Hi eb
I Have a model where all aggregations of all measures are calculated by ETL process and mapped in facts to members of all levels in (parent-child) hierarchies.
Since the cube adds no aggregations of its own i see no advantage of MOLAP storage design over ROLAP.
Further more - I dont see a reason for creating and maintaining multiple partitions on such a ROLAP cube .
That's because I understand all queries end up executed against the original relational fact table any way right?
So the important thing to do performance wise speaking is to partition and index the underlying facts table to match queries that will be sent from clients browsing that cube.
But I have come across this puzzling advice by microsoft to set the slice property of ROLAP partitions
(see here: http://msdn.microsoft.com/en-us/library/bb630308%28v=sql.105%29.aspx)
Why partitions? Why slice property?? What am i missing???
Another question regarding queries performance: cube's star schema is in Oracle.
Would it pay off by queries improvement to import it all to sql server ?
Thanks a lot!
Rea