BayWatch: Practical Internet-Scale Topology Monitoring with Dynamic Bayesian Estimation
summary
System that builds a layered model of TTL hops and uses a Dynamic Bayesian Network to infer actual load‑balancing reply (LBR) distributions rather than assuming uniform splits, implemented as BayWatch and run across ~14.46M targets. It estimates per‑hop transition probabilities, discovers 2.4× more nodes and 2.8× more links than D‑Miner in internet measurements, cuts probing by 6.3× while keeping completeness loss under 1%. It also detects real outages.
AI-written summary · about
chapters (14)
- 0:00BayWatch introduction
- 1:25Internet-scale topology monitoring challenges
- 2:31Multipath detection algorithms
- 3:56Limitations of uniform load balancing
- 6:46Non-uniform load balancing impact
- 7:49BayWatch Bayesian network model
- 9:12BayWatch parameter estimation
- 10:37BayWatch bootstrap mode
- 12:22BayWatch evaluation
- 13:27BayWatch accuracy vs prior methods
- 14:31BayWatch vs D-Minor discovery
- 15:12BayWatch predictive mode
- 16:37BayWatch key takeaways
- 17:10Q&A
related talks
23mNetBouncer: Active Device and Link Failure Localization in Data Center Networks
46mLessons learned from monitoring a large production network
32mActive Fault Finding in Networks
20mDiamond Miner: Comprehensive Discovery of the Internet's Topology Diamonds
17mReasoning about Network Traffic Load Property at Production Scale
17mNetHint: White-Box Networking for Multi-Tenant Data Centers
30mValidating Network High Availability With Batfish
30mKeeping the Balance: Internet-Scale Loadbalancing Demystified