| In recent years, there has been a growing interest in the Internet community to enhance the performance of operational networks. Traffic Engineering (TE) deals with the issue of performance evaluation and optimization of operational IP networks. One of the goals of performance optimization is load-balancing, which is accomplished by routing traffic in a way to utilize network resources efficiently. Traffic-sensitive routing is an older approach taken to achieve load-balancing. However, traffic-sensitive routing is associated with route-flapping and lack of convergence even in the absence of traffic or topology changes.; In this dissertation, we analyze the stability of traffic-sensitive routing. An analytical model has been developed to obtain the conditions under which routing with linear cost metric will be stable. A congestion-based metric has been proposed that leads to improved end-to-end throughput with significantly fewer routing updates. We have identified that the biased nature of traffic-sensitive routing is the main cause of instability and have proposed a loop-free, un-biased route computation algorithm.; We present Online Simulation (OLS) as a general network management tool and demonstrate minimization of packet loss in Open Shortest Path First (OSPF) networks by optimizing OSPF weights. In the area of inter-domain TE, we use OLS to achieve outbound load-balancing in Border Gateway Protocol (BGP).; This dissertation also presents a new connectionless traffic engineering framework for explicit, source-controlled routing in the Internet. The proposed framework does not require signaling, as in Multi-Protocol Label Switching (MPLS), or high per-packet overhead, as in IP source routing, and can be incrementally deployed in the Internet. The proposed framework decouples the path computation from the traffic splitting. |
