D9 - Research Methodology.lyx

#LyX 1.4.3 created this file. For more info see http://www.lyx.org/
\lyxformat 245
\begin_document
\begin_header
\textclass article
\language english
\inputencoding auto
\fontscheme pslatex
\graphics default
\paperfontsize 11
\spacing single
\papersize default
\use_geometry true
\use_amsmath 1
\cite_engine basic
\use_bibtopic false
\paperorientation portrait
\leftmargin 3cm
\topmargin 3cm
\rightmargin 3cm
\bottommargin 3.5cm
\secnumdepth 3
\tocdepth 3
\paragraph_separation indent
\defskip medskip
\quotes_language english
\papercolumns 1
\papersides 1
\paperpagestyle default
\tracking_changes false
\output_changes false
\end_header

\begin_body

\begin_layout Description
Name: Zhe Wei Chong
\end_layout

\begin_layout Description
Student\InsetSpace \space{}
Number: 40899013
\end_layout

\begin_layout Description
Supervisor: Marius Portmann
\end_layout

\begin_layout Description
Project\InsetSpace \space{}
Title: Analyzing and Extending a Wireless Mesh Network Monitoring
 System
\begin_inset VSpace medskip
\end_inset


\end_layout

\begin_layout Subsection*
Research Methodology
\end_layout

\begin_layout Standard
The main methodology for investigating wireless mesh network management
 for this project is constant experimentation to model a live mesh network
 while supported by an agile software development approach, and observation
 of results.
\end_layout

\begin_layout Standard
A small testbed is used to model a real world wireless mesh network with
 an arbitrary number of 802.11 wireless clients.
 This model will be used to study the optimal design of an SNMP-based wireless
 mesh monitoring framework, and the efficiency of the AODV\SpecialChar \-
MR routing protocol
 on multi\SpecialChar \-
radio meshes.
\end_layout

\begin_layout Standard
Two approaches have been identified for implementing the wireless monitoring
 framework using SNMP: gathering all statistics by having a designated manager
 poll each node, or let all nodes periodically send status notifications
 to the manager.
 Currently, a prototype of the wireless mesh monitoring module using the
 former polling approach has been implemented, but modifications will be
 made to delegate some of the message gathering to using SNMP notifications
 to improve the system's scalability.
\end_layout

\begin_layout Standard
Research on muti\SpecialChar \-
radio wireless mesh networks is relatively scarce.
 The wireless mesh monitoring system will allow close examination of the
 performance of AODV-MR, an extension to the AODV routing protocol to support
 multi\SpecialChar \-
radio multi\SpecialChar \-
interface wireless mesh networks 
\begin_inset LatexCommand \cite{1164791}

\end_inset

.
\end_layout

\begin_layout Standard
Software engineering for this project will carried out using an agile methodolog
y loosely based on the Dynamic Systems Development Method (DSDM).
 Like DSDM, the project is decomposed into a set of smaller parts.
 Each of these parts represent goals or research aims that should be achieved
 in the project, for example implementing a simple monitoring module for
 wireless mesh networks, or a web interface framework for mesh visualization.
 Development will be iterative and incremental, as software specifications
 is constantly revised when the focus of research shifts to different topics
 or angles.
 And more importantly, the project is largely driven by user input in the
 requirements definition and decision making process.
 I will be constantly communicating with my supervisor and assimilating
 his input into the project to develop new research ideas.
\end_layout

\begin_layout Paragraph
(325 words)
\end_layout

\begin_layout Subsection*
Collecting Results
\end_layout

\begin_layout Standard
The validation of the wireless mesh monitoring system will be based on the
 quantitative measure of the efficiency of such a framework.
 These metrics will be gathered using a number of methods from a live testbed,
 in which several implementation parameters will be changed to detect their
 effects on the overall status of the wireless mesh management framework.
\end_layout

\begin_layout Standard
Quantitative measurements may be collected using network monitoring software,
 for example Wireshark, RRDtool and Kismet.
 To do so, a machine is configured to actively listen to wireless transmissions
 (
\begin_inset Quotes els
\end_inset

promiscuous mode
\begin_inset Quotes ers
\end_inset

) and planted into the wireless mesh network.
 During the experiments, useful information such as wireless signal statistics,
 routing table dumps, AODV status dumps, traffic flow through multi\SpecialChar \-
radio
 interfaces, packets transmitted, bandwidth throughput and so on will be
 measured and recorded.
 Depending on the collection method, network information and routing tables
 can be retrieved either from SNMP modules, or read from the /proc virtual
 file system, or simply using readily available tools for Linux.
 Custom data, such as information related to the AODV routing protocol however,
 may have to be retrieved by kernel drivers or specifically written scripts.
\end_layout

\begin_layout Standard
During the experiments, parameters such as additional network traffic to
 be injected into the mesh and SNMP poll period may be adjusted to observe
 changes in the framework.
 Traffic injection can be done by generating a constant stream of network
 traffic from one end of the mesh to the other, which can be accomplished
 by third party software, or simply by initiating a data transfer.
\end_layout

\begin_layout Standard
It is expected that on a fairly large wireless mesh network, a polling period
 that is too short may cause the generated management traffic to saturate
 the mesh.
 By monitoring successful delivery rates for packets on the mesh while the
 parameters are being changed, it should be possible to deduce a threshold
 for the management traffic, which may possibly be used by the monitoring
 system to intelligently adjust its data gathering process.
\end_layout

\begin_layout Standard
Currently the wireless mesh testbed is based on conventional AODV.
 This model will be used as a comparison baseline with AODV\SpecialChar \-
MR, which should
 report an increase in routing efficiency.
 The monitoring methods is exactly the same as single radio meshes, but
 the monitoring system will have to be configured to retrieve statistics
 for the additional radio interfaces.
\end_layout

\begin_layout Paragraph
(377 words)
\end_layout

\begin_layout Subsection*
Analysis and Evaluation Techniques
\end_layout

\begin_layout Standard
The impact of a management framework on the wireless mesh network itself
 is not widely covered in literature.
 Qiu et al.\InsetSpace \space{}

\begin_inset LatexCommand \cite{Qiu2003}

\end_inset

 reports that data collection overhead is relatively low and will not significan
tly affect application traffic on the mesh.
 However, their experiments were carried out on a simulated model, and the
 scale of the test is relatively small with only up to 50 nodes.
 On the other hand, Henderson and Kotz 
\begin_inset LatexCommand \cite{Henderson2006}

\end_inset

 observes that an interval of below 5 minutes would have already saturated
 a wired LAN.
 Others vary greatly in implementation and size.
\end_layout

\begin_layout Standard
To deduce the optimal load of management traffic on the wireless mesh network,
 the traffic analyses will be compared against parameters for the mesh monitorin
g system.
 For example, it should be possible to observe the deterioration of packet
 transmission rates over SNMP polling period given a fixed number of nodes
 and constant traffic flow.
 On the other hand, the SNMP polling model will be used a comparison baseline
 against an optimized implementation later on that will employ SNMP notification
s for some messages.
 By only periodically sending status updates to the central manager if changes
 occur at the managed devices, we will have slightly reduced the traffic
 flow directly between them, thus reducing the management framework overhead.
\end_layout

\begin_layout Standard
Packet traces from actively monitoring the wireless mesh network can also
 be aggregated into their respective packet types and analyzed for better
 understanding.
 Similar to the SNMP polling period, we can expect that the percentage of
 management traffic to grow proportionally to the number of monitored nodes
 on the mesh network.
 It is also known that stream-based internet protocols (e.g.\InsetSpace \space{}
TCP) tend to perform
 badly on wireless adhoc networks due to latency caused by the routing protocols.
 On the other hand, experiments on AODV\SpecialChar \-
MR reports improved packet delivery
 rates and average latency, so it should be possible to reproduce the experiment
 using multiple TCP streams on the wireless mesh and observe if additional
 messaging overhead on the mesh may negate the improvements of multi\SpecialChar \-
radio
 wireless meshes.
\end_layout

\begin_layout Paragraph
(337 words)
\end_layout

\begin_layout Paragraph
A copy of this submission was emailed to my supervisor on 24/05/2007 at
 1pm.
\end_layout

\begin_layout Standard
\begin_inset LatexCommand \bibtex[ieeetr]{/home/zanglang/Thesis}

\end_inset


\end_layout

\end_body
\end_document