World's Fastest Scalable Join...?

Question

Hi everyone,
Our company has been having some performance issues with some of our data warehouse queries. The main issues seem to arise when we join or orders table to our customer location table (twice, once for bill to location and once for ship to location) and our organization table. We have about 10 million records in orders and about 7 million records in customer locations and about 7 million in organization. I decided to look around around and see what is good out there in terms of modeling the tables better to get faster queries.
I found the following article: http://www.orafaq.com/node/1446. It claims that a partitioned hash joins are the best thing to do.
In his example, he had an order table and customer table as well. From what I gleaned from the article, the recommendation is to partition the fact table (orders) on date and subpartition it based on a hash of the foreign key join to the dimension (customer). Then you also partition the customer table based on the same hash subpartition the fact has. This will allow oracle to do hash joins "combined with the scalability of partitions".
It sounded cool, so I wanted to give it a shot. But in my situation we have two dimensions, not just one. And some day our product dimension might also start getting in the millions, so we'd also want to be able to do this trick with that.
Is there a way to make this work for more than one dimension? I apologize but I'm not super familiar with hash sub-partitions, so I am not sure if we can define multiple versions.
Thanks for the help!
-=Joe

Answer 1

No, you have to pick one. The data is physically stored in the subpartitions - you can think of them as mini-tables made to appear as one table by the data dictionary.