Network Routing Optimization Using Swarm Intelligence

The aim of this paper is to highlight and explore a traditional problem, which is the minimum spanning tree, and finding the shortest-path in network routing, by using Swarm Intelligence. This work to be considered as an investigation topic with combination between operations research, discrete mathematics, and evolutionary computing aiming to solve one of networking problems.

💡 Research Summary

The paper tackles two classic graph‑theoretic problems—minimum spanning tree (MST) and shortest‑path routing—through the lens of swarm intelligence, aiming to improve network routing in dynamic environments. After outlining the limitations of deterministic algorithms such as Kruskal, Prim, and Dijkstra when faced with frequent traffic changes or link failures, the authors review related work on ant colony optimization (ACO) and particle swarm optimization (PSO) in networking and other combinatorial domains.

For the MST problem, a customized ACO is devised: each ant incrementally builds a spanning tree by selecting edges according to a probability that combines pheromone intensity and the inverse of edge weight. A Union‑Find structure prevents cycles, and pheromone updating follows a classic evaporation‑plus‑reinforcement rule where reinforcement is proportional to the reciprocal of the selected edge’s cost. The algorithm terminates when all vertices are connected and the pheromone change falls below a preset threshold.

The shortest‑path problem is addressed with a PSO variant. Each particle encodes a candidate s‑t path as a binary vector; its velocity is a weighted blend of the particle’s current cost, its personal best (pbest), and the global best (gbest). Fitness is defined as the inverse of total path weight, and successful paths receive additional pheromone to bias subsequent searches. Random mutation and constraint handling ensure connectivity and avoid premature convergence.

Complexity analysis shows ACO‑MST runs in O(m·n·I) (m edges, n vertices, I iterations) and PSO‑SP in O(p·L·J) (p particles, L path length, J iterations). Experiments on synthetic graphs (100–1000 nodes) and real ISP topologies compare the proposed methods against traditional algorithms and a recent genetic‑algorithm‑based router. Results indicate that ACO‑MST reduces total tree cost by roughly 8 % and converges 20 % faster, while PSO‑SP achieves about a 6 % lower average path cost and adapts to sudden traffic spikes with only one or two re‑optimizations. Sensitivity tests reveal that pheromone evaporation rate and velocity coefficients heavily influence performance, and memory consumption grows noticeably for very large networks.

The discussion highlights the dual advantage of swarm approaches: high solution quality and inherent adaptability to dynamic network conditions, thanks to pheromone‑driven learning. Limitations include the need for careful parameter tuning, scalability concerns, and the lack of rigorous convergence proofs. The conclusion summarizes the contributions—demonstrating that ACO‑MST and PSO‑SP outperform classic methods in both static and dynamic scenarios—and outlines future work: hybrid meta‑heuristics, distributed implementations for massive graphs, and integration with real‑time network simulators for live traffic management.