1720 1724

ISSN: 2278 – 1323
International Journal of Advanced Research in Computer Engineering & Technology (IJARCET)
Volume 2, No 5, May 2013
1720
www.ijarcet.org

Abstract
In this is an era of drastic evolution
Telecommunications, packet data communication has gained
paramount attention due to increased internet multimedia
usage. One of the challenges faced by telecom companies is to
provide robust and powerful network components that are
capable to handle the tremendous load of traffic and data on it.
Companies evaluate their products performance before
releasing them to the market by applying a large amount of
generated traffic in order to measure their capability under
traffic load; powerful solutions are hence needed for generating
traffic and modeling various telecom protocols. The
performance is also need to be evaluated for other scenarios that
exists in the real network. The scenarios include capacity tests,
busy hour call loads, rainy day scenarios, and different types of
handoffs. Hence there is a requirement of a simulator tool which
evaluates the product’s performance. Packet Load Generator
(PLG) is a tool developed to evaluate Network Controller.
Index Terms— Network Controller, RF path loss,
Semi- marcovian Traffic, Traffic Models.
I. INTRODUCTION
The ever increasing need for the multimedia communication
over telecom network is guiding the telecom industry to better
the existing features and data rates. Thus developed new or
enhanced technologies cannot be introduced directly to the
public. The expected or the theoretical figures from the
network elements have to be tested rigorously for their
capacity before it is deployed. The vendors cannot test and
verify the figures and functionalities with the existing real
world network. So there is a need for the network simulators
that imitate the actual traffic scenarios and hence validate the
newly developed network components.
Network components being complex entities challenge
building such simulator tools. The telecom network
components are majorly IP packet networks and thus such
simulation tool can be equally employed for all kinds of IP
networks for capacity and functionality testing or the existing
packet generator tool can be employed with new
functionalities embedded.
The tool has to consider the intermediate protocols and
interfaces lying in connecting such network components. It
have to include the air interfaces between switching center to
Access Terminals, the Backhaul interfaces and protocols with
the main network element and the IP link with the PDSN
(Packet Data Serving Node) for internet connections. Major
parameters of the network components determined will be the
delays and losses of the packets transmitted.
Major challenge for the tool design is to create a virtual
environment almost similar to the actual 3G network.
Functionalities like Idle, Soft, Softer, Virtual Soft and Virtual
Softer Handoff's are to be simulated between the tool
simulated cells with the handoff's specified between which
particular cells or sectors clearly. Measurements like Busy
Hour Call Attempts( ), Connection Disconnects( connection
and disconnection requests per hour to a particular cell ) are to
be simulated which are generally the mathematical
considerations made from the times of Public telephone lines
capacity analysis. Traffic is to be measured on both the sides
Reverse Link (RL) and Forward Link (FL) and are measured
in kilobitspersecond(kbps) and displayed to the user so that he
can notice the rise and drops in the traffic with the required
functionality and required number of cells simulated.
Many mathematical models are suggested for the figures
used in generating the random numbers used in any of the
steps here in the simulation. Generally the number of requests
are said to be following Poisson distribution and some of the
parameters like traffic are said to follow Erlang's distribution
curve. Also Gaussian distribution functions holds similar to
the Register and Deregistering the calls.
Different people have put forth their ideas of different
aspects of this tool but a technical paper that enumerates all
the conceptual requisites of such a tool is hardly available.
This was our motive and this paper describes the key
requirements of building an efficient simulation tool that
exactly imitates the complex telecom network components
and hence aids for capacity and performance testing. The
traffic generation procedure, proposed in this paper, is highly
modular. The proposed traffic generator architecture must be
flexible and capable of producing traffic streams compliant
with fully general high-level traffic models, either stochastic
or deterministic. The software that drives the traffic
generation process should employ several kinds of
optimization, for removing limitations due to hardware speed
and memory constraints of the host machine, and for ensuring
that traffic streams corresponding to models with a wide range
of parameters can be generated in an efficient and reliable
manner. This tool can be used in conjunction with the traffic
generator to support measurement-based performance
evaluation in experimental test beds.
We give an example of the state of art of such a tool for the
CDMA EVDO case and same techniques or architecture can
be followed for UMTS networks with slight protocol and
interface changes.
This paper first gives an overview of a EVDO network
under next heading. The next heading deals with theory
behind modeling the internet traffic for data used in data
communication, then the basic architecture of the state space
model is described, later we discuss about the importance of
physical layer considerations in UMTS network and at the
end, we give a over view of how the tool architecture.
Packet Load Generator for Telecom Networks
Niranjan U, Ramacharyulu P Govardhanam, K Vijaya

ISSN: 2278 – 1323
1721
www.ijarcet.org
II. INTERNET TRAFFIC THEORY
Observations of traffic on network links typically reveal
intensity levels (in bits/sec) averaged over periods of 5 to 10
minutes which are relatively predictable from day to day.
Systematic intensity variations occur within the day reflecting
user activity. It is possible to detect a busy period (usually in
the afternoon between 2 and 5 pm) during which the traffic
intensity is roughly constant. This constancy suggests that
Internet traffic, like telephone traffic, can be modeled as a
stationary stochastic process here statistical variations occur
about an underlying constant intensity. Busy period
performance is then estimated by the long term average
behavior derived for the stationary process.[2]
But still there is some ambiguity in the literature that
whether the statistical theory of public telephone network can
be applied as it is to the TCP or UDP supported internet traffic
systems. The precise characteristics of this stationary process
depend on the composition of Internet traffic. Currently, some
90 to 95% of Internet packets use TCP and correspond to the
transfer of digital documents of one form or another (Web
pages, data files, MP3 tracks …). The congestion avoidance
algorithms of TCP cause throughput to vary elastically in
reaction to random changes in the set of transfers in progress.
A small but growing proportion of traffic relates to inelastic
streaming audio and video transmission for both interactive
and playback applications.
A. Traffic Objects
The traffic process can be described in terms of the
characteristics of a number of objects, including packets,
bursts, flows, sessions and connections, depending on the
time scale of relevant statistical variations. The preferred
choice for modeling purposes depends on the object to which
traffic controls are applied. Conversely, in designing traffic
controls it is necessary to bear in mind the facility of
characterizing the implied traffic object. This consideration is
particularly important in the design of the future Internet
where only the datagram and the broad destination-based
aggregate used in routing are currently recognized. Traffic
characterization proves most convenient at an intermediate
flow level.
A flow is defined for present purposes as the unidirectional
succession of packets relating to one instance of an
application (sometimes referred to as a microflow). For
practical purposes, the packets belonging to a given flow have
the same identifier (e.g., source and destination addresses and
port numbers) and occur with a maximum separation of a few
seconds. Flows are frequently emitted successively and in
parallel in what are loosely termed ―sessions.‖ A session
corresponds to a continuous period of activity during which a
user generates a set of elastic or streaming flows.
Some network service models define the notion of
―connection‖ and control resource allocation by means of
explicit signaling exchanges. The connection might be set up
for a particular flow or used over a long period for an
aggregation of flows between given network end points. A
significant difficulty resides in defining parsimonious traffic
descriptors representing the impact the connection is likely to
have on network performance. Signaling overhead may also
prove excessive, particularly when each connection relates to
an individual flow.
B. Arrival Processes And Service Requirements
It is well known that the arrival process of IP packets can
exhibit extreme rate variations at multiple time scales. First
reports of this behavior more than ten years ago have given
rise to a large amount of research aiming to explain the
so-called self similarity phenomenon and to predict its impact
on network. The main reason for rate fluctuations at time
scales greater than a few hundred milliseconds turns out to be
extreme variability in the size of the flows making up the
observed packet process. Yet more extreme variability (so
called multi-fractal behavior) occurs at smaller time scales
due to the burstiness induced by TCP. It proves much simpler
to describe traffic in terms of flows.
The arrival process of flows in a backbone link typically
results from the superposition of a large number of
independent sessions. Observations confirm the predictable
property that session arrivals can be assimilated to a Poisson
process. This means simply that the probability of a new
arrival in a short interval of length dt is equal to λdt, where λ is
the arrival intensity, and is independent of all past activity. A
Poisson process results naturally when traffic is due to the
independent activity of a very large population of users, each
individually having a very small intensity.
As a first approximation, it is not unreasonable to assume
that individual flows also occur as a Poisson process. To
ignore the correlation between flow arrivals within the same
session is not necessarily significant when the number of
sessions is large. It is also true that results derived under the
simple Poisson assumption are also often true under more
general assumptions.
III. TRAFFIC GENERATION THROUGH HIGH-LEVEL
MODELS
A traffic model is one that have different states of the
simulation tool and the advancements of the states are carried
on according to random process that suits the particular
physical process. Consider an example of Registering calls
and states of tool is the number of calls registered per minute
and this is modeled using Gaussian Process.
In [3], they discuss about the model for the system to
change its state from one to another based on the traffic
models and the present event.
A. The Structure of Semi-Markov Traffic Models
The traffic generator adopts a high level traffic specification
in terms of (time invariant) semi- Markovian models.
According to this framework, each traffic model is
characterized by a set of N states. During a sojourn at some
state i, a constant, state-dependent data rate ri is maintained.
(The modeling framework assumes a ―fluid-flow‖ approach,
according to which more complex cell level rate fluctuations
within each state can be ignored. This approach is entirely
adequate for representing the important burst level
phenomena, while at the same time avoiding excessive
low-level complexity.) Transitions among states occur
according to a discrete irreducible Markov chain with a
transition probability matrix P = [pij ]. The sojourns at a state
i are independent and identically distributed random
variables, of a general probability distribution function (PDF)
Fi(·). The well-known MMRP traffic models, where all

ISSN: 2278 – 1323
1722
www.ijarcet.org
sojourn times are exponentially distributed, are an important
special case of semi-Markovian models. [3]
B. Top Level Architecture
From a high level view, traffic generation corresponding to
a semi-Markovian model consists of a sequence of events
(high level traffic events—HLTEs). Each HLTE signifies the
emission of cells at a constant rate ri, depending on the
currently occupied state i, for a duration chosen through
sampling the respective PDF governing the sojourns at this
state.
Fig.1. Basic functional diagram for generating
semi-Markovian traffic.
With the notion of an HLTE at hand, the task of semi-
Markovian traffic generation requires an algorithm to decide
on the next state to be occupied by the model (next state
decision algorithm—NSDA) and an algorithm to sample the
sojourn time that will be spent at the chosen state (sojourn
time decision algorithm—STDA). Furthermore, the nature of
ATM and the structure of the traffic generation hardware
suggest the need for two more mechanisms: One to convert
data-rates, and the durations for which these rates are
sustained, to ETEPs, i.e., (ID,NC) pairs (rate to pairs
conversion algorithm—RPCA) and another to encapsulate the
high-level traffic event sequence into units of size HMS (high
level traffic events encapsulation algorithm—TEEA).The
relation between these building blocks is depicted in Fig. 3.
C. The Next State Decision Algorithm (NSDA)
Given as input an Integer value expressing the state
visited last time, the NSDA determines the state to be visited
next. For this purpose, NSDA makes use of the transition
probabilities matrix P = [pij ], as follows: During an
initialization step, performed once before the actual traffic
generation process, NSDA computes the aggregate transition
probabilities matrix A = [aij ], where aij = _j k=0 pik, for all i,
j = 0, ..., N − 1. During normal operation, given the state i
currently occupied, the next state is decided upon by drawing
a random number s in the interval [0, 1) and determining the
smallest index j, such that the quantity aij exceeds s. Since the
transition probability matrix P does not change over time, the
partial probability sums are also invariant and having them
ready into the elements of A reduces the time needed to
determine the next state.
D. The Sojourn Time Decision Algorithm (STDA)
STDA consists of two modules: STDA 1, an initialization step
performed once in startup, and STDA 2, which is executed
repeatedly during the generation process. The preprocessing
stage of STDA 1 undertakes the conversion of the sojourn
PDFs into a form that allows for the efficient selection of
actual sojourn samples by STDA 2 during on-line operation.
Introduction of the preprocessing makes sense because the
PDFs do not change over time. More specifically, STDA 1
reads data that determine the PDFs governing the sojourn
times at all states of the model. Then, for each state the
algorithm computes values of the respective PDF2 and creates
an array of size M that contains inverse distribution function
values (i.e., sojourn time values), which correspond to
equally spaced PDF values in the range from 0 to (M − 1)/M.
The dimension M (common to the arrays of all states) is
chosen large enough to ensure a sufficiently dense
representation of the PDF, while also avoiding excessive
storage requirements. Once the arrays have been produced,
STDA 1 exercises the RPCA algorithm upon each element of
each array, taking as input the sojourn time of this element and
the data rate associated with the corresponding state, and
transforms this input into a collection of ETEPs. This is done
in a way explained in the next subsection and further detailed
in Appendix-A. During the cell generation process, the STDA
2 performs (in a highly efficient manner) a much simpler task,
namely that of drawing a random index into the array
corresponding to the state picked by the NSDA. The low-level
traffic event information in the chosen element is then fed into
TEEA, which has the responsibility of generating HMS
ETEPs for download to the hardware. TEEA is described in
Section III-F.
IV. PHYSICAL LAYER CONSIDERATIONS:
The advance communication theory proves that the RF
multipath loss is less in the CDMA links compared to the
TDM or FDM links due to its wide band nature, Hence
physical layer consideration would not matter much in
CDMA network simulation but is a point to be considered for
UMTS networks.[4]
Most of the available commercial traffic generators utilize
the network resources on the fixed side of the cellular
network. This is a big disadvantage because such generators
produce traffic without involving the overload of the air
interface, which is the most important interface (Uu) at the
UMTS network. In this way, we have not a complete and
realistic view of the traffic generation, because it is well
known that several congestion problems on 3G networks
occur due to air interface limitations. The use of air interface
resources, from a 3G traffic generator is very important,
because CDMA networks like UMTS, have several crucial
mechanisms that are implemented on RF layer, such as the
open loop power control, admission control and fast close
loop power control at both directions uplink/downlink. On the
other hand, the most common use of a traffic generator is the
ability to produce enormous traffic, in order to produce
overload and congestion scenarios, for evaluating the
response of the network in such situations.
In an UTRAN network the traffic generator, which works
from the fixed network side will produce faulty and optimistic

ISSN: 2278 – 1323
1723
www.ijarcet.org
results comparing to traffic generator, which implements an
―end to end‖ traffic generation. This means that on an UMTS
traffic generator it is important to emulate not only the
produced Erlang at the specific traffic scenario with
determined services, but also to provide information about the
response of the network to random and complex RF pathloss
environment with multipath fadings etc.
Also, the majority of traffic generators need specific
equipment and hardware, like terminals and control consoles,
which is a minus point due to incompatibility. This
proprietary architecture sometimes provides lack of
scalability and of course increases the overall cost of the
platform. The available traffic generators could be controlled
only locally, which is not very practical. Also there is no
capability to send the results and statistics remotely to a
central database. [4]
In [5], a separate simulator tool design is discussed to
imitate the RF losses in UMTS network.
Rayleigh fading generation
Generally in wireless transmission scenarios where a receiver
is in motion relative to a transmitter with no line-of-sight path
between their antennas, Rayleigh fading is often a good
approximation of realistic channel conditions. The term
Rayleigh fading channel refers to a multiplicative distortion
h(t) of the transmitted signal s(t), as in y(t) = h(t) . s(t) + n(t)
where, y(t) is the received waveform and n(t) is the noise. The
TGP is going to implement the fading generation, by
performing a sampled version of the channel waveform h(t) in
a statistically accurate and computationally efficient fashion.
Thus, with the use of specific microwave filters and polyphase
interpolators, we could produce such fadings on the Uu
interface. But in our prototype we could emulate these
fadings, with very fast alterations on the attenuation level on
the programmable attenuators.
The channel waveform h(t) is modeled as a
wide-sensestationary complex Gaussian process with zero
mean, which makes the marginal distributions of the phase
and amplitude at any given time uniform and Rayleigh,
respectively, hence the term Rayleigh fading. The
autocorrelation properties of the random process h(t) are
governed by the Doppler frequency fD, as in[ ]:
where Jo({) is the zero-order Bessel function of the first kind.
V. WORKING PROTOTYPE
Let us consider a CDMA EV-DO network as shown in the
figure. Let us assume we are manufacturing an RNC and as
explained earlier, it is really important to make sure of its
capacity before rolling to down to the market. So, the tool that
does this task needs all the requisites explained before. Let us
consider we are using machines with any flavor of Unix
operating system and works influenced by the before
mentioned algorithms and design.
As we are analyzing an RNC, which is at the middle of the
whole network, with network components on both sides, and
packets sent from one side is received and sent to the
components on the other side. This requires a simulation tool
that should be capable of sending packets to the RNC and at
the same time should be able to capture the packets from the
other end and analyze it. Thus there is a need of a feedback
type of architecture and complicates the design.
Figure 2. Basic block diagram of EV-DO network
The figure 3 gives a high level diagram of the possible
simulation tool. One side, it acts as a generator and other side,
analyzer. Some literature describes building two separate
tools for these two purposes as Generator tool and Analyzer
tool [1]. But extra complexities are to be faced for
synchronizing both the tools and two separate architectures
required for two different tools. In measuring delay and
throughputs of the RNC under test, tools on both the sides
must be matched in time and parameters. Small variations
might lead to wrong measurements. Hence we, in this paper
propose a single simulation tool that is capable of doing both
the functions together. This includes the traffic model which
is very near to the actual traffic scenarios in the existing
networks. And the states changes are modeled by the next
state algorithm referred earlier.
Figure 3. High level architecture
Since the traffic analyzer has to cope with the real time
issues like computation of loss-oriented metrics and logging
of results, there is a need for an autonomous entity that can
work with the real time constraints. The performance of the
traffic analyzer depends on the reduction in time consumption
in processing the packets in real time. The module has to pass
system calls each time a packet has to be processed. Instead of
this, if the control is passed to the Unix kernel space and the
packet processing module be implemented to work in the
kernel space, then the overhead of system calls can be
eliminated. Hence UNIX kernel space will be the best choice
for the intended purpose.
The interfaces of the tool with other network components
like HCS (HDR Cell Site) or the link with PDSN takes place

ISSN: 2278 – 1323
1724
www.ijarcet.org
in different interfacing protocols in different levels on the
protocol stack. Hence the packets destined to the different
components are to be prepared with the respective protocols
and sent.
Apart from design and architecture, the simulation tool
needs a good User Interface for the users to effectively and
easily interact with the tool. Some literature deals with the
geographical maps of urban or rural areas based on the RF
path losses available from experimentation and GPS services
are used to calculate the attenuation at each Access
Terminal.[2]
Other useful information that that are desired by the
user are the data rates on both the sides, time of the simulation
and the number of calls registered and de-registered per hour
basis. These data serve as the observations of the capacity
testers and decisions are made based on these data.
Figure 4 Prototype of User Activity window
The figure 4 gives an overview of a User window that gives
the live details of some of the important details to the user.
There will be a heartbeat time interval within which the tool
gets the values from the network and refreshes the same on the
window. Here Register deregisters are the number of Access
Terminals registered and deregistered within the heartbeat
interval. FL and RL are the link between Cell and AT and vice
versa. The traffics on these links are also displayed. If the user
is running the functionality of any different kind of hand-off,
the number of hand-off's taken place per time interval and
hand-off type is displayed. One of the main scenarios tested
for telecom networks is Busy Hour Calling Attempts(BHCA)
is also considered in the tool and the numbers are displayed on
the user activity window. All these values are printed in their
respective windows in a line joining points pattern and user
can see the up's and down's occurred across the simulation
period. Decreasing the heartbeat time gives more precise
updates and come with disadvantage of consuming more
processor and memory resources.
VI. CONCLUSION
In this paper we have presented a prototype traffic generator
that can be used in CDMA networks and gives the ability
operators to validate their networks. This is very important
tool for the companies working on development and
enhancements of telecom network components. The paper
does not include any results, as the system cannot be
compared with similar one. However the systems are build on
such architecture and is used widely.
The system concept is simple; though innovative and
can strongly support the operator in its planning and
acceptance needs. In the future the prototype will become a
commercial product, including other air-interface modules as
well.
ACKNOWLEDGMENT
I am thankful to Alcatel-Lucent, Bangalore and my
manager, Ramacharyulu GP and guide Mrs. K Vijaya, for
their constant encouragement, guidance and support since
beginning, which has transformed me as a ―Subject Matter
Expert‖.
Last but not least, I would like to extend my gratitude to all
those who indirectly supported me for the success of this
paper.
REFERENCES
[1] Loukatos, L. Sarakis, K. Kontovasilis, C. Skianis, G.
Kormentzas, "Tools And Practices For Measurement–Based
Network Performance Evaluation", The 18th Annual IEEE
International Symposium on Personal, Indoor and Mobile Radio
Communications (PIMRC'07), 2007.
[2] Jim W. Roberts, France Telecom R&D, "Traffic Theory and the
Internet", IEEE Communications Magazine, 2001.
[3] Dimitrios Loukatos, Lambros Sarakis, Kimon Kontovasilis, and
Nikolas Mitrou ,―An Efficient ATM Traffic Generator for the
Real-Time Production of a Large Class of Complex Traffic
Profiles‖, Journal of communications and networks, vol. 7, no.
1, march 2005.
[4] Nikolaos Chr. Papaoulakis, Maria Marchitti, Sofoklis
Kyriazakos,Neeli Prasad, "Traffic Generation Platform in
W-CDMA Networks", IEEE 2007.
[5] Fratu, S Halunga, D N Vurreanu, ―Radio Link Simulation
Method For Umts Communications In Rayleigh Fading
Environment‖, IEEE 2003.
Niranjan U, presently pursuing M.Tech in Digital Communication at
BMSCE, Bangalore. I did B.E in Telecommunications from BNMIT,
Bangalore and published a paper earlier on Speech encoding in IJCA.
Presently I am doing internship in Alcatel Lucent in a tools development
team for EVDO technology.
Mr. Ramacharyulu GP is presently Technical Manages in Alcatel-Lucent
India Limited, Bangalore in a DO tools development team. He completed his
B.Tech in Andhra University and M.S. in BITS, Pilani. He has more than 25
years of experience. He have worked on various telecommunication projects
in companies like Ericson, HP, ALU on different Simulation tools for
Wireless, 2G, 3G technologies.
Mrs. K Vijaya is currently Assistant professor at BMS College of
Engineering, Bangalore. She has done B.E and M.Tech under Electronics
and Communications. She has 15 years of industrial experience and 6 years
of teaching experience. She has guided many M.Tech and B.E level projects
and published many technical papers.

1720 1724

More Related Content

What's hot (16)

Viewers also liked (20)

Similar to 1720 1724 (20)

More from Editor IJARCET (20)

Recently uploaded (20)

1720 1724