hesela.devPartial synchrony
Partial synchrony models timing bounds that are unknown in advance or become guaranteed only after an unknown stabilization time.
Dwork, Lynch and Stockmeyer distinguish always-existing unknown bounds from known bounds that hold after an unknown global stabilization time. Communication delays and relative processor speeds are separate model dimensions. [1]
A protocol satisfying the eventual-bound contract preserves safety despite arbitrary delays; guaranteed termination also depends on eventual timing conditions and its fault assumptions. Stabilization is not a client-visible completion deadline. [1]
Raft illustrates the distinction: log safety does not depend on timing, but maintaining a leader and making progress do. Its timing guidance is not a measured service guarantee. [2]
Relations
- related
- FLP impossibility, Failure detector, Raft
Citations
- Dwork, Lynch and Stockmeyer. Consensus in the Presence of Partial Synchrony. JACM 35(2), 288-323, 1988; sections 1.2 and 5.
- Ongaro and Ousterhout. In Search of an Understandable Consensus Algorithm, extended version (20 May 2014), section 5.6.
Sources & caveats
- DLS (1988), sections 1.2 and 5, defines the timing models. Raft section 5.6 provides a protocol-specific example, not a universal fault threshold or a deployment measurement.