| Summary: | 博士 === 國立清華大學 === 電機工程研究所 === 84 === A new routing algorithm is proposed in this dissertation for
large communication networks with dependent unreliable
components. The entire network is first partitioned into a
multi level environment without assuming that all nodes in the
same cluster be directly connected. An auxiliary network
associated with each cluster is then constructed. Based on the
auxiliary networks, a new neural network model capable of
handling dependent component failures of communication
networks, is used to calculate the shortest path for each
origin-destination pair. Therefore, hierarchical shortest path
tables can be established in parallel for each cluster of the
same level from bottom up. Finally, by the hierarchical
shortest path table, a new routing protocol finds the resultant
transmitting path. Simulation results indicate that the routing
path can be found not at the expense of the table size and the
storage space for each node. The proposed neural network model
is proven to have a stable solution. Moreover, useful upper and
lower bounds for the design parameters are derived to
facilitate their selection during implementations. The
algorithm''s performance for one-level and multi-level clustered
networks are analyzed. Results indicate that the shortest path
table size can be reduced and the storage space saving for each
individual node is approximately one-third of that proposed by
Baratz and Jaffe.
|
|---|
