The Importance Sampling based Traffic Engineering (ISTE) technique proposed in this thesis is a brand new approach. Regardless of how large and how complex the network is, as long as the network behaviors of interest can be modeled as stochastic events, and the input of the network can be modeled as a stochastic process, ISTE can predict how the network is going to behave when the network input changes. ISTE will take much less simulation time in comparison to the traditional ad-hoc heuristic approaches which typically require step-by-step simulations or measurements. Our simulation results show that the ISTE approach is fast, simple, and practical for network traffic engineering. Our ISTE approach is much more powerful than all analytical approaches which typically become intractable when more than one node is involved. When applied to more complex scenarios where congestion may be caused by multiple traffic flows, an Alternating Twisting (AT) technique is developed to overcome the increasing noise when twisting single traffic flow. |