Partial synchrony model

The partial synchrony model is a theoretical framework for distributed systems that occupies the middle ground between the synchronous system model — where message delays are bounded by a known constant — and the asynchronous system model — where delays are finite but unbounded. In partial synchrony, the network is asynchronous most of the time, but there exist periods during which messages are delivered within known bounds. These periods of synchrony are not predictable, but they occur often enough to make progress possible.

This model is the theoretical foundation of practical consensus protocols like Paxos and the Raft algorithm. The FLP impossibility result proves that deterministic consensus is impossible in fully asynchronous systems; partial synchrony is the minimal assumption that makes consensus achievable. In practice, this translates to timeout-based protocols: nodes wait for responses, and if no response arrives within a timeout, they suspect the sender has failed. The timeout is a bet that the network is in a synchronous period. When the bet is wrong — when the network is genuinely asynchronous — the protocol may stall or trigger unnecessary leader elections. But when the bet is right, consensus proceeds.

The partial synchrony model captures something essential about real networks: they are not maximally adversarial. Routers fail, congestion spikes, and packets are lost — but not forever. The model acknowledges this reality without pretending that networks are predictable. It is the theorist's way of saying: we cannot guarantee when messages arrive, but we can guarantee that they arrive eventually, and that is enough.