Routing Scalability, Convergence, and Stability: A Comparative Analysis of Network-layer Design Trade-offs (2016–2026)

Abstract

This study investigates the trade-offs among routing scalability, convergence, and stability across traditional routing protocols (BGP, OSPF, and IS-IS), Software-Defined Networking (SDN), and emerging AI-driven routing approaches. Following PRISMA-inspired guidelines, we systematically reviewed 1,247 records (2016–2026) from IEEE Xplore, ACM Digital Library, Scopus, Web of Science, and arXiv. After deduplication (n=312) and title/abstract screening, 156 full-text articles were assessed, yielding 30 primary studies for detailed analysis. Evaluation of metrics included convergence time (ms–min), scalability limits (nodes/prefixes), stability characteristics (route flapping frequency), and control-plane overhead (CPU/memory utilization). Results demonstrate that no routing architecture simultaneously optimizes all three dimensions; for example, BGP convergence ranges from 30 seconds to 15 minutes for >900k prefixes, whereas OSPF/IS-IS achieves sub-second convergence (150ms–2s) but faces practical scalability ceilings at ~1,000 nodes per area due to O(n²) flooding overhead. SDN controllers process 1k–10k flows/sec under centralized architectures, while hybrid SDN-BGP designs offer the most balanced performance but lack standardized integration interfaces. Unlike prior surveys, this work introduces a unified comparative evaluation framework that enables direct comparison across distributed, centralized, hybrid, and AI-assisted routing paradigms. Findings are constrained by heterogeneous experimental methodologies and limited production-scale evidence for AI-driven routing systems.