Securing Public Cloud Networks with Efficient Role-based Micro-Segmentation
summary
Introduces an end-to-end system, ZTS, that infers roles from large flow logs using contrastive learning with sparse labels to reach average ARRI ≈ 0.77 on real deployments, and builds a two-stage pipeline (scalable pre-processing + SQL-based batch stage) that processes ~39M records/hour while making graph generation cost ≈0.05% of VM spend. Implementation outperforms a tuned Apache Flink baseline—up to 21.5× more cost-efficient than Apache Flink and 7.5× faster on the evaluated workload.
AI-written summary · about
chapters (10)
- 0:00Cloud network security challenges
- 1:19Ring fencing vs micro-segmentation
- 3:27Large-scale micro-segmentation challenges
- 4:32Role inference algorithm limits
- 6:20Contrastive learning for role inference
- 9:51ZTS algorithm design
- 10:33Role inference results
- 11:39Cost-effective graph pipeline
- 14:57Flink streaming comparison
- 16:20Conclusion
related talks
49mA Real-World Example of Using Network Behaviour to Detect Cyber-Compromise
49mRelational Observability for Cloud-Native Security and Data Science
25mReal-Time Fraud Detection at Scale
16mAnalysing Real time Activity Streams for Security
29mBuilding Resilient AI Platforms: Financial Fraud Detection
17mPrivateEye: Scalable and Privacy Preserving Compromise Detection in the Cloud
32mFrom Edge To Insight: Cybersecurity Intelligence With Kubeflow...
24mReal-Time Streaming Segmentation on Azure