OB-tree: A new write optimisation index on out-of-core column-store databases

Abstract

The column-store database is a representative of next generation databases featuring a high reading speed. Write optimisation in the out-of-core column-store database remains a well-known challenge. Timestamped binary association table (or TBAT) and asynchronous out-of-core update (or AOC update) have shown improvements in write performance. However, a common restriction shared by the timestamp-based approaches is that, after a time period of updates, the searching performance will gradually decrease. In this work, we introduce a new index, called Offset B+-tree (or OB-tree), to further improve the data retrieval speed after many updates have taken place. OB-tree is a flexible and robust index that employs a special pointer elimination strategy to reduce the storage costs. Succinctly designed, OB-tree can be easily integrated into existing timestamp-based column-store databases. Extensive experiments show that OB-tree can be efficiently constructed and significantly improves the data retrieval speed on the TBAT even when a large number of updates occurred.