the
network (closer to the clients). Remote clusters can communicate with the centralized servers
as well as with its local server. Each local server can support a number of customers
connected to it through a cluster switch via proxy. The customers are connected to the central
server's through the connected cluster proxy, which acts as an interface between the client
cluster and the broadband network. The cluster proxy and proxy will be used interchangeable
throughout this thesis. The idea of distributed proxies in VOD system (as shown in Figure.
2.2) are used to distribute the centralized multimedia server functions within the network
using the concept of local storage [Ref]. If the user cannot be served by the local multimedia

12
server for any reason, such as the blockage at the local multimedia server, or the multimedia
information is not available in the proxy server as well as local server, then the request of the
user will be transported to the centralized multimedia server. By locating the proxy server
close to the user, it is expected that there will be significant reductions in the load on the
system as a whole [26-27]. Another advantage of the distributed proxy in video-on-demand
system is that it can be expanded in a horizontal manner for system scalability and evolution.
It can start from an initial two-level system (with a centralized multimedia server and one
local storage server) to a system with as many local servers as needed. Compared with the
centralized multimedia system, the distributed system may utilize a lower than average
network bandwidth and has higher system reliability, but at the expense of needing a
significant amount of local storages in the system.

Client Cluster
Client Cluster Home
Home
Cluster
Switch

..
. .. ..
.. Cluster . .
.. .. . Switch
. .
Set-top
Set-top Cluster Cluster Box
Box
Proxy Proxy
Local
Local Server
Server
Broadband
Network

Central ..
.. . .. ..
.. .. . . .
. .

Cluster Cluster
Proxy Central Proxy
Switch
Local Server Local
Server

Media Storage Servers

Figure 2.2: Distributed local proxy in video-on-demand network

13
2.3 VOD Network Planning Methodology and Traffic Models
This section describes the overall bandwidth provisioning methodology, based on a functional
view of a typical VOD network architecture. Traffic models are developed for the purposes of
estimating the amount of bandwidth requirements for VOD system. The user requests are the
concurrent normal request and interactive request. The number of concurrent requests for
accessing video at a certain time point can be estimated as the sum of the active requests
lengths within an interval time is divided by this interval time according the Proposition 2.3.0.
Proposition 2.3.0
Statement: The number of concurrent requests for accessing video at a certain time point
can be estimated as the sum of the active requests lengths within an interval time is divided
by this interval time.
Proof: First we present the login session of the customer, then we proceed to estimate the
Concurrent

average number of (approximate) concurrent users. A login session is a time interval defined
users

by a start time and end time. There are one or more than one system resources being held. For
the web based video on demand system, it requires user authentication. A login session starts
from the time when the user logs on the remote server or to the video storage via proxy
server. The session is ended when the users logs out from the system. A user session (which
consumes system memory) is created for each individual login session. The length of the
login session is the difference between the start time and the end time. So, the number of
concurrent ‘video on demand’ user at a particular time instant is defined as the number of
login sessions fall into that time instant, presents in figure 2.3 .
Time

Login session 2
Login session 1

Login session 3

Figure 2.3: login session of concurrent VOD users

Login session 4

14
The normal request and the concurrent normal request will be used interchangeably
throughout this thesis. The horizontal line presents the time. Each horizontal line segment
parallel to the time axis represents a particular login session. According to the figure 2.3, the

vertical line at time t0 intersects with three different login sessions. Here, the number

concurrent users at time t0 is equal to three. If there are ' r ' number login session and out of

which ' q ' number of login session

( q p r ) intersects the vertical line through the time t0 , then

the number of concurrent users at time t0 is ( ) . The time variable t is considered in the time
q

t �[ 0, T ]
interval from 0 to an arbitrary time instant T , i.e. 0 �t �T such that .
( t)
Let, f (t ) be a real valued function, presents the number of concurrent users at time . Now,

the consider time period from 0 to T are divided into

( n ) number of equal length sub-

( T n ) and the i
(i ) �(T )
th
intervals. So, each sub-interval has a length interval, ends at n for
i = 1, 2,3,K , n

Figure 2.4: Discrete step time interval

When n is large, the average number of concurrent users is approximated by

1 n �iT � 1 n �T � �iT �
�f � �= �� �f � �
n i =1 �n � T i =1 �n � �n � (1.1)

So, the average number of concurrent users over the period of time from 0 to T is
approximated as the value of summation that is expressed in equation 1.1. If n � �then the
series summation becomes a definite integral. As a result, the average number of concurrent
T
1
f ( t ) dt
T�
user is approximated as 0

Suppose there are m login sessions, in the period from 0 to T and those login sessions have
been numbered from 1 to m . Let the start time and end time of the i login session are si and
th

15
di respectively. We consider the function d i ( t ) is associated with the each i th login session
such that
1, t �[ si , di ]
�
�
di ( t ) = �
0, t �[ si , di ]
� (1.2)
m
f ( t ) = �d i ( t )
So the number of concurrent ‘video on demand’ users is obtained as i =1

On integrating both sides from 0 to T , we proceed

T T m m T
f ( t ) dt = �
� �di ( t ) dt = ��
d i ( t ) dt
0 0 i =1 i =1 0

T m T
f ( t ) dt = ��
� d i ( t ) dt
i =1 0
0 (1.3)
�1 �
��
Multiplying both side of the expression (1.3) with �T �, we get
T T
�1 � �1 �m
� �� f ( t ) dt = �� �� d i ( t ) dt
�T �0 �T �i =1 0
T
d ( t ) dt
� i
di ( t )
The integration, 0 presents the area under the curve . According to the

expression (1.2) we get, i i

( d - s ) �( 1) = di - si
(d -s ) th
Since, i i is the length of the i login session for a user in the VOD system, so the sum
of the active requests is presented in the following expression (1.4).
T
�1 � �1 �m
��
�T �0
� f ( t ) dt = �� �
�T �i =1
{ length _ of _ the _ i th _ log in _ Session}

T
�1 � 1 �m
� � �� f ( t ) dt = �
�� �( di - si )
�T �0 �T �i =1 (1.4)
The right side of the expression (1.4) is the sum of the active request lengths within an
interval time divided by this interval time.

16
 th
Figure 2.4: length of i login session

2.3.1 User Activity Model

Once connected, the user is in one of the two states, the normal or the interactive state. User
starts in the normal state, i.e. the video is played at the normal speed. The user stays in this
state for a period of time, which is exponentially distributed with parameter a , and then the
user issues an interactive operation, such as stop, speed up, etc. The user stays in this
interactive state for some duration of time, which is also exponentially distributed with

parameter b . Then the user goes back to the normal state, from where he/she may again go to
the interactive state. This may be repeated many times until he/she disconnects. Different
types of interactive operations will affect the system in different ways, and these include
various common VOD watching operations such as: Play/Resume, Stop, Pause, Jump
Forward, Jump Backward, etc. [25]. In the steady state, concurrent request for a video movie
or video clip is serviced through a multicasting mechanism [Ref JU IEEE]. The viewers
watch the same broadcast movie or video clip and share one stream of broadcasting, which

lasts for an average period of ( tn ) minutes (equivalent to the normal request exponential with
parameter a ). For movie surfing, the steady state assumption is that the surfer’s interactive
requests are serviced by the network through sending an uncast (one per viewer) short-lived

stream, each for an average period of ( ti ) minutes (equivalent to the interactive request

exponential distribution with parameter b ). Due to their uncast nature, each interactive
request superimposes a significant additional demand on top of the steady state normal
demand. Therefore the bandwidth capacity planning and traffic analysis provision must
include surfing effects. The average client view activity is assumed, as shown in Table 2.1,
but represents typical empirical data [38]. This chapter will cover two types of services

17
available to VOD system users, namely: high definition (HD) and standard definition (SD)
VOD movies.

Table 2.1: VOD service features assumed in traffic analysis

VOD Service Feature Value
Peak viewing period Fri./Sat. nights (3 to 4 hrs./night)
Peak busy per subscriber 7hrs
Normal (steady state) request attempts per movie per 1to 5times
period
Holding time of movie 2hrs

Holding time of interactive request 1to 6 seconds

Interactive request attempts per movie per period 1-4 times

Bandwidth per port Variable; 3-20Mb/s
HD movie bandwidth 8-10Mbps
SD movie bandwidth 3Mbps

2.3.2 Outbound Video Request Distribution

The distribution of request in VOD system generally follows a Zipf-like distribution. The
1 1 1
1, , ,K K , ,K
video pages are marked according to the popularity as
2 3 ( i + 1) .

The relative probability of a request for i th

popular page is proportional to
( 1 ia )
. Here, a is
the Zipf parameter, with 0 < a < 1 and typically taking on some value less than unity. The
request distribution rarely follows the strict Zipf law (for which a =1) [36].
For Zipf-like distribution, the cumulative distribution that “one of the k popular” movie is
accessed (i.e. the probability for a popular movies request) is given asymptotically by

18
 k
d
y ( k ) = �a
i =1 i

Here, d is the proportionality constant and 0 p a p 1 , a is the Zipf parameter.

k
1 � 1 1 1 �
y ( k ) = d �a = d �1+ a + a +L + a �
i =1 i � 2 3 k �

k 1
d �a dx
Since a f 0 , this series is convergence and related to the integral 1 x

We get,

d
y ( k) = k 1-a - 1�
�
1-a � �

Neglecting the negative part we get

d
y ( k) � k 1-a �
�
� �
1-a (2.1)

For, k � N as N is the total number of movies or video in the VOD system. We get,

�d ��k 1-a ��
1-a � � ��
lim k � N y
�
� ( k ) �
�= lim k �N �
�
k ��= lim d �
1-a � �
k �N
� � �1 - a �
� �
Left hand side is the cumulative distribution function and the total number of movies or video
present in the system is at most N .
y ( N) y ( N ) �1
The cumulative probability value of is approximately consider as
So we get,
�N 1-a �
d� ��1
�1-a �
1-a
d ޻ 1-a
N

(1 - a )
d � (1-a )
Here a is the ziff parameter, and N where N is the total number of movies or
videos in the system.

So, Y ( k ) is approximated as:

19
 Y ( k ) �( k / N ) (1-a ) (2.2)

Because,
( k / N ) p 1 for all meaningful ( k ) , and a larger a increases Y (k ) , most of the
requests are concentrated on few popular movies [24, 28]. Based on this, it can estimate the

probability of a request for an unpopular movie, for a VOD system with N total movies and
k popular ones, as:

Punp = 1 - (k / N )1-a
(2.3)

The analytical methods for provisioning links are assumed to be in a steady state in busy hour
traffic. In steady state, multicasting is used to reduce VOD traffic volumes. The network
needs to deliver only one video stream (one video server port) for a group of viewers
(multicast group) watching the same video or broadcast program segment. The steady state
demand is therefore the total bandwidth of all video streams (or server ports) in use. There are
two types of request in the VOD network; the first one is the request for initializing or
starting the video (labelled normal request in the study). The other type is the request for
interactive service (e.g.1 stop/pause, jump forward, 2fast reverse, etc.) to be performed on the
viewed movie (labelled interactive request in the study). Since each of these requests is
independent from each other, and the arrival requests come from large numbers of client set-
up terminals, the arrival process of concurrent normal requests, as well as of interactive

requests, to each video server can be modelled as a Poisson process with average rates ln and

li respectively. With this assumption, the distribution of the sum of k of independent

identically distributed random variables, represent the request inter arrival times, is then the
Erlang distribution.

2.3.3. Traffic Model for Centralized Architecture

This subsection presents the traffic models used to determine the VOD system bandwidth
required for a ‘no blocking’ service. The model is based on the Erlang-B formula with
different values of blocking probability. Main aim is to estimate the number of server ports

20
supported by the down broadband link channel (from video server to client set-top box).
Using the estimated number of server ports, the VOD system bandwidth required is then
simply determined by multiplying the movie rate (according to the movie being HD or SD)
with the number of server ports, determined from the Erlang -B formula. The total demand
equals to the sum of the server ports and the movie bandwidth per port streams.
It can be preceded as follows:
Let Mc = Total network traffic in Erlang for a centralized system.

( x) �(h) �( p) �(ln ) �(tn ) ( x) �(h) �( p) �(li ) �(ti )

Mc = +
( Z ) �(T ) T (2.4)

Where x = Number of VOD system cluster areas,

Z = Multicast factor (i.e. number of viewers who request the same multimedia movie within a
short period of time, thus it can be served from the same server port),
h = Number of houses in VOD system cluster service area,
p = Penetration of service in a VOD system cluster area,

ln = Average number of normal (concurrent) request attempts per movie per period per

household,
li = Average number of interactive request attempts per movie per period per household,

tn
= Holding time of a normal (concurrent) request for a movie in minutes,
ti
= Holding time of an interactive request in minutes,
T = Peak busy period in minutes.

Now, the number of server ports supported by the VOD network

( SC ) can then be found

using the Erlang-B formula, from the total calculated network traffic
( MC ) with a given

blocking probability PB , where

( M C ) SC
( SC )
PB = SC
(M C ) N
� N
N =0 (2.5)
And the required bandwidth is then

21
 Wc = ( S c ) �( r )
(2.6)

Where r is the movie rate (e.g. 3Mbps for SD movie), and the allocated bandwidth per
household (Wch) is given by

Wch =
( r ) �(Sc )
( x ) �(h) (2.7)

2.3.4 Bandwidth Requirements for Distributed Local Proxy

This subsection conducts a traffic analysis to determine the number of servers needed for a
distributed local proxy in VOD architecture. The assumption is that the system will store the
unpopular materials on a high capacity optical disk, while the popular materials are stored as
on-line mass storage, in both the local and the centralized servers.
To simplify the analysis, it is assumed that most of the cluster-area traffic through the
broadband link is for non-popular requests. Therefore, there is no blockage for popular
request traffic at the local server. This assumption is justified as it is expected that local
servers will contain all popular movies.

We proceed as follows:

MCL = Total traffic supported by broadband network from all cluster areas in the distributed

local system,
M L = Local traffic at the local cluster area,

S L = Number of server ports supported by the Cluster area VOD local servers,

SCL = Number of server ports supported by the centralized primary servers,

Punp
= Probability of unpopular request,
which is given by Equation (2.3) as
Pun = 1 - ( k / N )1-a

22
Now M L can be found from:

(h) �( p ) �(ln ) �(tn ) ( h) �( p) �(li ) �(ti ) �

�
ML = � + �(1 - pun )
�
� ( Z ) �(T ) T � (2.8)

tn , ti , ln , li , T , Z , x
With, 3 , and p as defined before in Equation (2. 4).

SL can be found from ML using the Erlang-B formula for a given blockage probability PB as:

( M L ) SL
( S )!
PB = SL L N
(M L )
�
N =0 N! (2.9)

Now MCL is calculated as follows:

( h ) �( p ) �( ln ) �( tn ) + ( h ) �( p ) �( li ) �( ti ) �� P
�
M CL = � � ( un )
� ( Z ) �( T ) T � (2.10)

Similarly it can find SCL from MCL using the Erlang-B formula. The local cluster-area
bandwidth WLL and the bandwidth to central servers WLC can be found by multiplying the
corresponding number of ports with the movie bandwidth. Finally the total required server
ports per cluster area are then the sum of SCL and SL, the total bandwidth per area is given by:

TWLC = ( r ) �(SCL + S L ) = WL + WLC

(2.11)

The allocated bandwidth per household Wch is given by:

( r ) �(SCL + S L )
Wch =
h (2.12)

The overall system bandwidth is

TW = ( x ) �( TWLC )
(2.13)

2.3.5 Outbound Channel Requirements

Answer: This subsection presents the study for requirements of network outbound channel
that supports the VOD system. The client links are implemented using trucking technique to
provide many users with accesses to VOD system by sharing multiple lines or links. This is
needed as client links may implemented using asymmetrical connection technologies such as

23
ADSL or VSAT. Such links work under the assumption that the clients will download more
information than they send. For the video clients, the analysis includes an evolution of the
requirements in terms of the number of cluster areas that can be supported as a function of
request message length and rate.
Let the outbound channel bandwidth is equal to WOC , the number of bits per normal request is
equal to Ln . The number of bits per interactive request is Li . The interactive request rate per
house per period is li , normal request rate per period per house is ln . Now, ' T ' is the busy
peak period. The number of VOD system cluster areas are ' x ' . Number of houses in VOD
system cluster service area is h and penetration of service in a VOD system cluster area is p .
(t )
Average duration of a normal request is tn and average duration of an interactive request i .
The rough estimate of Woc for the (both centralized and distributed local proxy) VOD system
is as follows:

( p ) �( x ) �( h ) ��
( ln ) �( Ln ) + ( li ) �( Li ) �
� �
Woc =
T (2.14)

The traffic intensity offered by each user is equal to the request rate multiplied by holding
� ( l ) �( Ln ) + ( li ) �( Li ) �
time. That is each user generates a traffic intensity of � n �Erlang.
Erlang is a dimensional less traffic unit.

The VOD system containing

( x ) �( h ) i.e. xh number of users and an unspecified number of
channels.

( ( x ) �( h ) ) ��
�
The total offered traffic intensity is given as �
( ln ) �( Ln ) + ( li ) �( Li ) �
� �
�
�Earlang.

In a ' c ' channel trunked system, if the traffic are equally distributed among the channels, then
the traffic intensity per channel Ac (say) is given by

( x ) �( h ) ��
( ln ) �( Ln ) + ( li ) �( Li ) �
� �
Ac =
c (2.15)

2.4 Simulation and Results

This section presents the traffic analysis to determine the VOD system’s inbound and
outbound channel bandwidth requirements for a ‘non-blocking’ service. The traffic analysis

24
is carried out using the developed traffic models with different values of blocking probability.
The VOD system is analyzed for various video movie resolutions including standard
definition (video and audio) and high definition video with an average rate per port of 3Mb/s
and 8Mb/s respectively.

2.4.1 Traffic Analysis for Centralized Architecture

The effects of some of the centralized system parameters (such as the multicast factor, movie
holding time, and the average number of concurrent requests arriving to the system during the
peak period) on the required centralized system, VOD channel bandwidth are shown in Table
2.2.

Table 2.2: Bandwidth required for centralized VOD system (Mb/s)

Multi Cast VOD System Number of Require Bandwidth Wc
Factor
Traffic Mc (Erlang) Server Ports Sc Mbps (SD) Movie
Z

tn = 120, ti = 6,T = 420,

Key parameters: x = 250, p = 0.4, h = 600, l n = 1.5, l i = 4,
blocking probability =0.01

HD Movie rate =
Total SD Movie rate = 3Mbps 8Mbps
Centralized Number
network of VOD Centralized
traffic Mc servers network Bandwidth
ports Sc bandwidth per house
Multicast Wc Gbps Wch Kbps Wc Gbps Wch Kbps
factor (Z)

8571 8553 25.659 171.06 48.024 456.16

5
6000 6003 18.009 120.06 41.216 320.16
10
5143 5152 15.456 103.04 37.808 274.77
15
4714 4726 14.178 94.52 35.768 252.05
20
4457 4471 13.413 89.42 34.408 238.45
25
4286 4301 12.903 86.02 33.432 229.39
30

25
 4163 4179 12.537 83.58 48.024 222.88
35

It is clear that the VOD system with multicast streams for this sample data can reduce the
required bandwidth by 50% (12.537/25.659 = 48.8%), that a unicast system (z = 1) would
use. Also, high video movie resolution would increase the required bandwidth significantly.
Figures 2.3 and 2.4, present the plotted relationship between the inbound channel bandwidth
needed to serve those customer requests with various values for the multicast factor, the
number of cluster areas, and the interactive request rate.

Figure 2.3: Bandwidth versus multicast factor for VOD centralized system

Figure 2.4: Bandwidth versus Number of local cluster areas X for VOD centralized system

As expected, the bandwidth of the system is also dependent on the size of the client
population for a given resolution and blocking probability. For a low multicast factor (e.g. z

26
less than 10), reducing the number of local cluster areas (X) results in a significant reduction
of the required bandwidth.

Figure 2.5, presents the relationship between the inbound channel bandwidth and the
interactive traffic request rate and multicast factor. As shown, the increment of the interactive

traffic, (i.e. increment of li due to movie surfing), return savings in the bandwidth is made by
using multicasting for movie delivery.

Figure 2.5: Bandwidth versus Interactive traffic request rate for VOD centralized system

2.4.2 Traffic Analysis for Distributed Local Proxy Architecture

The impact of system parameters (such as the multicast factor, movie holding time, and the
average number of requests arriving to the system during the peak period) on the distributed
local proxy in VOD system with the required channel bandwidths are shown in Table 2.3.

Table 2.3: Bandwidth required for distributed local proxy VOD architecture
Z l n = 1.5, l i = 4, tn = 120, ti = 6, T = 420
x = 250, p = 0.4, h = 600,
ML SL WLL MLC SCL WLC TWLC Wch
Mbps Mbps Mbps Mbps
5 27.4 39 117 6.9 14 42 159 0.265

10 19.2 29 87 4.8 11 33 120 0.200

27
 15 16.5 26 78 4.1 10 30 108 0.180

20 15.1 24 72 3.8 10 30 102 0.170

25 14.3 23 69 3.6 9 27 96 0.160

30 13.7 23 69 3.4 9 27 96 0.160

35 13.3 22 66 3.3 9 27 93 0.155

In Figures 2.6 and 2.7, present the relationship between the outbound channel bandwidth and
the concurrent normal video request with increase rate and multicast factor.

Figure 2.6: Bandwidth versus normal (concurrent) traffic request rate for local proxy VOD architecture

Increasing the mean of the concurrent normal request rate will require more servers to satisfy
the system blocking probability, and consequently need a higher network bandwidth.

Figure 2.7: Bandwidth versus multicast factor for local proxy VOD architecture

Figure 2.8 shows the VOD system bandwidths for the local proxy architecture as a function
of interactive request arrival rate. Again, this rate has a strong impact on the system
bandwidth.

28
 Figure 2.8: Bandwidth versus Interactive traffic request rate for VOD local proxy architecture system

2.4.3 VOD Outbound Channel Analysis

This subsection presents the requirement analysis of the network outbound channel (from
cluster area clients to VOD servers). The analysis includes the evaluation of the requirements
in terms of request message length and rate. Figure 2.9, presents the effect of the normal
request message length in bits on the centralized system. Outbound channel bandwidth
requirement keeps the interactive rate and message length fixed at 1.5 requests per period and
512 bits respectively. As shown, the number of clients (or alternatively the required
bandwidth) increases as the mean of the message length or the normal request rate increases.
Thus, minimization of message length can significantly reduce the required outbound channel
bandwidth. In fact, the system capacity may be limited by the available asymmetrical
outbound channel bandwidth as it controls the maximum number of requests for movie
retrieval or surfing.

29
 Figure 2.9: Effect of the normal request message length on the
outbound channel bandwidth in VOD System

2.5 Conclusions

This chapter deals with the examination of traffic analysis by considering the basic planning
to determine the inbound and outbound bandwidth requirement with fixed blocking
probability in centralized system and ‘no blocking’ in the distributed proxies architecture.
The figures 2.3 and 2.4 present the bandwidth requirement with respect to the number of
clusters. The viewers from the different clusters initiate interactive request with various rates
in centralized VOD system. It has been observed from figures 2.3, and 2.4, that the demand
for the channel is influenced by the increasing number of multicast cluster, but the demand
for bandwidth sharply increases with the number of clusters. Similarly, figure 2.5
shows higher interactive users require more bandwidth.
The figures 2.6 and 2.7, present the relationship between the outbound bandwidth demand
and the concurrent normal video request, with the rate and multicast factor in distributed local
proxy architecture.
It exhibits that for gradual increase in the mean for the concurrent normal request rate will
require more servers to overcome the system blocking, and consequently, it deserves higher
network bandwidth.
Figure 2.8, 2.9, shows that the bandwidth requirement at the local server, in the distributed
local proxies in VOD system rapidly grows with the outbound interactive traffic request with
variable request size.
To enhance the performance of VOD system, admission control at proxy server, dynamic
page replacement at cache memory of the proxy server, estimation about the aggregate
bandwidth requirement at the proxy server during on demand page replacement is required.
The necessary modification is required for the traffic handle policy, to enhance the
effectiveness of VOD system for the exponentially growing number of users, for versatile
areas of deployment in Networking and Communication.

30