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

Alexander Polomodov and Alexey Svetlichny review Stonebraker and Pavlo’s What Goes Around Comes Around... and Around. The paper revisits a twenty-year-old predecessor to ask which database ideas disappeared and which returned under new names. Alexey adds experience from DBaaS and database operations at T-Bank.

Part one covers data models and queries. The relational model remains the baseline to which systems return. Key-value stores and Hadoop simplified particular jobs, yet SQL reappeared above MapReduce. Document stores reduced object-relational mapping, while MongoDB later added transactions required in production.

Cassandra shows the cost of leaving relational thinking: schemas follow queries, while an LSM tree favors writes over reads and compaction. Search uses inverted indexes. Vector databases store embeddings for approximate-neighbor search; Qdrant and Milvus are compared with PostgreSQL extensions.

Graph databases help when relationships and traversals are central, including social and corporate networks. Cypher adds pattern matching, but a large graph is hard to partition. The lesson is to choose by model value, read/write profile, and operational constraints rather than by a category’s novelty.