Typical Dimension

Question

Hi
The tools we are using are
OWB 9.0.2.56 for ETL and
Oracle Express 6.3.4 server, RAA/RAM and Oracle Express Web Agent.
The database is Oracle 9i
I have to design a very typical dimension for the following requirements which I guess should be common for most of the Data Warehouing projects.
We need to have a hierarchy for a department dimension in payroll module where we have 4 levels as follows
General Department --> Department --> Sub Department --> Section. Section being at the leaf level.
The doubts I have is
1. I was under an impression all these days that data should be available at leaf level but here I see some cases where employees does not have a section as they are directly into a sub department and similarly some employees does not have section as well as sub department as they directly fall in Department and Vise Versa. So the data is not always at leaf level. So how is the aggregation going to take place.
2. Say we have a measures like number of employees terminated, recruited, promoted etc As discussed in point1. if Section1 and and section2 are falling under Sub-Department1 so here say 200 employees are working in section one and 250 Employees are working in Section2 and 150 employees working in Sub-Department1. Here on aggregation at sub-department1 it sould show the sum of all the three figures(200+250+150) but not just the aggregation of section1 and section2(which is 250 + 150)
3. The another issue is that for some cases it is skip level like section1 directly falls under Department and so the level is being skipped. How do I address this in my dimension design.
Can any one through some light on this.
How can we design and implement the hierarchy in OWB
Thanks and Regards
Nanda Kishore

Answer 1

There is a way to design a dimension so that all levels of the data can be loaded to it and the aggregation hierarchy is preserved.
Firstly, the dimension implementation looks like the following:
ID Top Level Middle Level Detail Level
1 Total null null
2 Total East null
3 Total West null
4 Total East Part1
5 Total East Part2
6 Total West State1
7 Total West State2
8 Total West State3
The null's could be replaced with "Not Available" if from an EUL that is a preferable presentation.
In the ETL for loading the fact table I use Key Lookup Table (Index Organized) with the natural key for the dimension as the primary key, and the surrogate key of the ID as the output value. Giving a table which looks like:
Natural Key Dimension ID          
Total 1
East 2
West 3
Part1 4
Part2 5
State1 6
State2 7
State3 8
With this structure data in the fact table can be loaded and rolled up independent of the level of the dimension which the data comes in on. The roll up in the EUL also works happily (without any double counting).