Keskeiset käsitteet
DL2Fence introduces a novel framework utilizing Deep Learning and Frame Fusion for enhanced detection and localization of Denial-of-Service attacks in large-scale NoCs.
Tiivistelmä
1. Introduction
- NoC architecture in MPSoCs integrates various IPs like CPUs, GPUs, and memory controllers.
- High-intensive computation and communication tasks make systems vulnerable to DoS attacks.
2. DoS Attacks in NoCs
- Flooding attacks aim to exhaust resources, degrade performance, and increase power consumption.
- Traditional approaches monitor network traffic indicators for detection.
3. Proposed Framework
- DL2Fence utilizes Deep Learning models for classification and segmentation to detect and localize DoS attacks.
- Multi-Frame Fusion technique enhances the accuracy of victim localization.
4. Features and Model Selection
- VCO feature excels in detection while BOC feature performs well in both detection and localization.
- CNN models with minimal architecture are chosen for efficient performance.
5. Experiments and Results
- Detection accuracy ranges from 0.93 to 0.99, with precision close to 1.
- Localization accuracy varies from 0.91 to 0.98, with precision around 0.99.
- Hardware overhead decreases by 76.3% when scaling from 8x8 to 16x16 NoCs.
6. Conclusion
- DL2Fence achieves high detection and localization accuracies with reduced hardware overhead, making it scalable for larger NoCs.
Tilastot
DL2Fenceは、DoS攻撃の検出とローカライゼーションを向上させるためにDeep Learningとフレームフュージョンを利用する新しいフレームワークです。
DL2Fenceは、16x16メッシュノードで95.8%の検出精度と91.7%のローカライゼーション精度を達成します。
ハードウェアオーバーヘッドは、NoCサイズが8x8から16x16にスケーリングする際に76.3%減少します。