Rumors that reach everyone
In a large cluster, having one node tell all others directly is fragile and slow. Gossip dissemination instead has each node periodically pick a few random peers and share what it knows. Those peers pass it along, and like a rumor the information reaches the whole cluster quickly.
Why it scales
Each round, the number of nodes that know a fact roughly multiplies. Starting from one informed node, the count grows exponentially, so even a huge cluster converges in a number of rounds proportional to the logarithm of its size.
- Every node runs a periodic gossip round.
- It selects a small set of random peers.
- It exchanges state, and both sides merge what they learned.
Robust by design
Gossip tolerates failures gracefully. No single node is critical, since information flows along many redundant paths. A few dropped messages just delay convergence slightly rather than breaking it. The cost is that gossip is eventually consistent, nodes briefly disagree until the rumor finishes spreading.
Key idea
Gossip dissemination spreads information by having each node share with a few random peers each round, reaching the whole cluster in logarithmic time with strong fault tolerance but only eventual consistency.