Very slow MV fast refresh after "Merge" on dimension

Question

Hi,
We have a sales cube with 300Mio records and product dimension with 20.000 records.
We have a materialized view something like:
SELECT      COUNT(*) AS cnt,
     COUNT(cube.amount) AS cnt_amt,
     sum(cube.AMOUNT) as amount,
p.product_name
FROM sales_cube cube
JOIN product_dimension p on cube.product_dimension = p.product_dimension
group by day, product_dimension
There is tho mv_log on cube and on product_dimension defined.
After import process for sales_cube
In the night we run an product dimension update script implemented wit MERGE.
If something changes in product dimension then next fast refresh from materialized view takes 4 hours. If we have no changes in product_dimension then fast refresh takes 8 seconds.
My question is, why takes it sooo long and who can we prevent it.

Answer 1

I'm not authorized to publish EXECUTION PLAN or schema details.
But I found the problem.
ETL Prozess is generated by Warehouse Builder.
ETL Prozess is modeled as Mapping from product table to product dimension in Warehouse as "UPDATE/INSERT" by PK (product number).
Warehouse Builder generates an MERGE statement for such "UPDATE/INSERT" mappings.
But this mapping updates every record in Dimension Table, as result Materialized View Log for this dimension is confused. Fast refresh (join between fact table and dimension) is very slow (slower than compleate refresh).
Now, i simple must find out how i can change this OWB mapping not to update records in dimension if corresponding records are not modified in source table.
There is an "match by constraint" option in mapping, looks useful, i'll try it tomorow.