CAP Theorem — two decades and few clouds later
summary
Explains that CAP remains a hard constraint but is practical and nuanced: S3 claims strong read‑after‑write while buckets stay eventually consistent and S3 SLA is three nines with 99.9% replication under 15 minutes, Spanner is effectively CP by using TrueTime (bounded error ~7 ms) to get linearizability and trades availability for predictability, and Cassandra and Kafka expose tunable consistency (quorum, acks, unclean leader election) so behavior depends on configuration and SLAs.
AI-written summary · about
chapters (11)
- 0:00Start
- 7:14Cap Theorem
- 8:45Product Announcements
- 10:48Network Partition
- 13:13Node Failures
- 14:54Strict Consistency
- 16:27Hierarchy of the Consistency Models
- 34:39Tunable Consistency
- 38:09Kafka
- 39:56What Is the Consistency Recipe for Apache Kafka
- 1:14:34Q and a Session
related talks
52mCAP Theorem - two decades and few clouds later
20mCAP Theorem Reloaded
54mDay 1 Session 1
13mThe ACID BASE Conundrum
47mBeing transactional in 2019 - what does this actually mean?
24mStronger Semantics for Low-Latency Geo-Replicated Storage
25mThe shell game called eventual consistency
22mOrder And Consistency At Scale