What Goes Around Comes Around... and Around, Part II

In part two, Alexander Polomodov and Alexey Svetlichny move to system architecture. They start with row-oriented OLTP storage and explain why analytics favors columns. Vertica and ClickHouse read required fields and balance compression, scanning, and compute cost differently.

Cloud architecture separated compute from storage: processing can change while data remains durable. Aurora applies the idea to transactions, lakehouse systems to analytics, and Greenplum distributes queries across nodes. Commodity hardware and hot, warm, and cold tiers help control cost.

An OLTP schema change breaks downstream reports when the pipeline lacks a contract. ACID and BASE set different expectations, while distributed transactions need event order without shared clocks; Spanner pays for specialized infrastructure. A universal magic box requires expensive expertise and disciplined users.

The NoSQL cycle shows marketing and open source repackaging old ideas before restoring missing properties. ORMs simplify the start but hide machinery until failure. Autonomous databases and machine-learning optimization may be next, yet technology is useful only with an understood operational model.