OLTP systems (Postgres, MySQL, most application databases) optimize for many concurrent small transactions: insert a row, fetch a row, update a row, commit. OLAP systems optimize for the opposite pattern: scan millions of rows, group, aggregate, return a summary.

The data layouts diverge because the workloads do. OLAP engines store data by column so aggregation queries only read the columns they need, compress aggressively because adjacent values are similar, and parallelize scans across CPU cores. The trade-off is that single-row reads and writes are expensive.

Real-time analytics, user-facing analytics, and observability all sit firmly in OLAP territory. Tinybird is built on ClickHouse®, a column-oriented OLAP engine.