Investigating the Optimal Performance of Multicast Communication
by Simulation
Eleftherios Stergiou, Dimitriοs Liarokapis
and Gerasimos Meletiou
Department of Informatics & Telecommunication Technology, TEI of Epirus, Arta, Greece
Keywords: Simulation, Performance Evaluation, Multicast, Protocols, Multi-criteria, Quality Specification.
Abstract: The multicast mechanism is one method of data communication in data networks which aims to transfer
data to a group of receivers on a network in an efficient manner. In this work the performance of four well
known and widely used multicast protocols are investigated using OMNeT++ open software, which was
chosen for this purpose. Individual performance metrics are determined by executing simulation
experiments and in addition a unique overall performance indicator is defined to solve the multi-criteria
decision problem that is revealed as the network configuration and the service conditions vary. This
performance evaluation approach can be used by network protocol designers for building and exploiting
optimal protocols when setting up networks so as to achieve the best performance under the multicast traffic
load and quality specifications.
1 INTRODUCTION
Today, as multimedia applications are increasing,
they need efficient distribution to a large number of
end users over large distances. This trend demands a
multicast means of distribution rather than a unicast
one. Multicast transmission decreases the usage of
network resources compared to unicast and
broadcast transmission, which target one recipient
and all recipients in a segment, respectively. Figure
1 depicts the differences between unicast and
multicast transmission. In our effort to investigate
the performance of multicast communication in
various types of data networks, our interest is
focused on the performance of the protocols which
are used in this communication technique. By now a
number of multicast protocols have been
implemented and tested separately that work
perfectly on the internet and intranets. A well-
balanced and effective protocol must be able to cut
out the quantity of states that is stored in the routers.
In this paper, four well known and widely used
multicast protocols are selected and their
performance is investigated. Two of them belong to
sparse type protocol and that are the Core base Trees
(CBT) protocol and the Protocol Independent
Multicast-Sparse Mode (PIM-SM), while the other
two, the Protocol Independent Multicast-Dense
Mode (PIM DM) and the Distance Vector Multicast
Routing Protocol (DVMRP), belong to dense-type
protocols. The first pair of protocols is suitable for
groups where a very low percentage of the nodes are
subscribed to the multicast session. These sparse-
mode protocols assume relatively small numbers of
multicast clients. In contrast, the dense-type
protocols (PIM DM and DVMRP) are ideal for
groups where many of the end-users subscribe to
multicast packets. Dense-mode multicast routing
protocols flood packets across the network and then
prune off branches where there are no recipients,
while sparse protocols have the ability to explicitly
construct a tree from each sender to the receivers in
the multicast group.
The abovementioned protocols differ
significantly in terms of the approach used to
implement the many-to-many communication
model. They differ in terms of scalability and in that
some of them store state-related information in all
routers while in the remainder only routers currently
participating in the transmission keep these data.
Both types of protocols employ different types of
trees which have advantages and disadvantages that
need to be considered by the network designer. Thus
the comparison of results was considered as an
interesting topic.
To study the performance of protocols other than
by taking direct measurements, which is expensive,
it is possible to use a special purpose simulator or an
108
Stergiou E., Liarokapis D. and Meletiou G..
Investigating the Optimal Performance of Multicast Communication by Simulation.
DOI: 10.5220/0004479401080116
In Proceedings of the 3rd International Conference on Simulation and Modeling Methodologies, Technologies and Applications (SIMULTECH-2013),
pages 108-116
ISBN: 978-989-8565-69-3
Copyright
c
2013 SCITEPRESS (Science and Technology Publications, Lda.)
Source Source
node 1
node 2 node 3
node 4 node 5 node 1
node 2 node 3
node 4 node 5
Unicast Transmision
Multicast Transmision
Figure 1: Unicast and multicast modes of data forwarding.
open simulator. Developing a special purpose
simulator (Weijia et. al., 2005) is a custom solution
which is usually completed in one protocol. That
drawback forces us to use open simulation software.
For building up the multicast model in computer
engineering, there are various discrete simulators
such as, for example, OMNeT++ (Veselý et al,
2012), NS2 (Bartczak and Zwierzykowski, 2007);
(Tarik et al, 2001), and OPNET software (Wang et
al, 1999). In all cases it is possible to define static or
dynamic network models and to use graphical
interfaces. Finally it was decided that OMNeT++
open software would be used as well as it provides
an opportunity to verify the network design in a safe
and cheap environment. This open simulator
provides the ability to test a network with regard to a
failure hypothesis or to trace a flow of multicast
streams among sources and receivers. Discrete
events simulation like OMNeT++ provides a means
to check the behaviour of a network across a wide
range of test scenarios, which is not always feasible
in laboratory conditions. Contrary to this, obtaining
measures from a real multicast network architecture
is very expensive and complex as various
collaborating components are involved.
1.1 Previous Work
Stergiou et al. (Stergiou et al., 2007) present an
interesting multicasting architecture that operates
using a multicasting firewall over the Multicast
Control Protocol (MCP). They evaluate the
transition times of IGMPv2 reports on a multicast
control server. In 2008 the same group (Stergiou et
al, 2008) also analysed a broadband multicasting
system in an IPv4 environment using the IGMP and
MCOP protocols, where a Gigabit Ethernet was used
as the delivery network in the client’s segment. This
evaluation study provides measurements for the two
most significant performance metrics, the required
bandwidth and the round trip time of packets versus
the number of multicasting clients.
Chuang and Sirbu (Chuang and Sibru, 1998)
studied the cost advantages of a multicast protocol.
Their study took into consideration the link cost but
ignored the cost of routing table memory and the
CPU usage. In Chuang and Sirbu’s paper, the
normalized multicast tree cost is defined as the ratio
of the total length of the multicast distribution tree to
the average length of the relevant unicast path.
According to (Chuang and Sibru, 1998) this
normalized multicast tree cost is proportional to the
power,
k
N , where
N
denotes the number of
routing nodes that have subscribed to the multicast
group while
k
is a factor ranging between 0 and 1.
The previous sentence is called Chuang–Sirbu’s
Law. The weakness of this study is that it ignores the
cost of nodes as it is the cost of routing the table and
the bandwidth utilization. Moreover it is worth
noting that in (Chuang and Sibru, 1998) the pricing
model for the multicast mechanism was also
discussed. Chuang and Sirbu’s approach was the
first step in the study of the performance problems
of such protocols. Phillips et al. (Phillips et al.,
1999) tried to correct the abovementioned weakness
of Chuang–Sirbu’s Law, which is presented in
(Chuang and Sibru, 1998). In sequence to that, the
Phillips’s group analysed the above formula,
Chuang–Sirbu’s Law, for k-ary trees and for general
networks that were not k-ary. Another effort that
continued the research of Phillips et al. is the study
by Mieghem et al. (Mieghem et al, 2001), who
investigated the function of reachability, which
depicts the number of distinct sites that are equal to
a constant number. The number of nodes is a
specific constant number of hops starting from a
source that has been evaluated. The study of
Mieghem et al. shows exponential behaviour, as the
number of routings in the internet that can be
reached from a root grows exponentially with the
number of hops. Also the same group states that
multicasting provides efficient transmission of data
when the receivers are widely spread. Moreover,
Zaballos et al. (Zaballos and Segui, 2006) examined
a few routing properties of the IGP routing protocol
using an OPNET simulator. Finally, amazing and
enviable work has been presented in the field of
multicast protocols which are applied to wireless
networks (Gupta et al, 2010); (Bhasin et al, 2012).
All the aforementioned works were a considerable
asset which we held when we started this project.
InvestigatingtheOptimalPerformanceofMulticastCommunicationbySimulation
109
1.2 This Work
Here, a specific network configuration is obtained as
a testbed for our experiments. Multicast-type
streams running from load sources to multicast
clients via intermediate decreasing carrier links are
considered (see Section IV). Based on this network
scenario, a model was subsequently implemented
using OMNeT++ software. The specific model was
run for four multicast protocols which were selected
to complete the multicasting communication and
which were the main subjects for testing. Two of
them were dense-mode protocols (DVMRP, PIM
DM) while the other two were classified as sparse-
mode protocols (CBT, PIM SM). The major
motivation for this approach was to find a method of
distinguishing some counterparts of multicast
protocols in relation to receivers’ population
segments.
With a specific service scenario in a particular
network construction, the performance trends of four
natively multicast protocols were determined and
compared in a quantitative manner. OMNeT++ open
software was preferred as the platform for
simulation development because it is an open
simulator and does not have limitations in
supporting the tracing of data flows. Moreover, the
concept of the ANSA (Automated Network
Simulation and Analysis) extension module, which
expands the INET framework in the OMNeT++
simulation environment, was used and sufficiently
exploited here. This simulation comprises the called
INET framework in the OMNeT++ environment
with some external tools that are created for
implementing those specific multicast protocol
configurations. In addition to the above, the
OMNeT++ platform was chosen because it has the
capability to support hierarchically nested modules,
and has the ability to handle flexible module
parameters, although it is commonly accepted that
this product is a rather complicated software to
manage. Thus, one obvious contribution of this
paper is the simulation approach and the improved
router model incorporated in the OMNeT++ model.
Contrary to the majority of publications, which
present separate values of some performance metrics
without correlating the various results among them,
in this paper a general performance indicator is
introduced that can be extended to include a more
positive or negative performance factors, evaluating
any protocol’s strengths and weaknesses
respectively. So, the introduction and the usage of a
unique overall factor is another significant
contribution in the field of evaluation of protocols.
Finally, the rationale behind this performance
methodology is to provide a tool for estimating the
overall performance of protocols in an inexpensive,
quick, accurate, and transparent way for a given
number of receivers.
The remainder of this paper is organized as
follows: in Section 2 we introduce the performance
metrics as performance criteria that are related to
multicast protocols and that are taken into
consideration in our study, and in Section 3 the
simulation testbed used is presented. Section 4
presents part of the results of our performance
approach, which has been conducted through
simulation experiments, and finally Section 5
concludes with basic remarks.
2 METRICS DEFINITIONS
In order to consider the quality of service (QoS) at
the receiver side, we need to define some clear
parameters which will show the QoS in a
quantitative manner. Several parameters can indicate
the QoS at the end-user view such as the latency of
packets, the jitter, the packet loss, the availability of
connecting a node in a multicast tree, the time period
of joining, and so on.
Bearing in mind the need to have a reliable
approach, we chose to define the following
parameters which can show the QoS at the end-user
side sufficiently succinctly.
2.1 Relative Packet Latency
from Source to end User (
r
L )
The relative packet latency from source to end user
(
r
L
) can be defined as the ratio of latency which
packets presents when they travel from sender to
receiver in unicast mode (
u
L
) to the corresponding
packet latency when they travel in multicast mode
(
m
L
). Thus, the relative latency (
r
L
) from source to
end user is given by the following expression:
mur
LLL /
(1)
The packet latency transmitted in multicast mode is
basically dependent on the number of sources. If the
number of sources in a system proliferates then the
time consumed increases on each Randezvous Point
(RP) router considerably, which raises the total
packet latency from sources to end users.
SIMULTECH2013-3rdInternationalConferenceonSimulationandModelingMethodologies,Technologiesand
Applications
110
2.2 Relative Jitter Parameter (
r
J )
The relative jitter parameter (
r
J
) is defined as the
ratio of the average jitter of packets which are
transmitted by multicast mode transmissions from
source to receiver (
m
J
) to the corresponding
average jitter of the packets transmitted in unicast
mode (
u
J
). Hence,
umr
JJJ /
(2)
Jitter variation is a parameter that basically
determines whether a user who handles multimedia
will have a pleasant experience or not.
It is clear that the above two defined metrics
have a relationship and that they are directly
involved with the QoS at the listeners’ side.
Contrary to this, network utilization is an alternative
aspect of the performance problem. The efficiency
or the gain of multicasting in terms of network
resource consumption can be compared to the
corresponding unicast one. Similarly to the above
(the viewpoint of the performance of the end-user),
various parameters need to be defined in order to
obtain indications of the effectiveness of the network
when it manages and handles multicast traffic.
Subsequently, we chose to define two major
performance parameters through which the capacity
of the network is evaluated when it deals with
multicast traffic.
2.3 Relative Usage of Links )(
r
U
Considering one-to-many communication, in which
a source distributes messages (packets) to
m
different, uniformly distributed destinations along
the shortest path, in the unicast mode of transition,
these messages are sent
m times from the source to
each destination. One of the main properties of
multicasting is that it can economize on the number
of links traversed. Mieghem and Janic present a
study (Mieghem and Janic, 2002) which shows how
the number of links in a multicast tree changes as the
number of multicast users in the group changes.
According to (Mieghem and Janic, 2002), the
stability of a tree tends to a Poisson distribution for a
large number of receivers. It is obvious that the
greater the number of links that are used in a
multicast communication, the greater the utilization
of links and usage of resources of the network.
Hence a performance parameter is defined in order
to show the usage of the network. This metric is
called the relative usage of links and is defined as
the ratio of the number of multicast hops to the
number of unicast hops of a packet stream.
)
/()(
hopsunicastof
numberhopsmulticastofnumberU
r
(3)
where
10
r
U
.
It is apparent here that when the number of
members of a multicast group increases, the number
of hops also pullulates but this increment has a much
lower rate than that in the corresponding unicast-
mode communication.
2.4 General Performance Factor (GP)
From our general experience of multi-criteria system
performance it became apparent that the
performance metrics can be divided into two major
categories: those that cause the system (here the
performance of a protocol) to behave better when
they are maximized and those that cause the
performance of the system to improve when their
values are minimized.
We depict the first group of performance metrics
as
},...,,{
max,max,2max,1max
aaaa
and the other
group as },...,,{
min,min,2min,1min
bbbb
, where
and
are the number of performance factors to
be maximized and minimized respectively.
The optimal solution is to have high values of
the first group of performance metrics and low
values of the other group. Thus, it is interesting to
carry out a general evaluation using exclusively one
factor. We wish to have only one factor which will
reveal the overall performance. So, this required
overall performance factor is defined by relying on
the correlation of the two individual groups of
metrics. Nevertheless due to the fact that each metric
has different units and ranges, it is necessary to
normalize them to obtain a common reference value
domain. We call this factor the general performance
(GP) factor, which is formally defined as:


1
2
max,
1
2
min,
)
1
()(
i
j
j
i
b
aGP
(4)
Supposing we now have three or more protocols and
their general performance factors,
1
GP ,
2
GP ,
3
GP ,
and so on, respectively. If we have the following
inequality regarding the general performance,
....
321
GPGPGP , then we can say that System
1 is better in comparison to System 2, System 2 is
more powerful than System 3, and so on. It is clear
that the evaluation and comparison between
InvestigatingtheOptimalPerformanceofMulticastCommunicationbySimulation
111
protocols is transformed into the evaluation of a
unique factor and the ideal performance of the
system occurs when the value of the GP indicator
tends towards zero. It is noteworthy that the
GP
factor represents the length of the corresponding
vector in two-dimensional space and a smaller value
indicates a better performance of the examined
system. Following the above, another interesting
point when we study multi-criteria decision-making
problems is that the importance of each criterion is a
design problem and depends on the general
environment and special considerations. When we
want to set a special weight on each individual
metric, we have the ability to assign a weight to each
performance coefficient and the above formula can
be replaced by:


11
1
2
max,
1
2
min,
)
1
()(
),(
j
j
i
i
i
j
j
j
ii
ji
ww
b
waw
wwGP
(5)
where
i
w
and
j
w
are the corresponding weights of
the minimized and maximized performance metrics.
In all cases the reference value domain of the
general performance factor ranges from 0 to 1. The
assumption that
0
max,
j
b
must be satisfied in any
case.
In this work we limit our study to the use of three
performance metrics, knowing that it is feasible to
include additional performance factors that could be
chosen to evaluate the performance of the multicast
protocol with better accuracy. Additional
performance parameters that can be considered
include the multicast packet loss, the number of
states that are utilized by each intermediate router in
order to achieve the multicasting delivery, and so on.
3 EXPERIMENTAL TESTBED
AND SIMULATIONS
A specific networking schema is selected for use as
a basic testbed for checking multicast
communication (See Figure 2). This network
consists of three distinct zones related to the values
of speed and bandwidth. The carrier links that
connect these distinct areas is considered to have a
corresponding scalar structure, as they serve the load
from the sources to the multicast clients. According
to our scenario, three multicast senders create
multicast traffic at a rate of 256 kbps, obtained as a
root of traffic generation, and those servers send
load to the core network. The services that are
employed on the multicast servers are Web and FTP
services which are working over the TCP transport
layer, and also the deployed TFTP service which
operates over the UDP transport layer. The multicast
data streams traverse the network’s tree. specifically
the streams pass from the network backbone (WAN)
to an intermediate MAN area, then arrive at the local
areas and finally terminate at the end traffic
consumers. In the WAN area, 5 basic routers are
obtained which are connected together via E3 type
connections. Those WAN area routers are connected
via E2 carrier links with intermediate second level
routers. Each second level routing node (belonging
to the MAN area) is able to support a maximum of
10 different MAN type networks. Also, continuing
the tree hierarchy regarding the bandwidth and data
speed, the second level routing devices are
connected via E1 type links with the third level
routers. Each third level routing device directly
supports two Ethernet type LANs, which are
connected via 10 Mbps links. Finally, three different
nodes are connected on each LAN, which are the
final recipient of the multicast load. The end data
consumers of multicast traffic are selected randomly
in order to be spread throughout the network
architecture under study. Special care was taken to
ensure that the ultimate receivers do not belong to
the same group.
….
….
….
….
….
….
E3 link
E2 link
E1 link
WAN Area
(basic core of multicast traffic)
MAN Area
(intermediate level of routing)
LAN Area
(consumption of multicast traffic)
Multicast Senders
….
….
….
End receivers of data
….
(FTP, TFTP and Web Service)
Ethernet LANs
Figure 2: The under study networking schema.
The rationale behind our approach was to firstly
create a simulation model of multicast routing and
data delivery architecture including the visualisation
of the distribution tree and end-user behaviour. At a
secondary level we wanted to investigate optional
scenarios, accompanied by thorough analysis. To
accomplish this task, OMNeT++ version 4.2 and an
INET 20111118 framework is used to model the
network running the multicast traffic. We use the
INET framework modules as the state of art of
SIMULTECH2013-3rdInternationalConferenceonSimulationandModelingMethodologies,Technologiesand
Applications
112
simulation modelling. A module in the OMNeT++
environment is a list of commands enforcing a
required behaviour, which represents a counterpart
device in the network. Specifically, the model that is
created use the
inet.networklayer.ipv4:
INET’s module. At the start of modelling the
IGMPv2 protocol (RFC 2236) is used (
igmp:
INET’s module), since IGMP is a standard protocol
that has the ability to manage membership in
multicast groups. According to the IGMPv2
specification, this protocol uses three distinct types
of message: Membership Query, Membership Report
and Leave Group. As a consequence, the INET
modules
networkLayer, routingTable and
interfaceTable are used to build up and
manage a third level of communication in the
system.
Regarding the transport layer, corresponding
INET modules for TCP and UTP protocols were
used, as well as subsequently FTP, HTTP and TFTP
services for setting up corresponding sessions.
Unfortunately the implementation of the TCP
protocol in the OMNeT++ environment has not been
state of the art. Some important features of modern
TCP implementation, particularly Selective
Acknowledgements (SACK) and complete Flow
Control, were missing (Reschka et al, 2010).
However this simulation used the existing INET
module for the TCP sessions. Thus the modules
TCPConnection, addSockPair, etc. are
correctly exploited by this model. Furthermore,
among the other modules included in the INET
framework, the
OSPFRouter module is used for
typical router implementation in our system.
In the first phase of development we used all the
above described modules, which already exist in the
INET framework of OMNeT++. However, the freely
available OMNeT++ simulator is not fully-
compliant with the specific protocols under
investigation. That forces us to implement modules
for the following multicast routing protocols: CBT,
PIM SM, PIM DM and DVMRP. In the second
phase, special purpose modules are created which
include the required multicast features and which are
ultimately bound to the ANSA type router. The
additional modules are formed in XML language.
Simulation model for OMNeT++ was revealed
immediately after formed the XML configuration of
real devices and links. Our simulation tool was able
to setup all devices and a variety of links and initial
information, finally running on the network scenario
under study.
The
StandardHost module is used for
making simple host (or receiver) representations.
Moreover each end user is assigned a queue which is
capable of collecting and keeping 50 packets at
most. That value of the buffer size is considered to
be very common in such experiments. Originally 5
receivers were set up in our network system, and
step by step this reached a value of 30 receivers.
Those receivers were spread randomly throughout
the multicast tree.
We ran the simulation many times, again setting
all the parameters to separately obtain results in the
form of mean values for each multicast protocol.
Metrics such as packet’s latency of packets,
minimum variation of delay, and number of
multicast hops employed for each protocol, were
collected. However, the analytical structure of the
developed model, as well as the results presented
here, is limited by the boundaries of a typical
document. The obtained statistics from the simulator
were raw numbers that we subsequently handle in
order to produce normalised values which are
presented in the next section. Extensive simulations
to validate our results have also been undertaken.
4 PERFORMANCE RESULTS
4.1 Relative Packet Latency (
r
L )
Figure 3 illustrates the relative packet latency (
r
L
)
that appears when the packets travel from the
sources to the end consumers versus the number of
end multicast traffic consumers for each studied
protocol. It is obvious that the sharing mode
transmission needs more time cycles in comparison
to the unicast mode because the transmitted packets
have to be first multiplied and then transmitted.
According to the definition of relative packet
latency (Formula (2)) and the results depicted in
Figure 2, the relative packet latency of multicast-
mode traffic in the CBT, DVMRP, and PIM DM
protocols is
6.0
r
L . Thus the packet latency of
multicast traffic is:
666.16.0//
uurum
LLLLL .
That means that the maximum possible value of
multicast packet latency may be equal to 160% of
the corresponding unicast mode packet latency.
On the other hand, the relevant packet latency of
the PIM SM protocol (the second adjacent bar of
each four-tuple set of bars) reaches at most
9.0
r
L . Hence, the packet latency of multicast
traffic is
111.19.0//
uurum
LLLLL
.
InvestigatingtheOptimalPerformanceofMulticastCommunicationbySimulation
113
That means that the maximum packet latency of the
data when the PIM SM protocol is applied never
exceeds 111% of that of the corresponding unicast-
mode traffic.
0.4
0.5
0.6
0.7
0.8
0.9
1
5 1015202530
Number of End Users
Relative Latency
CBT PIM-SM DVMRP PIM-DM
Figure 3: Relative sources-to-end packet latency versus
number of end users.
Hence, from all of the above it is concluded that the
packet latency of the CBT protocol (the first
adjacent bar of each four-tuple set of bars) presents
the maximum deviation of the corresponding
unicast-type packet latency (and thus the worst case)
while the packet latency of the PIM SM protocol
shows the minimum deviation of the corresponding
unicast mode (the best case in terms of packet
latency), and Fig. 3 confirms and illustrates this
result quantitatively. In all cases of the multicast
protocols it can be noticed that when the number of
receivers become greater than 10, the traffic delay
from source to client tends to present stability.
Moreover it is worth noting that the gain in packet
latency shown by the PIM SM protocol over the
dense-type of protocols studied here can be
considered as marginal, especially when the number
of listeners becomes large and the differences are
negligible.
4.2 Relative Usage of Links )(
r
U
Figure 4 plots the relative usage of links (in relation
to unicast mode) when multicast traffic is employed
for various numbers of receivers and for each
examined multicast protocol for the given network
and service selection. From this graph it is obvious
that the two dense-type multicast protocols (PIM
DM and DVMRP; the first two successive bars in
each four-tuple bar group in Fig. 5) employ a
considerable number of links especially when the
number of receivers is small (less than 25). For
example, for an extremely small multicast client
population equal to 5 the quantity of links that are
engaged by dense-type protocols is approximately
25% over of the corresponding links’ quantity that
are employed by the sparse-type protocols.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
5 1015202530
Number of End Users
Relative usage of links
PIM-DM DVMRP CBT PIM-SM
Figure 4: Relative usages of links of protocols versus
number of end users.
On the other hand, when the number of listeners
exceeds this threshold (25 listeners) all of the
multicast protocols have similar behaviours
regarding the metric for the usage of links as there is
only a slight discrepancy between their employed
links. Also, as the number of receivers increases and
the service becomes more fully multicast, the
exploitation of the network is improved
considerable.
Especially, observing Fig. 4, it can be observed
that when the number of listeners is more than 25
the gain in network sources exploitation tends to
exceed 80% of the corresponding unicast service.
That clearly reveals the main advantage of the
multicast way of transferring data.
4.3 General Performance (GP)
Indicator
For calculation of the general performance indicator
three metrics were taken into consideration: the
packet’s latency and the jitter variation which were
considered with equivalent weights, equal to 40%
(w=0.4) and the usage of network’s links with its
weight equal to 20% (w=0.2).
The first pair of metrics is related to the
performance of the end user aspect, while the third
correlates to capacity of the network. Hence this
chosen pattern of weights is useful, especially when
there is more interest on the QoS on the receivers’
site and less interest on achieving a high network
capacity. Moreover, is noteworthy that the metric of
packet’s latency is a maximised type factor due to its
definition (see formula (2)) while the other two
metrics are minimised type factors in our
SIMULTECH2013-3rdInternationalConferenceonSimulationandModelingMethodologies,Technologiesand
Applications
114
performance scenario. Fig. 5 presents the general
normalised performance factors as the number of
receivers is increased (values domain: 5–30). In the
same plot, the general behaviour of multicast
streams according to the listeners’ view appears to
have a small variation (stability) for each protocol.
Weights: w
Lr
=w
jr
=0.4 and w
Ur
=0.2
0.3
0.35
0.4
0.45
0.5
0.55
0.6
0.65
0.7
0.75
0.8
5 1015202530
Number of End Users
General Prformance Indicator
CBT PIM-SM DVMRP PIM-DM
Figure 5: The general normalized performance indicator
versus number of end users.
Furthermore, the graph in Fig. 5 clearly shows that
the best overall performance is presented when the
PIM SM is applied; resulting in a lower value of the
GP indicator of approx. 0.45. In contrast to this, the
other sparse type protocol (CBT) presents the worst
overall performance which is about 28–30% worse
in comparison with the PIM SM case protocol for
the given network configuration, range of listeners,
obtained metrics and specific pattern of weights
between them.
It is worthwhile noting that the dense type
protocols endow an overall yield that is between the
performance of the two sparse protocols (approx.
22% worse in comparison to the PIM SM
performance and 10% better than the CBT protocol).
Also, it must not be ignored that the tension that
have all the protocols of slighting deterioration
(slight rise in the value of the GP indicator) when
the number of listeners exceeds 25 and is increasing.
Finally it is worth mentioning, and to be
reminded again, of the variety of network schemas
for the chosen metrics and obtained weights that can
be studied by applying this methodology for
detecting where the performance edge of each
multicast performance protocol exists.
5 CONCLUSIONS
Building up the simulations and making the
inevitable validation for each protocol was a difficult
task and a time-consuming process; also, the open
software OMNeT++ does not have a full range of
modules ready and does not provide fully compliant
tools for the protocols under study. Finally, to sum
up the above, in this paper two sparse-type multicast
protocols, CBT and PIM SM, and two dense-type
multicast protocols, DVMRP and PIM DM, were
investigated with regard to their performance. The
performance approach using ONNeT++ software
was one of the key contributions and the
introduction of a unique performance indicator was
the other. The simulation experiments focused on
only a few metrics out of many that can be explored
and studied. Nevertheless, we believe that the rest of
the quantitative properties can be extracted without
making considerable changes to the current
simulation approach. It is obvious that, so far, the
research on multicast protocols is far from
exhaustive.
REFERENCES
Bartczak T. and Zwierzykowski P., 2007. "Validation of
PIM DM implementation in the NS2 simulator", ISAT,
pp. 117-127
Bhasin Swati, Gupta Ankur, Mehta Puneet, 2012.
"Comparison of AODV, OLSR and ZRP Protocols in
Mobile Ad-hoc Network on the basis of Jitter",
International Journal of Applied Engineering
Research, ISSN 0973-4562 Vol.7 No.11
Chuang John, Sibru Marvin, July 1998. "Pricing Multicast
Communication: A Cost-Based Approach", Internet
Society INET 1998 Conference, Geneva, Switzerland
Gupta Anuj K., Sadawarti Harsh, Verma Anil K., April
2010. "Performance analysis of AODV, DSR &
TORA Routing Protocols", in IACSIT International
Journal of Engineering and Technology, Vol.2, No.2,
ISSN: 1793-8236
Mieghem P. Van, Janic M., 2002. "Stability of a Multicast
Tree", in IEEE INFOCOM 2002, pp. 1099-1108
Mieghem Piet Van, Hooghiemstra Gerard, Hofstad,
Remco van der, 2001. "On the Efficiency of
Multicast", in IEEE/ACM Transactions on
Networking, vol. 9, No. 6, pp. 719-732
Phillips G., Shenker S., Tangmurarunkit H., 1999.
"Scaling of Multicast trees: Comments on the Chuang
Sirbu Scaling law", in ACM SIGCOMM 1999,
Harvard, MA pp. 41-51
Reschka Thomas, Dreibholz Thomas, Becke
Martin, Pulinthanath Jobin, Rathgeb Erwin P., 2010.
"Enhancement of the TCP Module in the
OMNeT++/INET Framework", In: Proceedings of the
3rd ACM/ICST International Workshop on
OMNeT++, ISBN: 978-963-9799-87-5
Stergiou E. , Meletiou G., Vasiliadis D., Rizos G.,
and Margariti S., November 2008. "Evaluation Study
of a Broadband Multicasting Service over a Gigabit
Ethernet Delivery Network", AIP Conf. Proc. Volume
1060, p.p. 392-396
InvestigatingtheOptimalPerformanceofMulticastCommunicationbySimulation
115
Stergiou E., Meletiou G., Vasiliadis D., Rizos G., and S.
Margariti S., December 2007. "Evaluating the
Multicast Control Protocol on a Multicasting Network
Architecture", AIP Conf. Proc. Volume 963, p.p. 943-
946
Tarik Cicic, Stein Gjessing, Oivind Kure, 2001. “
Performance Evaluation of PIM SM Recovery”.
Proceedings of International Conference on
Networking ICN'01, Colmar, France, July 2001
Veselý Vladimír, Matoušek Petr, Švéda Miroslav, 2012.
"Multicast Simulation and Modeling in OMNeT++",
In: Proceedings of the IEEE 5th International ICST
Conference on Simulation Tools and Techniques,
Desenzano del Garda, IT, ICST, p. 298-301, ISBN
978-1-936968-47-3
Wang X., Yu C., Schulzrinne H. and Stirpe P., 1999. "IP
Multicast Simulation in OPNET", CiteSeerx
10.1.1.18.7365
Weijia Jia, Wanqing Tu, Wei Zhao and Gaochao Xu,
2005. "Multi-shared-trees based multicast routing
control protocol using anycast selectio", The
International Journal of Parallel, Emergent and
Distributed Systems, Vol. 20, No. 1, pg. 69-84, March
2005
Zaballos A. and Segui C., 2006. "Analysis and simulation
of IGP routing protocols", Technical report,
University of Ramon Llull, Barchelona
SIMULTECH2013-3rdInternationalConferenceonSimulationandModelingMethodologies,Technologiesand
Applications
116