Part 1 of 7 · 1 min read
Why this page comes after faults and elections
In this module you will
- Say what linearizability promises, using the recency rule, and tell it apart from serializability.
- Name the jobs that truly need linearizability: locks, uniqueness and cross-channel timing.
- Say which replication styles can be linearizable and which cannot, including the quorum trap.
- Place sequential, causal and eventual consistency on a ladder from strong to weak.
- Choose a model for a feature and name the price.
Module 4 showed that machines cannot trust clocks or even their view of each other. Replication (Level 4) keeps several copies of the data. The question now is what a reader may see when the copies differ for a moment. A consistency model is the written promise. The books used here are "Designing Data-Intensive Applications" and "Understanding Distributed Systems". The numbers are made up to show the idea, unless the text names a book.