Simplifying global-scale strong consistency
summary
A prototype global-commit protocol uses token-passing logical clock and event-sourced consensus groups to assign atomic commit timestamps locally and associate child groups to parent groups for a total order. It relies on optimistic locking, per-record registered/discovered lists for conflict resolution, and deterministic single-threaded replay from event logs. Implementation exposes a JDBC-compatible database, targets SQL snapshot-serializability, and claims sub-millisecond locality latencies with a worst-case global round ≈250 ms; software and consulting are already available.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 6:55Theory
- 11:45Model
- 16:43Token Passing
- 27:52Parent Group
- 36:02Questions
- 42:45Updating
- 44:46Commit
- 47:08Remarks
- 49:57Consequences
- 56:31Faults failures
- 59:17Multiple processes
related talks
37mGeo-Replicated Transactions in 1.5RTT
14mBolt-On Strong Consistency: Specification, Implementation, and Verification
42mBeyond Blockchain: Convergent Consensus
40mImplementing Distributed Consensus
35mDon't Give Up on Serializability Just Yet
24mStronger Semantics for Low-Latency Geo-Replicated Storage
40mwant don't You consistency no stinking
48mThe Accord Consensus Protocol, how to do ACID transactions at scale