Low-cost crossed probing path planning for network failure localization

Links inevitably fail in expanding networks, leading to user-perceived service interruptions. To localize link failures quickly and accurately is thus essential, and route-aware active probing makes it possible. Given the limited routing capacity and high traffic overhead, cross verification enables a light-weight probing scheme using reachability verification for distinct subsets of crossed paths to pinpoint the exact faulty links. Aiming to quickly optimize the crossed path design, we propose pruning genetic algorithm (PGA), which builds a pruning module on top of genetic algorithm to consistently produce high-quality solutions across various networks and avoid slow convergence in an exponentially large solution space by eliminating redundant paths. PGA also introduces extra repair operations to guarantee solution feasibility after crossover and mutation. Our experimental results on real-world network topologies demonstrate that PGA achieves a significant reduction of 23.0% to 58.3% in probing cost and 23.0% to 45.3% in forwarding cost in seconds or even milliseconds compared to its counterparts.