Abstract
The development of multimedia applications and distributed data retrieval systems that facilitate the exchange of real-time data in the Internet has increased in the past few years. As a result it has become evident that different types of traffic may have certain requirements that need to be addressed when the traffic is injected into the network. Mechanisms that consider such requirements and give preferential treatment to identified flows of traffic are called Quality of Service (QoS) mechanisms. QoS metrics that are often used and are specific to packet level traffic characteristics are bandwidth, latency (delay), jitter and loss probability. Streaming multimedia may require guaranteed bandwidth to maintain certain quality levels while Voice over IP (VoIP) requires low delay as round-trip delays may be noticeable and degrade quality. QoS has historically been achieved mainly through differentiated forwarding, i.e. by maintaining different queues on a router level for traffic of different needs. There has also been research on differentiated routing, i.e. routing traffic with different needs through different paths. This is achieved mostly by using Multiprotocol Label Switching (MPLS). In this thesis we present a QoS mechanism, Intra-Domain Differentiated Routing (IDDR), that provides qualitative delay differentiation in plain IP networks . We first show that we can get substantial delay differentiation, around 30% on average, between two traffic classes within a network whilst retaining in many cases the throughput of successfully delivered flows. We can further double the delay differentiation with a minimal drop on throughput by optimising link weights using a genetic algorithm. The work is validated by simulation on multiple topologies and against many different traffic matrices to increase robustness. We then show that the IDDR can be modified and adapted for inter-domain delay differentiation. To test the inter-domain variant of IDDR we need a valid AS-level topology. Therefore, another contribution of the thesis is the AS-level topology generation that retains the properties of the internet.