Channel Allocation , Power Budget And Bit Error Rate

Recent advances in computer processing speeds have resulted in a parallel processing environment in which the interconnection networks themselves are the limiting factor in terms of performance. One approach to circumvent this limitation is to arrange processing nodes in a hierarchical manner. In a hierarchical network, nodes are arranged in clusters of one or more nodes and these clusters are connected together at one or more higher levels. HORN 1] uses optically connected rings as the basic building blocks and connects these rings using an optical tree. Hierarchical topology, optical interconnects and unity diameter in HORN lead to a high throughput, low latency, scalable network. Two key issues facing HORN are addressed in this paper: dynamic channel allocation (DCA) and optical power budget (OPB) including bit error rate (BER) and signal to noise ratio (SNR). Four approaches to DCA are evaluated and a design trade-oo is performed between them. An alternative method is proposed that exploits the multiple paths between nodes at diierent levels in a hierarchical network. The evaluation of OPB shows that HORN is feasible and practical when optical ampliication is used at the initial signal insertion point for transmissions at higher levels in the hierarchy. The BER evaluation show that messages can be transmitted in HORN with very little loss of data.