Learning-Enhanced High-Throughput Pattern Matching Based on Programmable Data Plane
summary
Trochilus converts existing rule sets into a byte-level recurrent neural network BRNN, distills that model into a hardware-friendly soft multi-view forest (SMF) via semi-supervised knowledge distillation, and encodes trees into match-action tables with an entry clustering algorithm plus an overlapping sliding window. The system claims multi-Tbps throughput, reduces TCAM entries by an order of magnitude compared to direct DFA mappings, improves accuracy about 10% after fine-tuning, and automatically rebounds accuracy during weeks of real traffic updates.
AI-written summary · about
chapters (12)
- 0:00Research introduction and motivation
- 1:46Current pattern matching limits
- 3:33Design goals proposed system
- 4:16Pattern rules to BRN models
- 5:40Learning modules on data planes
- 7:50Semi-supervised knowledge distillation
- 9:39Tree encoding and clustering
- 11:54Sliding window mechanism
- 13:19Evaluation setup and baselines
- 14:46Throughput and storage results
- 15:31Long-term accuracy maintenance
- 16:58Conclusion and future work
related talks
14mBrain-on-Switch: Towards Advanced Intelligent Network Data Plane via NN-Driven Traffic...
18mLeo: Online ML-based Traffic Classification at Multi-Terabit Line Rate
20mContra: A Programmable System for Performance aware Routing
24mHyperscan: A Fast Multi-pattern Regex Matcher for Modern CPUs
15mAccelerating Rule-matching Systems with Learned Rankers
10mChip floorplanning with deep reinforcement learning
17mTraffic Engineering for AI Training Networks
17mSPLIDT: Partitioned Decision Trees for Scalable Stateful Inference at Line Rate