Everything you always wanted to know about highly available distributed databases
summary
Explains how large systems partition with a 32-part ring or similar hash ranges, keep data on three replicas for quorum durability, and use gossip membership to detect node liveness within seconds. Covers tradeoffs of master‑slave and master‑master replication, quorum read/write tuning, conflict handling with vector clocks and hinted handoff plus anti‑entropy, cites Netflix using Cassandra at ~1,000 nodes and Facebook using ~20,000 analytics servers, and points to Dynomite and Ringpop libraries.
AI-written summary · about
chapters (15)
- 0:00Start
- 4:03Complex systems can and will fail
- 9:51Multi-master is great, but: All the operations are replicated on all masters
- 13:32CAP Theorem
- 15:49Distributed DB design decisions
- 17:49Data distribution. The Ring
- 22:21Replication: durability
- 25:17Conflicts
- 26:24No-Conflict strategies
- 28:13Conflict resolution
- 28:55Vector clocks
- 31:08membership
- 32:40Gossip example
- 34:54Incremental scalability
- 36:11Operation under partition: Hinted Handoff
related talks
47mDatabase Resiliency
44mDistributed systems and scalability rules
43mDYNAMO: the paper that changed the database world
25mGrowing Distributed Systems
57mScaling the CAP Theorem for Modern Distributed Systems
57mLiving in eventually consistent reality
24mComparing the High Availability of MySQL with Modern Distributed SQL Databases
28mTobias Petry :: Scaling Databases