Scaling the CAP Theorem for Modern Distributed Systems
summary
Describes practical alternatives to the CAP theorem for building modern distributed systems, focusing on Dynamo-style architectures: consistent-hash rings with virtual nodes, gossip-based membership, and configurable quorum math (R+W>N) plus read-repair and hinted handoff. Covers conflict-resolution patterns from last-writer-wins and vector clocks to CRDTs and delta-CRDTs, and cites real-world incidents and deployments (MongoDB tail-latency and replica-lag issues, and production use of Aerospike/Cassandra and CRDT-style models at SoundCloud, Bet365, and Angry Birds).
AI-written summary · about
chapters (24)
- 0:00Introduction
- 1:25CAP theorem reality
- 3:35Consistency vs latency
- 6:50SQL myths
- 9:21ACID in distributed
- 12:56SQL for hot vs cold
- 14:41Dynamo alternatives
- 17:51Dynamo partitioning
- 20:40Cassandra ring
- 23:55Dynamo consistency
- 26:50Handling unavailable nodes
- 27:55Conflict resolution LWW
- 30:25CRDTs to avoid conflicts
- 32:31CRDT properties
- 34:18CRDT example Facebook likes
- 36:48CRDT in Twitter tweets
- 38:13Delta CRDT optimization
- 40:43CRDT in production
- 41:54CRDT vs Event Sourcing
- 45:46MongoDB scalability
- 47:36MongoDB load failure
- 49:42When SQL still wins
- 53:19SQL to streaming
- 54:43When not to use NoSQL
related talks
47mDatabase Resiliency
52mCAP Theorem - two decades and few clouds later
41mEverything you always wanted to know about highly available distributed databases
57mLiving in eventually consistent reality
01h 03mAdventures in Spacetime
20mCAP Theorem Reloaded
28mWhy Are Distributed Systems So Hard?
41mUnderstanding the CAP Theorem with Apache Kafka