The Internet is constantly evolving. My research seeks to understand
the behaviour of the Internet, to allow us to improve its efficiency.
Many of these projects will provide the opportunity for students to
collaborate with researchers in Caltech, Cornell and the
Hamilton Instutute in Ireland.
Energy efficient networking
With the rapid growth of the Internet comes a similarly rapid increase
in its energy consumption. In data centres and network
points-of-presence, this is causing immediate problems with heat
dissipation. In the longer term, the energy consumption is contributing
to the Greenhouse Effect, and the world energy shortage. The challenge
of providing improved service with reduced energy can be addressed in
several ways.
Speed scaling in servers
The power consumption of a CPU can be changed by altering its clock speed.
By dynamically adapting the CPU speed to the work load, energy
consumption can be dramatically reduced with minimal impact on
performance.
This project will investigate new speed scaling algorithms
which achieve different optimisation objectives, and seek to find the
underlying mathematical structure of the solutions.
Interaction between transport layer (TCP) and energy saving
TCP was designed to operate over constant-speed links which are always
connected. However, energy saving technology allows lightly-utilised
links to slow down, or power down completely.
This project will
consider both the impact that these link fluctuations have on TCP
performance, and also how small changes in TCP's traffic patterns can
improve the ability of links to enter low-power states.
Impact of peer-to-peer technology on energy consumption
Peer-to-peer file distribution dramatically reduces the demands placed
on data centres, allowing them to consume less energy. However,
many computers remain powered on 24 hours a day, simply to be avilable
to participate in P2P file exchanges when required.
This project will investigate the energy tradeoffs in converting applications
from client/server to P2P models, and investigate new algorithms which
can improve the energy efficiency of P2P networks.
Modelling and evaluation of networks
Mathematical modelling
Measurement
Mathematical models are very helpful in designing networks, but only if
those models accurately reflect the true networks. Even such basic
questions as which data flows lose data when a router buffer overflows
are not well modelled.
This project will involve taking measurements on CAIA's broadband access
research testbed (BART) and Caltech's high speed WAN-in-Lab testbed to
develop accurate models of the behaviour of upper-level protocols in
real networks. It will also form part of the TCP evaluation test suite
being developed by the main Internet research body, the IRTF.
Requirements
I am looking for students who
Have a strong interest in telecommunications research
Have an excellent mathematical background, or a strong background
in both mathematics and programming
