Diversity has been widely recognized as a critical factor for robustness in networks. In order to resist isolations due to link and node failures, we need to route along paths which are both span-disjoint (physically disjoint) and node-disjoint. In this way we can design the network for fault tolerance or graceful degradation. We demonstrate the procedure to obtain span-disjoint paths. This procedure may be applied to an arbitrary network topology, without restrictions on the span sharing configurations. In those cases where two completely disjoint paths do not exist, the procedure still tries to find a pair of paths which are as diverse as possible. The solution path pair we obtain is not guaranteed to be optimal, but gives us an optimal or near-optimal design, relatively quickly. The algorithm also accepts parameters which can be used to specify the cost-robustness tradeoffs.

The authors describe a simple adaptive routing scheme for datagram (connectionless) and virtual circuit (connection-oriented) transmission that relieves congestion resulting from nonuniform traffic patterns and network failures. The authors describe a fixed-routing algorithm for dedicated channel ShuffleNets. Based on the fixed routing algorithm, an adaptive routing scheme for datagram transmission is presented followed by performance results for uniform and nonuniform traffic patterns and fault tolerance. The adaptive routing of datagrams uses only the local queue size information available at the network interface units (NIUs) and redistributes the load as congestion develops. Since datagrams are individually routed through the network, they may not arrive at their destination in the order they were generated and may need to be resequenced. The authors compute an upper estimate on the resequencing buffer size for stream traffic. A virtual circuit version of the adaptive routing algorithm eliminates the need for resequencing buffers.>

The performances of centralized, fast packet switches based on the ShuffleNet and on multistage banyans are compared. Two cost criteria are used, viz., the number of crosspoints and the number of pin-limited ICs. The analytic comparison of the two networks built with arbitrary size infinite buffered nodes shows their cost performance to be similar. The analysis also provides the optimal node sizes for both the topologies. The best throughput per crosspoint is obtained with small nodes, but the best throughput per pin-limited IC is obtained with the largest possible nodes that can be built on an IC. With the IC criterion, for both the networks, small nodes are found to give small minimum delays (inside the network). With finite buffered nodes and a simple back-pressure flow control (like the one for the banyan network), the ShuffleNet based switch can have problems of congestion. It is also prone to deadlocks.>

Approximate Markov chains for the performance of banyan networks built with basic switches of arbitrary size containing input queues, output queues or a combination of input buffers and output queues are presented. These chains are also extended to model banyans where each link is replaced by a number of parallel links. Comparison with simulations indicates that the chains are reasonably accurate for practical network loads. It is shown that banyans built with switches containing a combination of single input buffers and output queues outperforms banyans built with switches containing only input queues or only output queues. Banyans built with larger nodes are shown to outperform banyans built with smaller nodes. It is shown that banyan-based networks with multiple links can be designed to have extremely low blocking probabilities.>