Al Brown

Vectorized query execution processes batches of values per operator call instead of one row at a time. It is the dominant execution model for analytical databases.

Row stores win point lookups and per-row updates; column stores win wide aggregations and compress 5–10× tighter. A measured comparison with a decision rule.

Columnar databases are 10-1000× faster than row stores on analytical queries because two principles compound: efficient execution (storage layout, vectorisation) and smart pruning (data skipping, late materialisation).

Columnar storage is an on-disk data layout where each column is stored contiguously. It cuts I/O for analytical queries and enables 5–10× compression.

How Apache Parquet, ORC, and Arrow lay out columnar data, and why they coexist instead of converging on one format.

Nine columnar databases in 2026, ranked using measured benchmark latency and compression rather than marketing copy.