IP Networking
Current and previous projects in this area include
MESCAL,
TEQUILA and
PRO-NET.
This work focuses on the provisioning of multi-service group communications in a Differentiated Services (DiffServ) environment. Specifically, we propose a QoS-Source Specific Multicast (QSSM) framework that is based on the promising Source Specific Multicast (SSM) architecture. From a routing point of view, the project involves constructing QoS-aware source specific multicast trees on a per-class-of-service basis inside DiffServ domains. Both intra-domain multicast routing protocols (e.g. PIM-SSM) and inter-domain multicast routing protocols (e.g. MBGP) have been investigated and extended so as to achieve group data delivery with the required end-to-end QoS guarantees. From a scalability point of view, the proposed scheme avoids imposing additional overhead for QoS state maintenance at DiffServ core routers. In effect, the most distinct advantage of QSSM is that it needs no extensions to the underlying multicast routing/forwarding table structure for the inclusion of heterogeneous QoS requirements from end users, and this is an important requirement for practical deployment at large scale. Meanwhile, we also address IP-level multicast traffic engineering issues, e.g., how to balance multicast traffic inside the network so as to achieve optimised flow distribution. Interoperability of unicast and multicast traffic aggregation in DiffServ-based traffic engineering will also be investigated.
The QoS testbed is maintained by the Networks Research Group and has been used for a number of research activities. It consists of six Linux-based software routers, realising a small-scale network topology. The main purpose of the QoS testbed is to accommodate and validate research activities carried out by the Networks Group as well as to assist MSc students with their projects. Two examples of projects which have used the testbed are as follows:
- The implementation of an experimental network dimensioning algorithm applied on DiffServ over MPLS-enabled networks that has been part of the research work carried out in the TEQUILA project;
- The implementation and validation of a QoS configuration, auditing and monitoring application based on the PARLAY specifications, for the MANTRIP project.
Internet traffic engineering is defined as that aspect of Internet network engineering that deals with the issues of performance evaluation and performance optimisation of operational IP networks. The Internet exists in order to transfer information from source nodes to destination nodes. Accordingly, one of the most significant functions performed by the Internet is the routing of traffic from ingress nodes to egress nodes, and therefore an important function of Internet traffic engineering is the control and optimisation of routing, so as to steer traffic through the network in the most effective way. The optimisation aspects of traffic engineering can be achieved through capacity management and traffic management. In this work we look into how to control capacity management and routing in order to achieve both network operational and service-oriented objectives. We assume the Differentiated Services service model, and we seek to accommodate customer contracts as defined in Service Level Specifications. The work has resulted in a holistic architecture for service-driven traffic engineering, with the associated algorithms for routing optimisation and capacity allocation. These algorithms have been evaluated through simulations (using Network Simulator - NS2), and we are in the process of validating the results on the Linux-based QoS testbed which is maintained by the Networks Group.
Internet traffic engineering is defined as that aspect of Internet network engineering dealing with the issue of performance evaluation and performance optimisation of operational IP networks (RFC3272). In order to monitor, evaluate and verify network performance it is important to understand the network traffic characteristics. Consequently, in this work we have initially modelled a single node Ethernet workstation's performance in terms of an analysis of its traffic measurements (packet size and inter-arrival time for each of WWW, FTP and Telnet traffic). We have compared measured results with theoretical mathematical distributions in order to understand the network traffic's characteristics. This work will be linked later in the study with network routing and queuing theories in order to explore the network's ability to support QoS. The figure illustrates how the mathematical distributions agree with the measured results.
Computer simulation provides an invaluable tool for engineers and scientists to better understand traffic control issues such as network control, protocols and network architectures. As part of this, accurate traffic source models are particularly useful for analysis. A major research topic in IP traffic engineering is the development of traffic source models that are consistent with empirical data obtained from operational networks. Such models should also be tractable and amenable to analysis. This research is studying the nature of IP traffic in some network models (classic traffic model and self-similar traffic) developed in Javasim. The results and their analysis show the characteristics of the IP traffic. This work will be linked later in the study with network routing and queuing theories in order to explore the network's ability to support QoS and modify traffic models for traffic engineering.
Traffic Engineering (TE) techniques have emerged over the last decade in an attempt to optimise operational IP networks by utilising network resources efficiently and reliably. To date, the primary use of TE has been focused within a domain (intra-domain TE). However, lack of incentives and support for traffic engineering between domains (inter-domain TE) results in poor management of inter-domain resources, which in turn can significantly affect the end-to-end performance. Network operators now recognise that the management of inter-domain traffic is part of their operation costs that should be optimised. In other words, traffic engineering not only needs to be considered within a domain, but also between domains. The current de factor inter-domain routing protocol is Border Gateway Protocol (BGP) which provides policies to select the best path in the global Internet. As TE concerns how to route packets efficiently, we consider modifying and optimising BGP routing as a primary goal of inter-domain TE. We are also investigating how to make quality-of-service and TE work together between administratively distinct domains with different capabilities and policies.
Differentiated Services are seen as an important technology that can support QoS in the Internet without the scalability problems of Integrated Services. The fact that in Differentiated Services per-flow information is kept only at edge and not at core routers allows for more efficient utilisation of bandwidth and buffering resources in the core network, but on the other hand makes supporting strict QoS guarantees on a per-flow basis a more difficult task. In this research we have considered bandwidth management and admission control. In the former, the traffic demands of aggregated traffic in terms of both bandwidth and buffering have been investigated and some implications on traditional traffic engineering techniques have been deduced. As regards admission control, a measurement-based approach has been employed with an attempt to be as non-heuristic as possible so as to be generic and not dependent on individual traffic sources' characteristics. All cases have been validated through extensive simulations.
