Minimizing packet delay (or packet latency, as it is often called) is an impor- tant goal in modern telecommunication networks. In a network... Show moreMinimizing packet delay (or packet latency, as it is often called) is an impor- tant goal in modern telecommunication networks. In a network with given physical and topological characteristics, uneven distribution of tra c between nodes may re- sult in parts of the network being underutilized while other parts may experience congestion and its related e ect of higher delays, and even packet loss if bu ers be- come overloaded. Overload results in longer queuing delays, which are a major, if not the major, source of packet delay. Internet nodes typically route based on a single best (shortest) path without taking into account link occupancy and without taking full advantage of all available network resources. So a method for improving network performance is to allow nodes to use multiple paths to route packets to a given desti- nation. In this research project the main objective is to nd the best algorithm that, applied to a multi-commodity network, produces multi-path ow assignments that minimize the maximum cross-network delay. The motivation for this objective is the tighter requirements for quality of service coming from real-time streaming services, such as voice and video, which restrict the maximum source-destination packet delay. An analytical approach, based on Lagrange Multipliers, was investigated which led to the design of an iterative algorithm that can be proved to converge to the opti- mal path and ow con guration. From that algorithm a simpli ed heuristic algorithm was derived, that achieves results close to optimal. Another heuristic but more intu- itive algorithm, similar to the algorithm described in [6], was also investigated and found to also o ers good results and to be computationally e cient. The two heuristic algorithms were simulated with Java programs and, as a benchmark, the single path algorithm that is commonly used in the Internet was also simulated. That simulation was also done with Java programs. Performance comparisons for the three types of simulations is also provided. For a more realistic evaluation, the intuitive heuristic algorithm was also simu- viii lated in the NS-3 simulation environment (in addition to the Java simulation). NS-3 is a state-of-the-art industry standard widely used in the academic world, and for it a distributed routing protocol, that implements the maximum delay minimization algorithms without centralized control, was designed. PH.D in Computer Engineering, July 2013 Show less
