Pretty Good BGP: Improving BGP by Cautiously Adopting Routes Josh Karlin, Stephanie Forrest, Jennifer Rexford IEEE International Conference on Network Protocols 2006
Outline What are current BGP security issues? What is PGBGP trying to solve? How does PGBGP solve it? How good is PGBGP? How bad is PGBGP? Shall we use it?
What are current BGP security issues? BGP4 (RFC1771) –Inter-domain routing, internet core –Link state protocol, distributed system Vulnerabilities –No encryption: eavesdropping –No timestamp: replaying –No signature: man-in-the-middle
What are current BGP security issues? Examples
What is PGBGP trying to solve? General requirements of a good solution –BGP is widely deployed: don’t modify the protocol –Route’s resource is stretched thin: don’t consume too much resource –ISPs are conservative: incremental deployable –ISPs are greedy: show good results!
What is PGBGP trying to solve? Prefix hijack –Shorter AS_PATH (man-in-the-middle) –MOAS (multiple origin AS)
How does PGBGP solve it? Basic idea –Suspicious Cautious –Use historical prefix-origin records –Damping suspicious prefix-origin announcement for 24 hours –Human investigation –Good for prefix/sub-prefix hijacks
How does PGBGP solve it? Algorithm History period – h hours clean Suspicious period – s hours quarantined Move h forward remove staleness, get freshness Parameters sensitivity h = 10 days : short FP, long repeat slips s = 24 hours : human response time
How does PGBGP solve it? Prefix Hijacks: conflict w/ unknown origins Sub-prefix hijacks: Conflict w/ known origins [Q1]?
How does PGBGP solve it? Mitigation –Avoid suspicious routes: lower preference Sub-prefix: quarantine, choose neighbor not having the suspicious routes (not really helpful) Never seen prefix / super-prefix will be adopted –Convergence consideration Obey relationship-based policy Dampened as if not announced
How good is PGBGP? Simulation –18,943 ASes, average 4 links per AS-AS –Simulator w/ policy-based routing –Deployment strategries: random -- p core+random (15 degree+) + p –500 attacks per setup –Parameters: h = 3, s = 1 –Day 1, O; Day 2 O’
How good is PGBGP?
Conclusion: pretty good –Core + random deployment, 90%+ effective –Incrementally deployable –Out-of-core computation possible –Centralized computation possible –Overhead is small, real time possible –Extension: IAR (internet alert registry)
How bad is PGBGP? Limitations: –FP: Origin change, multi-homed –DoS + no other choice –lucky slips –Man-in-the-middle (put itself in AS_PATH) Conclusion: not to bad
Shall we use it? Critiques for the paper –FP delay propagation: –Model human correction rate with prob. p1, FP rate p2 … –Some analysis is not thorough (e.g. Fig 3) –Undeployed ASes at risk (good & bad) –Distributed/Co-operated version Conclusion: try if you like
Shall we use it?
Questions Ask me: Josh Karlin: Interested in security research?