Abstract
We investigate design issues of optical networks in light of two conflicting criteria: throughput maximization (or equivalently, congestion minimization) versus delay minimization. We assume the network has an arbitrary topology, the flow can be split and sent via different routes, and it can be transferred via intermediate nodes. Tabu search heuristic is used to compare solutions with different weights assigned to each of the two criteria. The approach is tested on a benchmark data set, the 14-dimensional NSFNET T1 network with traffic from 1993. The results suggest that: (1) some connectivity matrices are quite robust and desirable regarding both criteria simultaneously; (2) forcing minimization of total delay unconditionally can result in significantly inferior throughput. Some decision strategies are outlined.