Skinny Fact Tables are good

Many of my customers have billions of rows in their fact tables, which means there cube processing time is, at least partly, dependent on how long it takes the relational database manager to scan the data from disk. With most rdbms' (eg SQL Server, Oracle, DB2 etc.) this ends up as how fast the disk subsystem can transfer the data. So one way of improving performance without changing hardware is to trim up the fact table. Naturally, incremental processing is useful for these very large fact tables, but everything will go faster with skinier fact tables.

Here are a few ways that you can trim the fat out of your fact tables.

Read more...