◄ WORLD V · SONNY 5DART 217 · a helldive at the net

THE GOSSIP PROTOCOL rumor reaches everyone in log(n) rounds

To spread an update across thousands of nodes without a central broadcaster, let it spread like a rumor: each round, every informed node tells a random other. The count of informed nodes roughly doubles each round, so the whole system learns in about log₂(n) rounds — and the redundancy makes it robust to lost messages and dead nodes. This is epidemic dissemination: the same math as a spreading infection, turned into a feature.

THE TECHNIQUE informed count doubles per round → full in ⌈log₂ n⌉ rounds

Seed one node and watch the rumor double each round until the whole cluster knows — the round count grows only logarithmically as you scale the cluster to thousands: live demo


HISTORY & CREDIT Demers et al. (Xerox PARC), 1987

“Reaching every node means a broadcast from the center — a bottleneck and a single point of failure.” — gossip has neither. No node is special, no node must know the whole membership, and the exponential spread reaches everyone in logarithmic time even as nodes fail — the losses are just re-covered by the next round's redundancy. cited

1987 · Demers et al. (Xerox PARC) — “Epidemic Algorithms for Replicated Database Maintenance”: anti-entropy and rumor-mongering to keep Clearinghouse replicas in sync.
1927 · Kermack & McKendrick — the mathematical theory of epidemics; the doubling curve is the simple (SI) epidemic phase, borrowed wholesale (the general SIR synthesis is Bailey's 1957 textbook).
now · Cassandra, DynamoDB, Consul, and blockchain mempools all gossip to stay converged.

The virtue is that no one is in charge and no one holds the full map, yet everyone converges — the same property that makes an epidemic unstoppable makes a gossip layer unkillable. Demers 1987

RECOMMEND FOR I-13 full dissemination in ⌈log₂ n⌉ rounds, computed

On the canonical compiler, a rumor seeded in one node of a 1000-node cluster reaches all of them in 10 rounds — exactly ⌈log₂1000⌉:

$ i13 run gossip.i13 # seed 1, doubles per round, n=1000 rounds = 10 -- ceil(log2 1000); 1,2,4,8,...,1000 -> everyone knows
Recommend: the gossip protocol is LIT for I-13 — verified a single seed saturates a 1000-node cluster in 10 rounds, the doubling of the epidemic curve capped at n. Logarithmic reach with no coordinator and no membership list — and ten million nodes would take only 24 rounds. Scale-free dissemination, computed.