The Accord Consensus Protocol: ACID transactions globally at any scale
summary
Claims a leaderless, single-network-round-trip consensus that delivers strict serializability and ACID multi-partition transactions at global scale using commodity clocks. Implemented as a standalone Accord library (~280 files, ~60k LOC) and integrated into Cassandra with only 27 files changed, work started in 2021 and targets Cassandra 5.1. Two new mechanisms are flexible fast-path electorates (shrink quorum per transaction) and a reorder buffer using Lamport-suffixed timestamps; common-case is one round trip, but failure modes cause graded degradation.
AI-written summary · about
chapters (12)
- 0:00Introduction to Accord
- 3:20Challenges in distributed data
- 7:22The need for ACID transactions
- 8:57Consistency and isolation levels
- 13:22Limitations of current methods
- 14:52Leader-based consensus
- 16:57Leaderless consensus protocols
- 20:38Comparative protocol analysis
- 25:02Core innovations of Accord
- 26:53Optimizing fast-path quorums
- 29:12Timestamp reordering
- 31:18Status and future of Accord
related talks
48mThe Accord Consensus Protocol, how to do ACID transactions at scale
46mData Replication with Apache Accord
34mCassandra on ACID: This Changes Everything
54mEvolving Transactions in Apache Cassandra
30mApache Cassandra: What’s New and What’s Next
13mThe ACID BASE Conundrum
47mBeing transactional in 2019 - what does this actually mean?
58mDon’t Drop ACID - Transactions in Distributed NoSQL