Efficient columnar storage in B-trees

Abstract

Column-oriented storage formats have been proposed for query processing in relational data warehouses, specifically for fast scans over non-indexed columns. This short note proposes a data compression method that reuses traditional on-disk B-tree structures with only minor changes yet achieves storage density and scan performance comparable to specialized columnar designs. The advantage of the proposed method over alternative storage structures is that traditional algorithms can be reused, e.g., for assembling rows with multiple columns, bulk insertion and deletion, logging and recovery, consistency checking, etc.