Date begun
|
Date ended
|
CAIA researchers
|
External collaborators
|
Title and Description
|
|
|
|
|
|
Oct 2009 |
(active) |
Lawrence
Stewart, Grenville
Armitage |
Greg Chesson (Google, USA) |
MCC:
Microburst Congestion Control
Microburst
triggered congestion affects environments where unsynchronised data
sources send data via a common path, which converges on the wire as a
highly synchronised and bursty traffic spike. Clustered data centre
environments with low latency, high bandwidth connectivity are the
obvious places where microbursts and incast can be observed. See the MCC project page
for more details. |
Sep 2009 |
(active) |
Mattia
Rossi, Grenville
Armitage |
Geoff
Huston (APNIC) |
Exploring
the Utilisation of IPv4 Address Space and Size of the NATed IPv4
Internet
With IPv4
address pool exhaustion being imminent, it is of major interest to
canvass to what proportion the allocated address space is utilised, to
allow the
development of new strategies for distribution of the remaining IPv4
addresses. It is also of major interest to approximately quantify the
amount of devices using
IP addresses to connect to the Internet to be able to plan proper IPv6
address distribution. With NAT (actually NAPT) being widely in use to
allow sharing of a single globally routable IPv4 address, and hosts
behind NATs being practically invisible, it is quite difficult to
provide detailed host counts. See the STING project page for more details. |
Apr 2009 |
(active) |
Warren Harrop,
Grenville Armitage |
Fred Baker (Cisco)
|
IPv4 and IPv6 Greynets
Based on various work (including earlier work at CAIA on Greynets),
we are developing an eventual IETF RFC documenting the potential
roll of IPv4 and IPv6 Greynets |
2009 |
(active) |
Lachlan Andrew |
Steven Low and Jayakrishnan Nair (Caltech) |
Avoiding heavy tails due to protocol interaction.
Many communication networks experience delays with "heavy tailed" delay
distributions, which means that very long delays are more likely than
traditional theory predicts. It was recently shown that retransmission
protocols can cause heavy tails where none existed. This work showed
that this effect is very fragile, and that most realistic protocols will
not induce heavy tails.
|
Feb 2008 |
(active) |
Lawrence
Stewart, Grenville
Armitage |
Dr Michael Welzl (University of Innsbruck, Austria;University of Oslo, Norway.) |
Evaluating next generation TCP congestion control
Empirical and simulation evaluation of how TCP congestion control algorithms like H-TCP and CUBIC interact with NewReno over consumer broadband links. |
June 2007 |
(active) |
Lachlan Andrew, Grenville
Armitage, Thuy Nguyen, Andrew Sucevic,
Dragi Klimovski, Adam Black |
Adam Wierman (Caltech),
Kevin Tang (Nithin Michael (Cornell))
Mung Chiang (Yannis Kamitsos (Princeton)) |
GREEN - Global Research into Energy Efficient Networking
The energy consumption of the network infrastructure underpinning the Internet is becoming increasingly important.
It is estimated to take about 2% of the energy consumption of industrialised nations, or about the same as the aviation industry.
The GREEN program is investigating ways to reduce this energy consumption.
See the GREEN project page for more details. |
2007 |
(active) |
Lachlan Andrew |
Taib Znati and Ihsan Qazi (University of Pittsburgh) |
Congestion control using minimal explicit network feedback.
Users want to send data as fast as the network will allow. When the
network becomes congested, it must signal users to slow down. This is
currently done implicitly, by discarding data. Many alternatives have
been proposed, but most require major changes to network infrastructure.
This project investigated a simple method by which the network can
tell users when to slow down, and when they can speed up.
|
2006 |
(active) |
Lachlan Andrew |
Fernando Paganini (ORT), Ao Tang (Cornell University) and Andres Feragut (ORT) |
Flow-level stability of TCP with general file size distributions.
How much traffic is too much for the Internet? If people try to send
too much data over a path through the Internet, congestion control will
slow down all of the data, making each transfer take longer. If
requests to transfer data come too quickly, the total number of
transfers in progress will grow indefinitely. It has long been known
that if file sizes are exponentially distributed, then typical
("alpha-fair") congestion control will be stable provided each
individual link is. This work extends that result to realistic
(non-exponential) file size distributions.
|
Nov 2004 |
Dec 2005 |
Irena Atov, Grenville
Armitage and David
Kennedy |
Bartek
Wydrowski (NETLAB,
California Institute of Technology, USA),
Lachlan Andrew
and A/Prof.
Stephen Hanly (CUBIN,
University of Melbourne, Australia)
|
Evaluation of FAST TCP using
Swinburne
University's Broadband Access Research Testbed (BART)
There is strong evidence that the
efficiency of the Internet is limited by its existing TCP congestion
control system. A replacement, called FAST TCP, is being designed at
Caltech to improve performance and it is emerging as a strong candidate
for a new IETF TCP standard. For its standardization and deployment, it
must be tested in a wide variety of environments, and it is necessary
that these tests be repeated by independent groups. To date, FAST has
been tested by Caltech and independent groups such as SLAC (Stanford
Linear Accelerator Center) and CERN (The European Particle Physics
Laboratory) in a wide range of high speed environments. However, there
is a pressing need for testing in low speed environments which are more
typical of the existing Internet. The current and medium term future of
access networks is in the 1-10 Mbps range, using such technologies as
ADSL and cable modem. FAST needs to work in these environments as well
as being able to scale to the high-speed regime. This project aims to
experimentally evaluate the performance of FAST under typical ‘edge of
network’ scenarios involving ADSL modems, cable modems and 802.11
wireless LANs. In particular, it will perform experiments using
Swinburne University's Broadband
Access
Research Testbed (BART) . It will seek to identify all
possible
failure modes of FAST in the test environments. The understanding
gained will also allow optimal parameter settings to be determined for
a range of conditions, such as link bandwidths, error rates and
propagation delays. More importantly, if weaknesses are discovered, it
will provide an opportunity to contribute to the evolving FAST
protocol. |
June 2004 |
Dec 2005 |
Sebastian
Zander |
Tanja Zseby (Fraunhofer Fokus,
Germany) |
Sampling
Techniques for Non-Intrusive
Statistical SLA Validation
Service
Level Agreements (SLAs) specify the network Quality of Service
(QoS) that providers are expected to deliver. Providers have to verify
if the actual quality of the network traffic complies with the SLA
without introducing significant additional network load and operational
costs. We propose a novel approach for non-intrusive SLA validation
that uses statistical SLAs and direct samples of the customer traffic
for the quality assessment. Based on pre-defined thresholds for QoS
metrics, we model the validation problem as proportion estimation of
non-conformant traffic. We compare the sampling errors of different
sampling techniques and present a novel solution for estimating the
error prior to the sampling. We also derive a solution for computing
the minimum sample fraction depending on the SLA parameters. Finally we
evaluate the proposed approach using real traffic from multiplayer
online games and prove that only a small fraction of the traffic needs
to be sampled to provide a customer with statistical SLA guarantees. |
March 2004 |
March 2006 |
Jason
But and Grenville
Armitage |
Urs Keller (Ecole
Polytechnique Federale de Lausanne,
Switzerland) |
NetSniff
- a Multi-Network-Layered
Real-Time Traffic Capture and Analysis Tool
The
recent widespread uptake of broadband access technologies has led to a
shift in how the Internet is being used. The availability of an
always-connected, high-speed Internet connection means that home users
are increasingly likely to use the Internet as an information
repository and content delivery resource. Higher content access speeds
coupled with a zero connection time means that Internet usage can
become more spontaneous rather than planned for. The ICE3
project
considers whether the traditional Internet access model (where
bandwidth at the edge of the network is orders of magnitude lower than
within the core of the network) could support an explosion in the usage
of new and evolving Internet applications, and particularly if the
network capacity hieracrchy was inverted (more bandwdith at the edge
than within the core of the network). In order to do this we need to
statistically analyse the performance of networked applications in
either environment. This is achieved using , a multi-layered network
capture and analysis tool. This tool is under ongoing development to
increase the number of supported applications and develop an increasing
dataset of real-world traffic statistics. For more information please
visit the ICE3
website. |
June 2004 |
Dec 2005 |
Irena Atov |
Richard
J. Harris (Massey
University, New Zealand),
Sanjaj K. Bose (Nanyang
Technological
University, Singapore)
|
Determining
Class-Based Bandwidth
Allocations on Links in Multi-Service IP Networks
The
growth of the Internet has brought with it problems of service quality
that were not really thought of when the “best-effort” design of the
network was originally envisaged. The Internet is now planned to be
used by a variety of different services which have different kinds of
service requirements. These range from the old-fashioned “best-effort”
services to ones which require real-time traffic like voice or video to
be carried with reasonable delay, delay jitter and data loss. The
control of the network with this kind of traffic requires careful
resource provisioning as standard Weighted Fair Queueing (WFQ) service
disciplines employed in IP QoS networks can only provide tight
end-to-end delay guarantees for the classes if an adequate level of
resources (in terms of bandwidth and buffer space) is allocated along
their respective data paths through the network. In this project we
focus on development and analysis of recursive methods that can be used
for inversion of some of the well known traffic decomposition models
(e.g., QNA) and can provide basis for network dimensioning with
multiple service classes. The goal is to develop computationally
efficient algorithms for determining class-based bandwidth allocations
on the links subject to satisfying varying end-to-end QoS constraints
for the classes. |
2003 |
2004 |
Philip Branch |
IBAP Pty Ltd. |
Lawful Interception of Emerging Technolgoies
This project investigates some of the issues dealing with lawful interception of emerging network technologies.
|
Sep 2002
|
Jan 2005
|
Grenville
Armitage |
Les Cottrell (IEPM group, Stanford Linear Accelerator
Centre, USA) |
An
Australian node of PingER
PingER
(Ping
End-to-end Reporting) is the name given to the Internet End-to-end
Performance Measurement (
IEPM ) project to monitor end-to-end performance of Internet
links. PingER involves hundreds of sites in many countries all over the
world. CAIA collaborated with IEPM to provide an Australian node to the
PingER project, which involved running regular 'ping' tests against a
list of international sites every 30 minutes and reporting our results
back to the IEPM team at Stanford's Linear Accelerator Centre. Our site
began operation in September 2002. As of January 2005 our PingER node
is temporarily offline. |