Performance issues of SQL access to AW

Question

Hi Experts:
I wonder whether there is performance issues when using SQL to access AW. When using SQL to access cubes in AW, the SQL queries the relational views for AW objects. And the views are based on OLAP_TABLE function. We know that, views based on any table function are not able to make use of index. That is to query a subset of the data of a view, we would have to full scan the view and then apply the filter. Such query plan always lead to bad performance.
I want to know, when I use SQL to retrieve a small part of data in an AW-cube, will Oracle OLAP engine retrieve all data in the cube and then apply the filter? If the Oracle OLAP engine only retrieves data needed from AW, how can she did it?
Thanks.

Answer 1

For most requests the OLAP_TABLE function can reduce the amount of data it produces by examining the rowsource tree , or WHERE clause. The data in Oracle OLAP is highly indexed. There are steps a user can take to optimize the index use. Specifically, pin down the dimension(s) defined in the OLAP_TABLE function LIMITMAP via (NOT)IN lists on the dimension, parent, level or GID columns. Use of valuesets for the INHIER object, instead of a boolean object.
In 10g, WHERE clauses like SALES > 50 are also processed prior to sending data out.
For large requests (thousands of rows) performance can be a problem because the data is being sent through the object layer. In 10 this can be ameliorated by wrapping the OLAP_TABLE function call with a SQL MODEL clause. The SQL MODEL knows a bit more about the Olap options and does not require use to pipe the data through the object layer.
SQL MODEL example (note no ADT defintion, using of auto ADT) This can be wrapped in a CREATE VIEW statement :
select * from olap_table('myaw duration session', null, null, 'measure sales as number from aw_sales_obj dimension d1 as varchar2(10) from geog ...rest of dimensions')
sql model dimension by (d1, d2, d3, d4) measures (sales, any attributes, parent columns etc...) unique single reference rules update sequential order ()
Example of WHERE clause with above select.
SELECT *
FROM (select * from olap_table('myaw duration session', null, null, 'measure sales as number from aw_sales_obj dimension d1 as varchar2(10) from geog ...rest of dimensions')
sql model dimension by (d1, d2, d3, d4) measures (sales, any attributes, parent columns etc...) unique single reference rules update sequential order ()))
WHERE GEOG NOT IN ('USA', 'CANADA')
and GEOG_GID = 1
and TIME_PARENT IN ('2004')
and CHANNEL = 'CATALOG'
and SALES > 50000;