Study On Buffering And Transferring Techniques In Optical Packet Switching Networks

Optical communication technique has been widely utilized in modern communication,especially in back-bone transmission. As one main approach of future high-speed and wide-band communication, all optical network (AON) is facing two key issues, optical trans-mission and optical switching. Since enormous transmission capacity yet with very lowcost can be obtained by ultra-high-speed transmission and wavelength-division-multiplexing(WDM), the study on AON has concentrated on optical switching techniques for taking fulladvantage of optical transmission. With small switching granularity and ?exible schedul-ing, optical packet switching (OPS) has been considered as the goal of optical switchingtechnology.In OPS networks packet contention occurs inevitably when two or more optical packetsare going to leave switching node from the same output port on the same wavelength simul-taneously. How to solve the contention becomes the key to realize practical OPS networks.Since appropriate buffering and transferring techniques will be very challenging solutions tosuch issue, this dissertation focuses on the study in this field including buffer device design-ing, buffer scheduling algorithm, buffer performance evaluation, packet transferring scheme,etc. The dissertation is organized as follows:In Chapter 2 optical buffer technology is investigated. A hybrid optic-electronic bufferdevice and a cyclic-queue-based access scheme are proposed. In order to deal with the lowefficiency of fiber delay line (FDL) used in optical buffering, the hybrid structure consistsof optical part and electric part. Especially the optical part is wheel-like cyclic FDL loopbuffer constructed by a number of shorter FDLs so that the buffer can offer much smallerdelay granularity with the same total length. Hence the utilization efficiency of FDLs can beincreased, and the contention solving capability of switching node can be improved remark-ably.In Chapter 3 a quantized delay buffer (QDB) model is proposed to analyze the perfor-mance of FDL buffer. Considering the delay quantization in FDLs, delay time and waiting time of packet are discussed. Without any specific assumptions of packet arrival process andlength distribution, the model presents a generic approach to study the delay time distribu-tion and modify the Lindley integral equation for the waiting time distribution. Analytic andexact results of the two distributions mentioned above can be obtained without any approxi-mation. The accuracy of the model is validated through simulation.In Chapter 4 the performance of FDL buffer system with finite waiting places is eval-uated. The probability generating function of delay time distribution can be obtained fromthe QDB model presented in chapter 3. By means of Embedded Markov Chain method it isable to figure out queue length distribution, loss probability, limiting maximum of modifiedaverage arriving rate, limiting maximum of modified traffic load, mean of delay time/waitingtime/reservation time, and other important performance measures. The accuracy of the pro-posed approach is also verified through simulation.In Chapter 5 the performance of FDL buffer system with finite maximum buffer timeis studied. By means of virtual waiting time (VWT) approach the distribution function forVWT is related to the one for waiting time in the QDB model. Then loss probability can beanalytically figured out and the minimum loss probability can be predicted as a consequencefrom extremum theory. The optimal normalized delay granularity is obtained for the cases ofboth variable-length and fixed-length packets. The study in this chapter can be used in FDLbuffer structure designing.In Chapter 6 a stochastic load-balancing scheme in OPS networks is proposed. First, astochastic model for packet transferring between two adjacent nodes is studied. The trans-ferring probability is obtained from stochastic performance measures of the neighbor node.Second, a load-balancing scheme is presented by using the transferring model. This schemecan decrease contention probability in OPS networks and achieve a tradeoff between de-creasing node state update cost and increasing network resources utilization efficiency.