Dear community,
What is the performance impact of access to attribute cubes in rules
vs. direct storage of data in cube cells?
Simplified example:
I have a product dimension with 20.000 products.
Each product has a price which is slowly changing.
Of course there is a time dimension on level year+month.
There will be rules on the cube:
Turnover = Price * Units
+ additionally there are several margin calculations on top
I would see 2 possibilities to cover this in my cube:
1.
Write the prices per month directly into the cube
2.
Transform the product dimension into a Type 2 SCD
via an additional hierarchy level.
Then store the price in an attribute and refer to it in a cube rule:
['Price'] = B: PALO.DATA("db","#_products","price")
Product A
Product A: 01.2013 - 06.2014; Price = 400
Product A: 07.2014 - 12.2015; Price = 420
Product A: 01.2016 - ; Price = 450
Annotation:
There is another dimension with ~20.000 elements as well in the cube, which is probably important for performance considerations.
I would tend to the SCD approach also considering, that future products will be created manually for planning purposes
without running through the source system ETL process.
But are there major drawbacks in terms of performance and scaling?
Thanks
JCO
What is the performance impact of access to attribute cubes in rules
vs. direct storage of data in cube cells?
Simplified example:
I have a product dimension with 20.000 products.
Each product has a price which is slowly changing.
Of course there is a time dimension on level year+month.
There will be rules on the cube:
Turnover = Price * Units
+ additionally there are several margin calculations on top
I would see 2 possibilities to cover this in my cube:
1.
Write the prices per month directly into the cube
2.
Transform the product dimension into a Type 2 SCD
via an additional hierarchy level.
Then store the price in an attribute and refer to it in a cube rule:
['Price'] = B: PALO.DATA("db","#_products","price")
Product A
Product A: 01.2013 - 06.2014; Price = 400
Product A: 07.2014 - 12.2015; Price = 420
Product A: 01.2016 - ; Price = 450
Annotation:
There is another dimension with ~20.000 elements as well in the cube, which is probably important for performance considerations.
I would tend to the SCD approach also considering, that future products will be created manually for planning purposes
without running through the source system ETL process.
But are there major drawbacks in terms of performance and scaling?
Thanks
JCO
