1. 前面提到了 Single-copy consistency 的例子。本节说的是 Eventual consistency with probabilistic guarantees 和 Eventual consistency with strong guarantees.<br />    Amazon's Dynamo is an eventually, weak consistency guarantees but high availability key-value store. 它使用了 Consistent hashing 来把 keys 映射到不同的节点，使用 partial quorums 来进行复制。<br />    W, R指定需要参与写入或读取的节点数。N是代表数据副本数。 一般来说会推荐 R + W > N.<br />    在冲突检测和读取修复，我们分别可以从 元数据(metadata), 时间戳(timestamp), 版本号(Version numbers), 矢量时钟(Vector clocks) 来代表数据的历史。<br />    后面的 PBS / CRDT / the CALM theorem 不太能理解。

目录

  1. ## Reconciling different operation orders
  3. ## Amazon's Dynamo
  4. ### Consistent hashing
  5. ### Partial quorums
  6. ### Is R + W > N the same as "strong consistency"?
  7. ### Conflict detection and read repair
  8. ### Replica synchronization: gossip and Merkle trees
  9. ### Dynamo in practice: probabilistically bounded staleness (PBS)
  11. ## Disorderly programming
  12. ## CRDTs: Convergent replicated data types
  13. ## The CALM theorem
  14. ## What is non-mononicity good for?
  15. ## The Bloom language
  16. ## Further reading