On the Difficulty of Scalably Detecting Network Attacks Kirill Levchenko, Romanmohan Patruri, George Varghese Presented by: Yaxuan Qi, 2007.11.29
Outline Problem & Contributions Background Study of Attacks Conclusion SYN-flooding Port-scan Connection-Hijacking Fragmentation Conclusion
Problem Most network intrusion tools (e.g., Bro) use per-flow state to reassemble TCP connections and fragments A high speeds at network vantage points, some form of aggregation is necessary. a number of problems have scalable solutions. No clear proof that such per-flow state is required for many of these problems
Contribution Proves: Exposes assumptions that Many well-known intrusion detection problems (detecting SYN Flooding, Port Scans, Connection Hijacking, and content matching across fragments) require per-flow state. Exposes assumptions that need to be changed to provide scalable solutions to these problems; Concludes with some systems techniques to circumvent these lower bounds.
Background Deployment of NIDS Per-flow state Vantage point: deeper inside the network Cost-saving: number and management As close to the attacker as possible Fewer legitimate users are affected (??) Per-flow state Provide wire-speed detection Reduce false positive
Background Related work Vantage point also requires per-flow state However, high-speed devices rely on cache or on-chip SRAM Still smaller flow aggregation Load-splitters Expensive Also split attacks
Methodology Abstract Problem Formulation Example Lower Bound definition Example illustration Lower Bound Spatial complexity Proof (see appendix) Practical Implications Scalability
Ingress SYN-flooding
Ingress SYN-flooding
Ingress SYN-flooding
Ingress SYN-flooding
Ingress SYN-flooding
Egress SYN-flooding
Ingress Port-Scanning
Ingress Port-Scanning
Egress Port-Scanning
TCP-Hijacking
TCP-Hijacking
Fragmentation Detection
Fragmentation Detection
Conclusion
Questions?