Landmark Cloud Systems

Engineers meet DynamoDB, S3, Spanner, Lambda and Kubernetes every day as black boxes with documented limits: a per-partition throughput ceiling, a read-after-write guarantee, a cold start, a reconciliation loop that "eventually" converges. Those limits follow from a handful of design decisions, each taken against a named alternative for a stated reason. This course teaches the decisions, read from the systems' own papers and engineers, so that you can predict a system's behaviour beyond its documentation, model your data and workload to its grain, and reuse its ideas in your own designs. It builds on System Design and the courses beneath it: replication, consensus, clocks and the rest are recapped in a line and linked, and each lesson spends its length on what this system decided and why.

There is one module per system: DynamoDB; Aurora; S3; EBS and Physalia, the consensus store behind Amazon's block volumes; Firecracker and Lambda; Spanner; Aurora DSQL; Borg to Kubernetes; and Zanzibar, Google's authorization system. Each opens with the problem that forced the system into existence, then the contract it exposes, the decisions behind that contract and what it forces on its users. Modules are independent: take any system once you have its prerequisites. Reading the systems against each other closes the course by comparing them on the ideas they share — where consensus sits, how blast radius is contained, and how a contract shapes the data models built on it.

69 topics across 10 modules, in dependency order.