03-09-2012, 11:34 AM
Analysis of simulation environments for mobile ad hoc networks
2Analysis of simulation.pdf (Size: 213.36 KB / Downloads: 32)
Abstract
This document is directly derived from Deliverable D11 of the EU-funded project BISON [4].
It discusses the characteristics of simulation environments for telecommunications networks,
and, more specically, for mobile ad hoc networks. Different available network simulators are
discussed and compared according to a number of criteria ranging from the reliability of the
provided simulation models, to the degree of usability of software and graphical interfaces. The
aim of the document is to identify the simulation environment which is the most appropriate to
run experiments to test and validate the design of novel adaptive routing algorithms developed
in the framework of BISON. However, most of the discussions and results hold in general.
The document concludes that the (commercial) QualNet [21] simulator appears to be the most
satisfactory product currently available.
Introduction
This document is directly derived from Deliverable D11 of the EU-funded project BISON [4].
It discusses the characteristics of simulation environments for telecommunications networks,
and, more specically, for mobile ad hoc networks. The aim of this document is to identify the
simulation environment which is the most appropriate to run experiments to test and validate
the design of novel adaptive routing algorithms developed in the framework of BISON. In spite
of the focus on BISON's specic objectives, most of the reported discussions and results hold
in general.
Discrete-event simulation is the main tool used to study the characteristics and predict in some
extent the behavior of communication networks, and, more in general to study complex stochastic
dynamic systems modeling real-world situations of practical interest. Generally speaking,
simulation is an extremely powerful and exible tool. It potentially allows to study a large
number of possible system congurations controlling the amount of complexity and realism
included in the simulation model. The purpose of a simulation study consists to draw conclusions
that are at the same time statistically sound, meaningful, and of practical interest. The
exact denition of what aspects of the system under study should be included in the simulation
model and their level of detail are the central design choices constraining the quality of the
nal output and of the derived conclusions.
General characteristics and requirements of simulation studies
Generally speaking, simulation (e.g., [3, 2]) can be dened as the process of designing a model
of a real system and conducting experiments with this model for the purpose of understanding
the behavior of the system and/or evaluating various strategies for its control. To be this
process useful, the behavior of the model is expected to faithfully mimic the response of the
system under study.
Given the intrinsic discrete nature of the events happening in a communication network (e.g.,
start/end of user sessions), in BISON we are mainly interested in discrete-event simulations.
In a discrete-event simulation model, state variables only change in correspondence of events
happening at discrete points in time. Events occur as a consequence of activities and delays.
Active simulation elements may compete for system resources, possibly waiting in queues to
get access to an available resource. A discrete-event simulation model runs over time through
a mechanism that moves simulated time forward according to the sequence of events. The
system state is updated at each event along with the capturing and/or freeing of resources that
may occur at that time.
A number of logical and practical steps have to be followed in any simulation study [3]. In
particular, three major phases of the process of building and using a simulation model can be
identied: (i) denition of the model of the real system, (ii) collection of data from the real system,
and (iii) denition of the experimental design and of the actual run of the simulation using the
dened model
Simulation model and selection of the level of detail
It is well understood that in general terms all the physical and logical components listed in
our abstract representation are equally important to determine the overall dynamics of the
network system. Therefore, all these components should be also included in our simulation
model. However, the open question is: which is the right level of detail and which are the core
specic characteristics that shall be used to model the different network components inside the
simulation architecture? For instance, do we need to implement all the OSI layers and let every
packet passing through them at each network node, or, is it sufcient to limit ourselves to the
MAC, network, and session layers? How faithful has to be the implementation of each selected
OSI layer? Is it a meaningful choice to model the environment simply as a small open at area
given that it is hard to think of a possible application scenario with these characteristics? Must
the mobility of the nodes reect some precise model of human behavior, or, can it be modeled
in terms of little more than a random walk? Do the transmission range and the modality of
access to the medium have to be adapted to the values of devices currently available on the
market, or, can it be envisaged the use of smarter and more powerful technologies? Should the
input trafc patterns be dened in the generic terms of negative exponential distributions, or,
should they be close to the patterns usually encountered in GSM or other wireless networks?
And so on . . . The appropriate choice always depends on several factors starting from the
specic objectives of the research.
Modules for the OSI protocols, mobility models, and radio propagation
A simulator coming with a rich set of already implemented and tested modules is preferable
to a simulator coming with no implemented modules. However, according to the previous
discussions on the extreme sensitivity of the obtained results to the model characteristics, it is
clear that not only the number but also the characteristics (quality) of the implemented modules
are equally important.
The OSI modules that are really interesting for us are primarily those relative to routing protocols.
It is also very important to have a reliable implementation of the MAC layer. The physical
and link layer, if considered in the perspective of the radio emission/reception are also important.
At the application or session layer it is probably enough to have simple FTP-like or CBR
modules. For what concerns the transport layer, only a best-effort (unreliable) UDP-like service
is probably sufcient. A TCP-like service would be in principle very useful in the context of
mobile ad hoc networks, but unfortunately there are a number of difcult problems related
with this issue. Therefore, we will likely not use any form of reliable transport protocol.