Chapter 11: Replication & Consistency Models

The chapter establishes a conceptual foundation through shared memory abstractions and register classifications—safe, regular, and atomic—to illustrate how concurrent operations interact within distributed environments. A comprehensive analysis of consistency models follows, starting with strong guarantees like linearizability, which ensures operations appear instantaneous and maintain real-time ordering through carefully defined linearization points. The exploration continues with progressively weaker models including sequential consistency, which maintains global operation ordering consistent with individual process histories without real-time constraints, and causal consistency, which preserves causally related event ordering using vector clock mechanisms while permitting unordered concurrent operations. The chapter distinguishes between system-wide consistency guarantees and client-centric session models, examining read-your-writes, monotonic reads, and monotonic writes patterns that ensure coherent individual user experiences. Practical implementation approaches are detailed through tunable consistency mechanisms and quorum-based systems, demonstrating how configuring read and write replica requirements enables adjustable consistency guarantees while witness replicas provide storage optimization opportunities. The discussion concludes with advanced concepts including strong eventual consistency and conflict-free replicated data types, such as counters and sets, which enable independent update acceptance across distributed nodes while guaranteeing mathematical convergence to correct states without explicit conflict resolution mechanisms.