Postgraduate Training & Research
Current Projects
See the honours or masters projects pages for current projects on offer.
The following projects are ideas which I am interested in but not currently actively pursuing. If one appeals to you, let me know. I am likely to be willing to supervise it.
Analysis of mobile ad-hoc routing protocols (honours)
Mobile ad-hoc networks are formed by mobile hosts which dynamically
join, leave and move within the network. No permanent network
infrastructure or administration is assumed. One of the main
challenges in supporting mobile networks is in designing routing
protocols that can efficiently handle the mobile nature of both the
hosts and routers. This project will identify current mobile
ad-hoc routing protocols and analyse their performance within the ns2
network simulator. Advantages and limitations of the routing
algorithms will be identified.
Ensuring fairness in mobile ad-hoc networks (honours/masters)
A second challenge of mobile ad-hoc networks is ensuring fair access.
Given the absence of centralised administration, it is feasible for a
user to flood the network with data, preventing others from
communicating. This project will investigate current methods of
ensuring fairness in networks and evaluate their applicability and
performance in mobile ad-hoc networks.
Multi-threaded packet network simulator (honours/masters)
The REAL network simulator is used widely in the research
community. It is accurate in the results it produces; but the
original design introduces many limitations to portability and to use
in a distributed environment. As one example, the original
program implements context switching through assembly language calls,
making it tightly coupled to the architecture and requiring assembly
language knowledge to port the program to new architectures. This
project will develop a new design to overcome these limitations using a
modern multi-threaded language (Java).
Graphical User Interface client for network simulation
REAL-edit is a graphical editor and client, written in Java, for
developing and running simulations on a (possibly remote) REAL network
simulator server. The model used by REAL edit has several
limitations. The current interface to the user is non-intuitive
and results in incorrect use. REAL edit also uses an active model
for it's connection with the server, which is typically blocked by
firewalls, limiting its use by clients on a different local area
network than the server. In addition the original architecture is
not well designed. This project will address these limitations
and develop an OO architecture for REAL edit with an intuitive user
interface, and passive connection to the server. Interviews
with students using the package in semester II is expected to be part
of the project.
Early Bird II - Network traffic anomaly detection using kernel
methods (honours)
The "Early Bird" project in 2003 investigated the use of Hidden Markov
Models for detecting Internet worms in network traffic. Some
encouraging results showed that HMM's were capable of detecting the
Saphire worm, which had just spread through the Internet.
However, HMMs were limited in their ability to train and classify
traffic quickly enough to be used as a live system. A more modern
approach to classification is to use kernel algorithms which typically
outperform HMMs in training and classification speed. This
project will implement a kernel algorithm for detecting Internet worms,
compare the performance of kernel methods to HMMs and investigate the
practicality of kernel method based Internet worm detection.
Compliant network architecture for supporting application service
requirements (honours/masters/PhD)
The existing protocol layer model separates applications from network
policy decisions that affect their performance. For instance, a
real-time video conferencing application is separated from the
congestion control
decisions made by the transport layer. In the event of network
congestion,
all network connections on the host begin to degrade.
Applications,
however, may be better served if they are able to specify and implement
policy in response to network events such as congestion. The
video
conference may prefer to drop the video channel in order to preserve
the
voice channel. To provide such capabilities to the application
will
often involve violation of the protocol stack layers, allowing the
application
to access information associated with lower protocol layers. This
project will investigate the application of the compliant architecture
model
to the problem of providing a structured method for accessing
information through mulitple protocol layers. The compliant
architecture model has been successfully applied in the context of
allowing applications to implement policy in
operating systems and the experience there can be translated to the
network
protocol stack.
This project will provide the TCP/IP base for the Middleware for
User Control of Priorities in Data-Rich Mobile Systems project - In
mobile computing environments reliable, secure and economical access to
central servers is not guaranteed. Thus support for the prioritisation
of I/O
by database and network middleware is an important driver in the uptake
of mobile technology. This project will investigate the constraints
imposed
on handling information in mobile environments, including the
modifications
that prioritisation may require for DBMS and TCP/IP architectures and
functionality. We will develop flexible prioritisation policies and
prototype middleware
able to provide mobile access to storage, transfer, management and
processing
services. In summary, we will develop a vertical model for I/O
prioritisation
in data-rich mobile environments.
Reducing delay bounds for generalised processor sharing (GPS) based
packet schedulers through a more accurate traffic arrival model.
(masters or honours with strong maths background)
Generalised Processor Sharing (GPS) is a theoretical algorithm that
is used as the basis for a wide range of scheduling mechanism that are
the proposed for providing guaranteed delay bounds within the
Internet. The published delay bounds for these schedulers are
based on network traffic flow assumptions that result in a correct, but
pessimistic bound on
delay. This project will apply previously developed techniques
for more accurately
bounding delay, to GPS-based scheduler queuing. Significant
improvement has been shown for FIFO and the extension of this model to
GPS-based schedulers will represent a significant advancement in
providing delay bound guarantees while minimizing the resources that
must be reserved to provide such guarantees.