The growth of the Internet and the range of applications it now supports has created a need for improved traffic engineering techniques. One protocol which shows promise in this regard is Multiprotocol Label Switching (MPLS). MPLS inherits a mix of attributes from earlier protocols such as IP and ATM, and potentially combines the simplicity of IP and the Quality of Service (QoS) capabilities of ATM. MPLS is now a mature standard widely deployed in the Internet. This thesis concerns the development of new mechanisms that can further extend the MPLS capabilities for traffic engineering. Web service remains a key application in today's Internet. The traffic demands at popular Web-sites and the requirements of redundancy and reliability can only be met by using multiple Web servers. A new solution to Web server load balancing based on MPLS is presented in this thesis. This solution features a novel Web switching architecture featuring switching at layer two. An extended solution for providing differentiated Web services is also proposed . It has been implemented in a soft MPLS router using the Linux operating system. The performance of soft routers is significantly affected by the packet processing time. An MPLS-based framework to increase the average packet size and consequently reduce the traffic frame-rate is described in the thesis. This has been implemented in a Linux-based soft router and its performance evaluated experimentally. As transmission rates continue to rise, such aggregation techniques will be needed if packet processing time is not to become a bottleneck. The switching technology at the core of tomorrow's Internet, featuring GMPLS and optical switching using , perhaps, optical burst switching technology, will not work efficiently with short packets. A new class of scheduling algorithms is also described, intended for deployment in MPLS networks. Their operation is based on an analogy with the workings of the human heart. This class of algorithms achieves the optimal fairness for packet based schedulers and has low hardware complexity. It can be combined with the packet aggregation mechanism above to provide an effective interface between the edges of tomorrow's Internet and its high-speed core.