This dissertation studies the packet scheduling problem in unidirectional optical buses, i.e. light-trail and opportunistic hyperchannel, which is also a medium access control problem. In a unidirectional optical bus, a collision is arisen when more than one packet are transmitted simultaneously and the bandwidth is wasted when there is a collision in the system. Further, upstream nodes (resp. downstream nodes) may experience some inherent high (resp. low) priority for the one directional transmission of optical signal. Therefore, system throughput, fair service among competing nodes, and/or Quality of Service (QoS) are the issues integrated with the packet scheduling problem. To provide efficient and practical solutions, we take system architecture and algorithm co-design as the theme of the dissertation. We summarize the results obtained in this dissertation as follows. (1) By making some practical assumptions, we show the equivalence of the packet scheduling problem with assured QoS in a distributed unidirectional optical bus and a centralized multiplexer. (2) An innovative unidirectional bus architecture, named an opportunistic hyperchannel, which takes advantage of the recent development of ultra-fast optical shutter where the state change of an optical shutter can be done within femtoseconds (10^-15 second). (3) A distributed dynamic scheduler for an opportunistic hyperchannel, called 1-persistent Carrier Sense Multiple Access/Collision Free protocol, setting goal to maximize system throughput with fair servicing among competing nodes, of which the time complexity is O(1) per packet. (4) A distributed dynamic scheduler for an opportunistic hyperchannel, named min-SrcRR, on purpose of maximizing system throughput with QoS assurance. With parallel receiving and transmitting, the time complexity of minSrcRR is O(1) per packet. (5) A centralized dynamic scheduler for an opportunistic hyperchannel, identified as PMIS, which can achieve better system throughput with QoS assurance compared with minSrcRR by consuming more channel resource. In PMIS, one round calculation can be carried out in O( k · log n · log m) time, where k is the number of transmission decisions made in the PMIS computation, n is the number of nodes in the opportunistic hyperchannel, and m is the number of waiting packet transmission requests in the system. (6) A generalization of linear opportunistic hyperchannel, named non-linear opportunistic hyperchannels. There are two interesting non-linear opportunistic hyperchannel architectures: gathering tree and multicast-and-select tree. The corresponding implementation of 1-persistent CSMA/CF protocol in gathering tree is provided.

The performance of these schedulers and designs is analyzed by theoretical proofs and/or simulations.