Mitigating request flooding attack in named data networking using federated learning.

Area of Science:

• Computer Science
• Network Security
• Distributed Systems

Background:

• Named Data Networking (NDN) architectures are vulnerable to Interest Flooding Attacks (IFAs).
• Existing IFA mitigation methods suffer from high false positives or significant overhead.
• Adaptive and efficient solutions are crucial for NDN security.

Purpose of the Study:

• To propose FL-IFAshield, a novel federated learning framework for adaptive IFA mitigation in NDN.
• To enhance the security and performance of NDN against Interest Flooding Attacks.
• To provide a resource-efficient solution for constrained edge computing environments.

Main Methods:

• Developed FL-IFAshield, integrating dynamic Poisson-EMA thresholding for detection.
• Implemented entropy-aware federated aggregation for non-IID traffic.
• Incorporated Byzantine-robust mechanisms with differential privacy guarantees.
• Evaluated on a 100-router FIT/IoT-LAB testbed.

Main Results:

• Achieved 93.1% F1-score for attack detection with only 5% false positives.
• Demonstrated average end-to-end latency of 28 ms.
• Maintained over 90% legitimate Interest Satisfaction Ratio under collusive attacks.
• Showcased minimal computational overhead (<9% CPU utilization).

Conclusions:

• FL-IFAshield offers a significant improvement in NDN security against IFAs.
• The framework provides superior accuracy and lower communication overhead compared to existing approaches.
• FL-IFAshield achieves an optimal balance of precision, low latency, and resource efficiency for edge computing.