• NOTE: The Cambridge Networking Group has been merged into the Cambridge Systems Group. Please visit those pages for the most up-to-date information.

Our research targets the development of network-aware applications that can provide a desired Quality of Service to users. Existing network applications are typically unresponsive to bandwidth and latency fluctuations. The transport protocol attempts to optimize performance for a single flow rather than for the aggregate system.  Our aim is to advance the underlying theory and develop algorithms whereby applications can measure and respond to changing network conditions by co-operatively sharing bandwidth between flows in order to optimize system performance.

More generally, we study problems of decentralized control. How can we co-ordinate the actions of a large number of agents, with a minimum of communication between them, in order to achieve some objective? Each agent must act on the basis of very limited information. For example, agents might only be able to observe the actions of their neighbours, or might only know about the average of all actions. Can they nevertheless select their actions so that the system collectively behaves in a desirable manner?

Networking research is also carried out in Redmond and at the Bay Area research labs.

Admission Control and Rate Adaptation

Communication networks carry a mixture of elastic traffic, such as Web downloads, and inelastic real-time traffic. For the latter, quality degrades appreciably below a specified throughput and so bandwidth sharing is inappropriate. It is preferable instead to accept such traffic only when it is possible to guarantee that it receives a specified bandwidth for a specified duration. Rather than do this in a centralized manner, we study a decentralized solution, where a call or flow probes the network and decides whether to admit itself on the basis of these measurements. More generally, a flow may choose among a small set of rates on the basis of network measurements.

Self-Organising Overlay Networks

Self-organising overlay networks are the keystone of modern peer to peer networks. The design of the overlay dramatically varies the efficiency of the network along multiple axes including: bandwidth, latency and robustness. We are conducting research into theory related to such networks.

Network Inference

Knowing the topology of your network is beneficial both for fault diagnosis and resource provisioning. While other researchers have concentrated on mapping networks using SNMP information from switches and routers, we map networks without this support. This is useful in home or small office scenarios, where users cannot be expected to invest in expensive networking equipment. We also use the topology components to permit inference of network bandwidths and conditions.

Mobile IPv6

Using the Windows IPv6 stack developed by the Systems and Networking group at Microsoft Research Redmond, we have been connected to British Telecom's early trial IPv6 ISP service since early 2000. As part of this activity, we developed a Tunnel Broker to aid sites in getting connected to the IPv6 Internet (using tunnels through the existing Internet). You can get this software along with the IPv6 stack for NT here. We have also been supporting Mobile IPv6 work occurring at Cambridge, more details of which are here.

Primary Contact: Peter Key
Affiliate Members
Maltz, Dave	Thaler, Dave	Photo Not Available Wang, Helen	Zhu, Bin

Exploiting Explicit Congestion Notification (ECN)

How can resources such as bandwidth be shared fairly between users of an Intranet or the Internet? This involves a combination of information and incentives.  Currently, in TCP, packet loss is the only indicator of congestion. We aim to improve performance by using better congestion information, for example as in the recent ECN proposal.  We are also exploring ways of providing differentiated Quality of Service and mechanisms for introducing incentives through usage charges. (Read more...)

Large Deviations

If you roll a die a thousand times and compute the average of the numbers seen, how likely is it that the average is close to 4? And if the average is close to 4, how many 1s, 2s and 6s were there? Answers to such questions are given by the theory of large deviations, which plays an important role in the probabilistic analysis of large systems made up of homogenous components. We apply this theory to communication networks and traffic modelling.

Congestion Pricing and a Distributed Game

We are looking at novel ways of controlling a network by using Congestion Pricing to achieve differential QoS. Signals are related to shadow prices, the marginal cost of congestion, and fed back to the user. The users are free to react as they chose, but will incur 'charges' when resources are congested, so in effect the users are playing a 'game' against the network. (Read more...)