ATM switche fabrics based on the cell extracting banyan networks

Abstract

An ATM switch must handle various kinds of traffics simultaneously at very high speed. And it should maintain low cell loss rate. Therefore new switch architectures are needed. In this thesis two switches fabrics suitable for the large-scale ATM switches are proposed and their performances are analyzed. A testing methodology for highly reliable implementation is also proposed. Deflection routing and cell extraction are used to increse the throughput of banyan networks. CEBN-2 and CEBN-4 are proposed networks. CEBN-2 is bulit from $2 \times 2$ switching elements, and CEBN-4 is bulit from $2 \times 4$ switching elements. Balancing and balancing extraction function is devised for CEBN. Analytical throughputs. But a single CEBN has too much cell loss rate to be used as an ATM switch. So several CEBNs are cascaded to make an ATM switch fabric. Applying the analytical models to the cascaded CEBNs, we can know which cells have the highest probability for each switching element. When those cells are extracted, minimum cell loss is possible. It is an optimal extraction funtion. Cell loss rates of CEBN switch fabrics with linear, balancing and optimal extracion are computed and compared with other fabrics. CEBN switch fabrics are very complex system and should be highly reliable. Implementation of such systems require parametric fault test. A test generation algorithm is developed for intra-gate bridge faults. Experimental results on the ISCAS 85 circuits show that it can cover more faults with less time.