| This thesis is concerned with both physical and networking aspects of packet switching with very high throughput.; We discuss a new network design that is based on a TDMA technology comprising the terahertz optical asymmetric demultiplexer (TOAD) and rapidly tunable delay line (RTDL), both developed by professor Paul Prucnal and his group. Due to the good perfoiniance of RTDL, this technology provides efficient switching on a packet-by-packet basis with high throughput. We propose further “mining” of the optical bandwidth in a TDMA optical network by using a spectrum encoding technique demonstrated by professors Warren Warren and Karen Bergman and their groups. In a summary, our network design is based on the experiments conducted in Princeton University laboratories. We examine possible modulation schemes in the proposed network.; We also propose medium access control (MAC) protocols that efficiently use the switching capacity provided by the above technology. A new feature of the physical layer is the increased number of channels that users may select for transmissions. We discuss two scalable protocols: an adaptive fixed assignment protocol (AFAP), and a protocol with reduced control information (PRCI). AFAP is a hybrid of random access and fixed assignment access schemes, and, therefore, provides low latency under light traffic load, and high efficiency under heavy traffic load. In PRCI, users contend for the channel reservations through the short reservation slots. A fair access of the channels is ensured by using the global parameters stored by all users. PRCI reduces the transmission overhead in a comparison with the previously proposed protocols, yet it achieves an excellent performance.; Finally, we propose a protocol for an electronically controlled optical switch. Unlike in a single-hop network, we assume centralized control. The proposed protocol is termed round-robin greedy scheduling (RRGS). RRGS satisfies stringent timing requirements by using a pipeline technique. Consequently, the delay in RRGS is non-negligible for the light traffic load, but it does not increase significantly even for the traffic load as high as 0.9. RRGS does not leave the input/output pairs unmatched even for a capacity of several Tb/s. |
