Different aggregation operator for two measures in single cube

Question

Hi,
I have three measures age_month and age_year,pages for which aggregation operator should be like this
for age_month and age_year
time - max
account - sum
sales org - sum
for pages
time - sum
account - sum
sales org - sum
I am creating MOLAP cube in OWB and deploying to AWM. I can create above structure in OWB but when I deploy it to AWM I see that for all the dimensions and all the measures aggregation operator as sum. Off course I can change the aggregation operator at cube level but it make changes to all underlying measures too.
Also I can see in XML of the cube and see the operator for two measures is as max but UI shows as sum. After load also I am seeing in report it is calculating as sum instead of max along time dimension.
Any help would be highly appreciated.
Thanks
Brijesh

Answer 1

If you have an existing cube (already defined and aggregation setup) then modifying the aggregation behavior like changing order of dimensions/aggregation operators is not very robust (likely to fail due to inability of internal objects to sync up to modified defintion). Its always better to take a backup of cube definition and drop/rebuild a fresh cube with new dim order/agg operators.
How can you have a compressed cube and also set the aggregation property per dimension?
I was under the impression that if it is a compressed cube, then not only is there a restriction that all measures have same aggregation properties (defined at cube level) but also the aggregation behavior defined at cube level should use the same operator (note: any one operator out of sum or max or min) for all dimensions.
Edited by: Shankar S. on Dec 19, 2008 11:54 AM
Perhaps this additional restriction is only required if we intend to use the Cube as a source for relational queries using Cube organized MVs.
Another way to do this is given below:
I'm assuming that when you say that you require Max along Time and Sum along other dimensions, you mean that you want LAST (chronological) along time at leaf level and SUM along other dimension. Also assuming that Time hierarchy has the following levels: DAY, MONTH, QUARTER, YEAR. Big assumption :) as finding out Max along time requires the help of the fact/measure whereas finding out last along time requires us to use the dimension (along with dimension metadata) alone.
Define 1 Cube: Cube1 ... Structure: Compressed Composite, datatype: DECIMAL
Set Aggregation: SUM along all dimensions
Create 3 stored measures ( age_m_sto, age_y_sto and pages)
You may want to set description for age_m_sto/age_y_sto as "******** INTERNAL: DO NOT USE ******** Stored Measure used for age_month/age_year"
Create 2 calculated measures - age_month and age_year defined as follows ...
olap expresssion to be given in AWM:
<age_month>: OLAP_DML_EXPRESSION('cube1_age_m_sto(time if time_levelrel eq ''DAY'' then time else statlast(limit(time to bottomdescendants using time_parentrel time(time time))))', DECIMAL)
<age_year>: OLAP_DML_EXPRESSION('cube1_age_y_sto(time if time_levelrel eq ''DAY'' then time else statlast(limit(time to bottomdescendants using time_parentrel time(time time))))', DECIMAL)
NOTE: The caculated measure performs the LAST along time action using the stored measure. For every higher level time dimension member, it reaches along the dimension to the last leaf level member and reads the stored measure for that lowest level time member.
Map and Maintain Cube.
From the SQL Cube View: Use only the columns corresponding to stored measure: pages and calculated measures: age_month, age_pages (ignore columns corresponding to age_m_sto, age_y_sto)
HTH
Shankar