Researchers at Nanjing University collaborated with China Mobile to develop a novel algorithm, BI-TE, which enables multi-domain networks to protect sensitive internal data while maintaining high bandwidth use, cutting cross-domain routing time by nearly a quarter. It addresses the growing need for internet service providers (ISPs) and enterprises to collaborate on traffic management without exposing their private network details.
Imagine several chefs, each guarding their secret recipes, needing to cook a joint banquet. Instead of sharing full recipes, they exchange taste-blurred samples. A trained food critic then learns to predict the authentic flavors from those samples and suggests a menu that delights everyone, without ever revealing any chef’s private ingredients. BI-TE works similarly with network data.
“By blending controlled noise with a Graph Neural Network, we’ve shown that network operators no longer have to choose between performance and privacy. BI-TE delivers nearly a 25% speed-up in cross-domain routing while keeping each domain’s secrets intact,” says Prof. Jingyu Hua.
Private Multi-Domain Networks Finally Talk Traffic: Smarter Path Prediction Boosts Collaboration
Modern networks span multiple administrative domains—think data centers, campus networks and wide-area links. Each domain often refuses to share its detailed topology or resource data, hampering efforts to optimize traffic and meet service guarantees. BI-TE’s blend of privacy protection and smart path prediction offers operators a way to collaborate more effectively, with potential benefits for cloud providers, telecom operators and policymakers looking to strengthen both network efficiency and data confidentiality.
BI-TE’s Results Unveiled: Key Findings on Speed, Privacy & Efficiency
The evaluation of BI-TE across both real-world and simulated environments reveals several noteworthy improvements:
- BI-TE’s privacy layer perturbs each domain’s reported topology and bandwidth figures using controlled noise, keeping internal layouts and usage patterns indistinguishable.
- A Graph Neural Network (GNN) model predicts end-to-end bandwidth use on these perturbed views, guiding the central controller to choose near-optimal paths.
- In tests on real and simulated network maps, BI-TE cut the time for multi-domain path calculation by 24.35%.
- Even with privacy perturbations, overall link utilization stayed around 90%, matching the performance of non-private methods.
- BI-TE also maintained fair allocation across flows, avoiding bandwidth hogging by any single traffic stream.
How It Works
Local controllers add small random changes (“noise”) to their network maps and available-bandwidth reports before sharing them. A central coordinator then runs its usual routing algorithm on this masked data. To compensate for the inaccuracies introduced by the noise, BI-TE trains a GNN that learns to predict actual bandwidth use from the noisy inputs. This allows the controller to pick efficient routes without ever seeing exact private details. The complete study was published in
Frontiers of Computer Science in January 2025 (https://doi.org/10.1007/s11704-024-40551-2).
Privacy and Performance Combined: Machine Learning-Driven Differential Privacy Powers Fast, Secure Inter-Domain Routing
BI-TE demonstrates that strong privacy safeguards and high traffic-engineering performance need not be mutually exclusive. By combining differential privacy techniques with machine learning, operators can effectively protect their network secrets while still achieving fast and efficient inter-domain routing.
DOI:
10.1007/s11704-024-40551-2