Internet Traffic Engineering

Researcher : B. Quoitin

Research team: INL (IP Networking Lab)

Collaborations :

• Research Unit in Networking, ULg, Belgium,
• Unité de gestion logistique et modélisation, UCL, Belgium,
• DANTE, UK,
• Laboratoire d'Informatique de Paris 6, France,
• Polytechnic University of Catalogne, Spain,
• Intel Research, Cambridge, UK,
• Micro Research, Belgium.
• Telefónica
• France Telecom
• Algonet
• University College London
• University of Surrey

Funds : FRIA; TOTEM and AGAVE projects

Description :

Today, the Internet connects more than 18000 Autonomous Systems (AS), operated by many different technical administrations. The large majority are stub ASes, i.e. autonomous systems that do not allow external domains to use their infrastructure, except to reach them. Less than 20% of autonomous systems provide transit services to other ASes. They are called transit ASes. The Border Gateway Protocol (BGP) is used to distribute routing announcements among routers that interconnect ASes.

The objective of the research carried in this area is to develop techniques and tools that allow IP- and MPLS-based networks to engineer their traffic. Our focus in on techniques that are applicable between ASes on the global Internet.

Three projects are carried out in this area:

• the TOTEM project,
• the IPv6 multihoming and traffic engineering,
• the AGAVE project.

The objective of the TOTEM project (TOolbox for Traffic Engineering Methods) is to develop a toolbox of algorithms for traffic engineering purposes (http://inl.info.ucl.ac.be/projects/totem). Therefore, we will unify the algorithms which we already proposed these last years and we will develop new techniques of traffic engineering applicable in a single network as well as between distinct networks. These techniques will take into account the distribution of the traffic, the fault-tolerance requirement and the support of quality of service. We will develop generic algorithms for the optimisation of networks of big size which will apply, on one hand, to IP networks, and on the other hand, to networks operated with (G)MPLS. Some of these algorithms will require extensions to the routing (OSPF-TE, ISIS-TE, BGP) or signalling (RSVP-TE) protocols which we will submit to IETF. Our toolbox will be available in open source and will be designed such that its elements can easily be integrated in various platforms such as Linux PCs or routers.

The second research activity deals with multihoming domains. Internet connectivity takes a strategic importance for a growing number of companies. Therefore, many ISPs and corporate networks wish to be connected through at least two providers to the Internet, primarily to enhance their reliability in the event of a failure in a provider network, but also to improve their network performances such as network latency. Nowadays, at least 60% of stub domains are multihomed to two or more providers, and it is expected that IPv6 sites will continue to be multihomed. In order to preserve the size of the BGP routing tables in the Internet, every IPv6 multihoming solution is required to allow route aggregation at the level of their providers. Most IPv6 multihoming mechanisms proposed at the IETF rely on the utilisation of several IPv6 provider-aggregatable prefixes per site, instead of a single provider-independent prefix. No IPv6 multihoming solution today provides at the same time sufficient traffic engineering techniques, full fault-tolerance, full route aggregation, provider independency and the preservation of active TCP connections. We focus on the traffic engineering issues related to IPv6 multihomed stubs.

In this work, we show that there exists a multihoming solution for IPv6 that does not jeopardise the interdomain routing in the Internet, and that still allows ASes to engineer their interdomain traffic. In particular, we show and provide a technique for stub ASes to easily balance their outgoing traffic on the links with their providers. Moreover, we show that IPv6 multihoming with several IPv6 provider-aggregatable prefixes not only allows the Internet to better scale, but also provides wide opportunities to develop the quality of service in the Internet. We show for instance that the use of IPv6 multihoming can reduce end-to-end delays by leveraging the Internet path diversity.

Next, we propose a new technique that allows IPv6 stub ASes to make use of new paths that are otherwise hidden by BGP. Using the proposed technique, stubs are able to select paths with low delays in a scalable way. The technique is based on the use of synthetic coordinates, computed by a decentralised algorithm that we improved. We finally provide a similar technique but for IPv4 multihomed sites.

Finally,  AGAVE (A liGhtweight Approach for Viable End-to-end IP-based QoS Services) is developing solutions for open end-to-end service provisioning based on the notion of Network Planes that may be interconnected across multiple providers to create Parallel Internets tailored to service requirements. The project is investigating a range of Traffic Engineering techniques to realise Network Planes. A lightweight QoS approach is being developed, based on the principles of differentiated routing with inherent load balancing and resilience, without requiring universal deployment of differentiated forwarding.