Many large scale core data networks are now based on the new generatio
n of intelligent time division multiplexers. These allow permanent vir
tual circuits between any network nodes and automatic rerouting of los
t connections should link failures occur. The design requirements for
such networks are very different to those more usually associated with
standard packet switched network arrangements. For a given traffic ma
trix the number of possible network topologies, link capacity selectio
ns, and circuit routings, even between a small number of nodes, is rel
atively large. When taking into account the additional requirements of
optimising the overall link capacity of a network and ensuring its re
silience to failure by specifying link disjoint primary and backup pat
hs, the design complexity becomes very large. It is also necessary to
use commercial link costs for all the link specifications likely to be
selected from the various PTT service providers to ensure realistic d
esign results. The following paper describes the modules of a heuristi
c method that has been successfully used on a standard personal comput
er to provide minimum cost optimum designs based on the described netw
ork characteristics. The heuristics incorporate a speed versus accurac
y trade-off factor so that rapid approximate designs may be examined f
or differing traffic conditions. (C) 1997 Elsevier Science B.V.