CRDTs in Practice
summary
CRDTs give always-accept updates with deterministic merge rules and actual libraries (React implements an observe-remove set used in production at a major betting site with bursts to ~2.5M concurrent users and ~1 TB working set). Prototypes include an RGA sequence, a bounded counter that distributes “rights” and fails locally when rights are unavailable, and SwiftCloud (3 AWS datacenters) showing much lower latency and higher throughput than a single strong-consistent replica while preventing numeric-invariant violations. Limits appear when invariants require unstable preconditions.
AI-written summary · about
chapters (24)
- 0:00Intro
- 0:28Cloud Computing
- 2:50Conflictfree replicated datatypes
- 4:09Why use these datatypes
- 5:31Design Concepts
- 9:43Bet365
- 13:30New Data Type
- 15:35Linked Lists
- 21:32Availability
- 22:43Invariants
- 25:35Replicas
- 26:44Bounded Counter
- 29:17Transferring Rights
- 30:08Merging
- 33:00Summary
- 33:47MicroBenchmark
- 36:35Swift
- 37:28Results
- 38:19Important Message
- 39:33Conclusion
- 41:02Questions
- 41:37Operational Transformation
- 44:38Weak vs Strong
- 47:07Graph Data Types
related talks
35mData Types for Eventually Consistent Databases
22mHappily Ever After: A CRDT Fairy Tale
29mHow to Read the Exact Time When You Have a Billion Clocks
49mDistributed Data: Conflict-free Data Structures in JS
46mPractical data synchronization with CRDTs
39mUsing Erlang, Riak and the ORSWOT CRDT at bet365 (...)
46mCoordination-Free Computations
21mCRDTs for Mortals