Distributed Systems — Episode 7

The seventh Distributed Systems episode focuses on replication and consistency. Replicas move data closer to consumers and scale reads, but every copy delays change propagation. The guarantee must therefore follow product semantics rather than a desire for the abstractly strongest model.

The speakers read operation timelines, compare sequential and causal consistency, and examine causal dependencies. Data grouping leads to locks, while transactions introduce serializability: a database must present concurrent changes as a valid sequence.

Eventual consistency is explained through operation properties: commutative changes converge under different orders. Continuous consistency measures divergence by unapplied updates, staleness, and ordering. Client-centric guarantees such as read-your-writes protect one user journey.

The finale turns to replica management and protocols: placement, operation propagation, Calvin-style ordering, primary-based replication, and quorums. Web caches and CDNs show the practical result: lower latency and origin load, paired with explicit update, validation, and placement rules.