DataWarehousing

Data Warehousing Features

• On-disk Bitmap Index (anyone game to finish GP patch?)
• Error Handling in COPY
• Parallel Query
• Windowing Functions
• MERGE
• Parallel Index Build

Note that there's much overlap between here and the Simon Riggs' Development Projects planning.

Error Handling in COPY

Handled OK by pg_loader, so lower priority

Parallel Query

2 main kinds are single-node and multi-node parallelism Fairly easy to get something working to improve SeqScan performance, but will be more difficult to get something working that applies further up into the executor. Challenges are

• Planner changes
• How to manage the pool of query slaves
• Deciding which parts of the data are accessed by which slave

Notably the last two aren't an issue at all in most multi-node parallel architectures, since there is one query slave per node, plus each node operates only on the data that has been statically partitioned to it, often using a hash partitioning scheme.

Windowing Functions

SQL:2003 feature

MERGE

SQL:2003 feature

Parallel Index Build

Josh Berkus: not sure how this works exactly, but it speeds Oracle up considerably

Slides

Overview of "Data warehousing with PostgreSQL" presented by Gabriele Bartolini at PGDay.EU 2009 (Data warehousing with PostgreSQL)