| Summary: | A widely adopted approach for interconnecting local area computer networks (LANs) is to use MAC bridges, and building or site-wide backbone subnetworks. When using backbones to interconnect a large number of segments, however, it is important to ensure that the backbone itself does not form a bottleneck in the network. To overcome this problem, high-speed networks are normally used as backbones, especially with applications involving substantial cross-site traffic. To realise the full benefits of backbones, the overheads associated with relaying frames to and from the backbones must be minimised. In interconnecting high-speed LANs and backbones, the desired features of a bridge relaying/routing scheme are: minimum processing at bridges, optimal route selection between end-stations and transparency to end-stations and higher-layer protocols. Each of the two types of bridge routing schemes that have been standardised - transparent spanning tree, and source routing - do not fully meet all of these criteria. The former requires a large filtering database in all bridges that slows down their performance and the spanning tree algorithm does not make optimal use of bandwidth. The latter is considered more efficient because it requires simpler processing at bridges and derives optimum routes. It is not transparent to end stations, however, and route discovery overheads can be very high in complex topologies. This thesis describes research that addresses each of these limitations and also the results of a set of simulations relating to the performance of high-speed backbone networks. A modification to the existing source routing scheme, termed as reduced broadcast source routing, was developed to minimise unnecessary broadcast of route discovery frames.
|
|---|
