Providing high throughput and controllable performance in burst -switched optical networks

Optical fiber networks, and in particular wavelength division multiplexing, are the only technologies capable of supporting the huge bandwidth demand that has accompanied the explosive growth of the Internet. Optical burst switching (OBS) has emerged as a very promising candidate architecture for next generation optical networks because it combines the simple optical technology of optical circuit switching with the flexibility and bandwidth efficiency of optical packet switching. The most distinguishing feature of OBS is the use of a time offset between the header and payload of each burst. These offsets allow for the time required to perform switching and forwarding operations in each node and directly affect the complexity, throughput performance and end-to-end delay of OBS systems.;By increasing the understanding of the effect of OBS offsets and by providing mechanisms by which they can be precisely controlled, the contributions of this work provide OBS researchers with a powerful toolset for providing the reliable and predictable end-to-end performance guarantees required by next-generation Internet applications and brings optical burst switching one step closer to commercial viability.;In this thesis, I examine the problem of providing high throughput and controllable and predictable performance in OBS networks. I present new analytical models which, for the first time, allow one to quantify precisely the effect of offsets on the performance of OBS networks. I then present a new OBS signaling architecture called dual-header optical burst switching (DOBS), which allows for precise control of offsets in every node of the network. I show how DOBS signalling can reduce the complexity of core-node burst schedulers from O(log W) to O(1), where W is the number of wavelengths in the system. I also present a new scheduling algorithm that minimizes burst blocking in core nodes of a DOBS network. This allows the system to support an offered load that is between 10% and 50% higher than that of a single-header OBS system for a given target blocking probability.