Methods available for storing Custom Measures

Question

Just a general question.
Using 11g. If I have a complex DML program, which calculates values using existing stored measures within the cube and would like to store the end results of the values calculated during the run of my program. What methods are available?
Should my program be run as part of DBMS_CUBE package, should it be part of a maintenance script? I appreciate any info on this.

Answer 1

It would probably be useful to understand what the calculation needs to do, but in general you can embed this as David describes or you can do it directly using OLAP DML. Working off David's example:
exec dbms_cube.build('UNITS_CUBE USING (SET UNITS_CUBE.STORED_MEASURE = UNITS_CUBE.CALC_MEASURE'));
The direct OLAP DML method would look something like this:
SET units_cube_store_measure = units_cube_calc_measure
Or, let's say I want to assign data to sales_cube.sales with the expression sales_cubes.sales = sales_cube.quantity_sold * price_cube.unit_price. This might like like:
SET sales_cube_sales = sales_cube_quantity_sold * price_cube_unit_price
Note that these expressions refer to the OLAP DML objects (that is, the physical objects) rather than the API level objects.
If you want to constrain the values being set, use the LIMIT command. E.g.,
LIMIT time TO time_levelrel 'MONTH'
LIMIT product TO product_levelrel 'ITEM'
LIMIT geography TO geography_levelrel 'CUSTOMER'
SET sales_cube_sales = sales_cube_quantity_sold * price_cube_unit_price
David also notes that you need to pay attention to looping. The above example will "loop dense", that is loop over the base dimensions. Since data is usually sparse, that's not going to be as efficient as it can be. So, introduce the ACROSS command and loop over the composite dimension of the sales_cube. Assuming that the cube is partitioned (almost all cubes should be partitioned), the program now looks like this:
LIMIT time TO time_levelrel 'MONTH'
LIMIT product TO product_levelrel 'ITEM'
LIMIT geography TO geography_levelrel 'CUSTOMER'
SET sales_cube_sales = sales_cube_quantity_sold * price_cube_unit_price ACROSS sales_cube_prt_template
The ACROSS command will cause the SET command to loop only values in the sales_cube where data exists (for any store measure).
Finally, don't forget to save your work with UPDATE and COMMIT commands. E.g.,
LIMIT time TO time_levelrel 'MONTH'
LIMIT product TO product_levelrel 'ITEM'
LIMIT geography TO geography_levelrel 'CUSTOMER'
SET sales_cube_sales = sales_cube_quantity_sold * price_cube_unit_price ACROSS sales_cube_prt_template
UPDATE
COMMIT
David also mentions that dbms_cube.build will run jobs in parallel (assuming the cube is partitioned). That's really nice. You can parallelize this in OLAP DML. See http://oracleolap.blogspot.com/2010/03/parallel-execution-of-olap-dml.html for an example.
So, which method should you use? It depends on what you are trying to do. If it's easy to do with dbms_cube.build, I would probably do that. But, if you need more power and/or control, the OLAP DML method might be best.