Media Access Protocol Selection for Hierarchical Optical Ring Interconnection Network (HORN)

Optical fibers offer high potential for interconnection networks (INs) for multiprocessing and multicomputing due to their ability to carry a large number of wavelengths at high transmission rates but current technology limits their use to only a few dozen wavelengths. One approach to circumvent this limitation is to arrange processing nodes (PNs) on fibers in groups and spatially separating those clusters. Grouping those clusters together in a higher level structure, or hierarchy, allows any node to communicate with any other in a large network while using only a few wavelengths. One potential hierarchy presented previously [1], Hierarchical Optical Ring Interconnection Network (HORN), uses a clustered hierarchical approach in which clusters consist of a set of PEs optically interconnected and these clusters are optically connected by higher level rings. As such, a highly scalable architecture is achieved, overcoming the fixed number of available wavelengths through spatial reuse of the same set of available wavelengths. Under hierarchical topologies like HORN nodes are separated spatially and spectrally but these two methods only reduce substantially the number of collisions of data from multiple sources, they can’t completely prevent them. An additional method, or protocol, is needed to achieve single-hop, non-blocking, and collision-free communication with minimal delay and high throughput. There are a wide variety of potential protocols for this type of application but many are inappropriate or unacceptable for one reason or another. Preliminary analysis narrowed consideration and indepth analysis down to five promising protocols: time division multiple access (TDMA), TDMA with arbitration, FatMAC [2, 3], DMON [4], and token hierarchical optical ring network (THORN). These are analyzed in terms of average delay, system throughput, node complexity, bandwidth and scalability. They are also assessed more qualitatively with respect to their abilities to accomodate bursty communication traffic. The first four protocols are documented in the literature but the fifth, THORN, was developed expressly for HORN and is an adaptation of the token ring and DMON protocols. Not suprisingly, the protocols that have variable cycle lengths (FatMAC, DMON, and THORN) are found to be superior in terms of delay than the protocols with fixed cycle lengths (TDMA and TDMA/arbitration). While all the protocols support the scalability objectives of HORN, THORN is shown to have the lowest delays with minimal increases in node and algorithm complexities. While this protocol analysis is performed for a HORN only, the results should be general enough to apply to other hierarchies.