ETL patterns/strategies/templates/"best practices"

Question

The OWB Users Guide is good for defining the OWB operators, but I'm having some difficulty deciding how to best combine those operators to formulate a mapping. I'm hoping someone will be able to point me to documentation that will fill the gap. It wouldn't necessarily need to be OWB specific.
Some example questions that I'm hoping an "ETL patterns" book or document would address:
1. High level implementation decisions (Is the source data staged locally? How/when are the changed records identified?)
2. Options for locally staging data.
3. Options for implementing a Type I/II/III SCD.
4. Options for implementing a cube.
I was able to find "Slowly Changing Dimensions with Oracle9i Warehouse Builder" (http://otn.oracle.com/sample_code/products/warehouse/files/SCDWhitePaper.zip) that gave a lot of guidance on the SCD issues. The Type II SCD mapping presented in that document uses 10 operators. I never would have been able to formulate that on my own based on just reading the OWB documentation.
Thanks,
Brian

Answer 1

Hi Torsten
I assume you want to follow the framework described in the SCD Whitepaper.
Where do I have to add the unbound table?You will need to insert data into a table, and not a dimension. More than this, since you have a need for using a sequence to feed the items with a surrogate key, there comes a matching problem (this is described in the whitepaper).
The table must be added as the final destination of the data, after the last union (as shown in the Whitepaper).
Is it an OWB table in the target module where the target dimension is located? Or do I have to create it in the database using sql*plus?It is neither, read the next question
Why do you call it "unbound"? Because it is not a dimension or fact-table?The reason I call it unbound is that OWB does not have a "true" match between the object in the mapping and the object in the database. There is probable more to it than this, but let's just say, that it could be compared to some temporary table or something - usually it is not good to use unbound tables, but since we edit the "bound name" to an existing repository table, it's no problem - but be aware, that you will get a warning ever after...
But is it true that I don't need the target dimension in this mapping now?No, you should use it as in the whitepaper, since you need to check if there is a trigger column that should instantiate a new record, or, if an insert or an update should be required.
Or how do I connect the unbound table to the target mapping?You change the "bound name" (right-click the table, and choose "properties" and change the bound name to the name of the dimension).
Torsten, generally you should follow the whitepaper, except for the last table/dimension :)
Don't sweat if your stuck - I've clicked a gazillion times in OWB, in order to make this work - ask again, if I don't make sence :)
Yours
Kim