Daniel Suo

Scientific progress goes 'boink'

Ph.D. Candidate
Princeton University
Department of Computer Science

Column-Stores vs. Row-Stores: How Different Are They Really?

link

Elevator pitch:

Column-stores are more I/O efficient for read-only queries since they only have to read from disk (or from memory) those attributes accessed by a query

If this is the case, why can’t we achieve column-store performance with row-stores by vertically partitioning the schema or by indexing every column?

Answer: must make fundamental changes to both storage layer and query executor.