Gossip protocols for Eventual consistency
summary
Explains that rumor‑mongering gossip achieves convergence in O(log n) rounds and implements it in Rust with a Tokio-based runtime and a retry mechanism so lost messages are re-sent; tests run under Maelstrom+Jepsen with a fan‑out 3 gossip strategy across 10 simulated nodes and a partition nemesis; the run sends ~127,564 messages (≈426 per operation) and Maelstrom’s analysis shows correctness with retries and stale results without them.
AI-written summary · about
chapters (12)
- 0:00Gossip protocols intro
- 3:15What gossip means
- 6:22Why gossip protocols exist
- 13:25Rust gossip node
- 23:15Push-based gossip algorithms
- 24:58Push-pull protocols efficiency
- 25:59Gossip simulation orchestration
- 29:05Network partitions & fault tolerance
- 30:52Live demo with Maelstrom
- 33:21Where gossip is used
- 35:29Gossip limitations
- 36:31Q&A
related talks
37mLet's build our own message streaming platform
45mImportance of Transactional Metadata (Or How We’re Improving Upon Gossip)
37mGeo-Replicated Transactions in 1.5RTT
44mPolyglot Data
41mProgramming Distributed Systems
01h 01mLogic and Algebras for Distributed Computing: Applications and Open Questions
47mHow to scale a distributed (file) system
01h 02mConsistency and Agreements in Distributed Systems