Traffic Engineering for Quality of Service Provisioning in IP Networks

Panos Trimintzios

Centre for Communication Systems Research (CCSR), University of Surrey, Guildford, UK.

Internet Service Providers (ISPs) strive to make efficient use of their investment. They try to slow down the pace of overprovisioning, and effectively the induced cost. Even when network resources are overprovisioned, performing traffic engineering is desirable because a loadbalanced network will reduce the need for further upgrades and will achieve better performance.

Quality of Service (QoS) delivery in IP networks is also an important area for ISPs, pointing to new business opportunities for value-added services. The Differentiated Services (Diffserv) architecture has been proven to be the service model that overcomes the scalability hurdle, with aggregate QoS differentiation that pushes complex per-flow tasks at the edges of a domain. The objective of this thesis is to investigate the application of traffic engineering techniques for the provisioning a Diffserv domain in order to meet specific performance targets.

Services are currently offered on the basis of Service Level Agreements (SLAs), which set the terms and conditions for both providers and customers for a single level of QoS guarantees. We define the Service Level Specification (SLS), the technical part of an SLA, which can support multiple-level QoS guarantees. The provisioning of SLSes in a single domain is based on the concept of the edge-to-edge QoS Classes (QCs). SLSes and QoS Classes are the missing links in offering value-added services.

We take a holistic view to the issue of provisioning a Diffserv domain, by identifying a single architecture with all the required functionality and the interfaces between the functional components. The design of the architecture includes multiple timescale operation levels.

We focus on the offline traffic engineering aspects of this architecture. We devise the serviceoriented traffic estimation approach to complement measurement based approaches found in the literature. We propose algorithms for the problems related to Network Provisioning. Our targets are to meet the traffic demands, and efficiently balance the load in the network at the minimum cost. This engineering approach effectively avoids transient congestion of the network, reducing the expected load to acceptable levels as we show in our performance evaluation.

Finally, as part of our Network Provisioning algorithms we look into the problem of efficiently provisioning for point-to-multipoint traffic demands; this is known as the hose demand and provisioning model. The latter is an important provisioning model because it reflects better than the point-to-point (or pipe) model the demands resulting from interdomain routing processes and of value-added services such as Virtual Private Networks (VPNs). We study the effectiveness of tree computation algorithms with QoS constraints for provisioning in the hose model.

Key words: Traffic Engineering, Network Provisioning, Quality of Service, Differentiated Services, Route Computation, Hose Provisioning Model

PhD Thesis, April 2004.

The full thesis in Acrobat pdf (2.2M) can be made available by contacting the author (ptrim (at) ics.forth).