Abstract
Interconnection-network design requires a trade-off among reliability, redundancy, connectivity, and implementation cost. This paper presents an Improved Genetic Algorithm (IGA) for reliability- and cost-constrained topology optimization. The network is modeled as a probabilistic graph whose processors are vertices and communication links are candidate edges with cost and reliability attributes. The IGA combines reliability-guided initialization, adaptive crossover and mutation, tournament selection, elitism, and feasibility-preserving repair. To address the limitation of a link-average reliability surrogate, the revised study uses estimated all-terminal reliability as an independent validation metric. Experiments are performed on standard Mesh, Torus, and Hypercube topologies with 64 and 256 nodes. The IGA is compared with a conventional Genetic Algorithm and a minimum-cost spanning-tree-based heuristic over five independent runs per configuration. Monte Carlo reliability estimates use 5,000 link-state trials for each optimized topology. The results show that the IGA provides competitive reliability under the same cost budget, with the clearest advantage on the 256-node Mesh and on the 64-node Hypercube, while the conventional GA is slightly better on some Torus instances. The results therefore support the usefulness of reliability-guided evolutionary search without claiming universal superiority.
Sushil Kumar Dwivedi1*, Rakesh Kumar Katare2
1Research Scholar, Department of Computer Science, APS University, Rewa-486003, M.P., India
2Professor, Department of Computer Science, APS University, Rewa-486003, M.P., India