A new methodology in network reliability

Computer networks have turned out to be a more and more indispensable way of electronic communication. Not only the internet with its vast number of networked computers but also local or regional high-speed data networks are on their way to replace any other classical communication equipment like telephone, video conferences, radio, television, magazines, newspapers but also sensitive control devices. This makes the user of electronic media increasingly dependent on the proper functioning of the devices and connections which build the basic components of communication networks. This dissertation contributes to optimizing the dependability of highperformance networks. Graph-theoretical considerations combined with examinations of real-world networks and their predicted developments lead to a test framework which is used to analyze a novel high-speed restoration methodology which is capable of rapid calculation of network restoration paths. The methodology supports an arbitrary number of node and link failures in point-to-point or point-to-multipoint networks. Besides this universal approach, as far as the author knows, the algorithm reveals the lowest runtime of all currently presented restoration algorithms for general network graphs. Its operation is simulated on specially elaborated random graph models which can be tuned to represent most of the network topologies that currently exist. An additional chapter presents new methods for gaining a single similarity index, which allows the comparing of graph topologies. Finally the prediction of the structural development of existing network topologies is discussed. Models of this kind allow realistic performance and behavior tests in not yet existing network structures.