Do we need Dimension delta loads?

[mamba]

Hi,
this is something like a best practice question.

Is it necessary to implement delta processes for Dimensions or
can i always unconsidered load the full dimensions again and again?

My Question is about what happens within the cube cells,
if i do a full dimension load with the same Elements, does it has impact on existinct Cube cells,feeders, rules?

I would like to go for full dimension load and implements delta processes only for fact, is this a common approach?

[Steve Rowe]

I'd consider the following the best practice for dimension builds, assumes having a good and well controlled external source for the metadata.

1. Never do a dimension delete all elements.
2. Flatten your dimension, i.e. remove all parent child relationships.
3. Rebuild the dimension from your source, testing and blocking the conversion of N type to C type
4. Add all elements with no parents or children to a control consolidation "Orphans".
5. Children of Orphans are (probably) elements that are no longer referenced in the data source, perform analysis on these to understand why this would be and delete / correct as required.

Golden rules.
Never do a dimension delete all elements or automate the deletion of elements in any way.
Prevent the conversion of N level elements to C level elements to minimise the risk of data loss.

The above gets significantly more complex when dimension structure has multiple sources so try not to get into this situation.

HTH