25 September 2003

Data warehouse fact tables contain one or more measures (such as 'volume sold', 'cost price' and 'retail price') and one or more dimensions. When the data model is first put together, the number of dimensions is usually quite small, and centre around common business areas like organization, time and geography. As the data model is further refined, and the requirements for reporting are incorporated into the fact table design, a number of additional items often get added to meet someone's specific requirements for a report. Examples I've come across include 'Invoice Status', 'Payment Type', 'Matched/Not Matched', and so on, and the question then becomes - do we treat these items as proper dimensions in the data model, or do we just leave them in as a special type of measure in the fact table?   9:40:12 PM

Lillian Hobbs, co-author of the definitive Oracle 9i Release 2 Data Warehousing book, has written a new article for OTN on Query Rewrite. The article explains the different types of query rewrite scenario, techniques such as aggregate computation and joinback, and how dimensions and hierarchies can be used to enable a single materialized view to support multiple levels of rollup. There's also mention of a useful feature coming in Oracle 10g where, if a materialized view only contains a subset of the columns required to satisfy the query rewrite rules, it can retrieve some of the rows from the materialized view and the rest from the detail table. Lillian also put together an interesting article for Oracleworld on the SQL Access Adviser, a new feature being delivered with Oracle 10g that builds on the existing Index Tuning Wizard and Summary Adviser Wizards provided with Oracle 9i. The SQL Access Adviser will be available either as a GUI, or through the command line, and recommends and implements the best combination of summaries and indexes for a particular schema or set of tables, based on current or previous access patterns. 1:45:30 PM