SSAS: dimension tables: 50 million rows x 50 columns

SQL SERVER 2005 / 2008 ANALYSIS SERVICES

I have 3 tables (more I have to consider later) and all 3 contain approx. 50 million rows. The first table contains details of tranactions (and will be a fact table) and the other two tables are dimension tables called 'Client' and 'Policy'. 'Client' and 'Policy' both contain around 50 columns. Many of the columns contain codes of a sort which suit the use of look-up tables, for example Client contains codes for 'occupation category' and 'employer occupation category', etc.

Would it be better to:
a. simply create dimensions based upon the Client and Policy tables as they are, or
b. create separate dimension tables attached to the Client and Policy tables for all the different codes (i.e. look-ups for their descriptions) and therefore create dimensions for these tables as well ?
(This would equate to a snow-flake design in which the Client and Policy tables have lots of lookups hanging off them).

Also, there is a direct relation (one to many) between the Client and Policy tables. Would I also set this relation to each other (i.e between the Client and Policy tables) as well as the relation of Client and Policy to the fact table ?


Des
DerekRobertsAsked:
Who is Participating?
 
agandauCommented:
Based on your description, I would create two dimensions.  One Policy dimension and one Client dimension.  In each of these two dimensions, if the performance isn't miserable, I would snowflake in the lookups and use the codes as member keys and the descriptions as member names creating an attribute in the dimension of any of those that are necessary for reporting.

Since it sounds like the relationship exists between Client and Transaction I would avoid using the referenced dimension relationship and have the fact table be regularly related to both the Client and Policy dimensions separately for performance reasons, especially considering the number of members you'll likely have in both of these dimensions.  In the DSV then I wouldn't bother with having that relationship between Client and Policy defined.

My main concern with what you're describing is what kinds of changes happen in the dimensions.  I'm guessing a policy doesn't change.  If a client changes a policy perhaps the existing is expired and a new one created?

A Client on the other hand, may change demographics (get married or move to another state), or their employer might change from one occupation category to another, etc.  What happens to the records in the Client when that happens.  Is a new record cut with the changes, and then do subsequent transactions point to that new record, or is it updated in place?  It's the whole "slowly changing dimension" debacle, which has to be taken care of via ETL in the database.
0
 
agandauCommented:
Is the Policy table something more like a "Client Has Policy" table, where a record will regardless even if there aren't any records in the transaction table?
0
 
DerekRobertsAuthor Commented:
there will be always be records in the transactions table for a policy table entry.  The transactions table is an amalgamation of a stats (events) and ledger table which records something for everything.
Clients have policies and policies always have transactions associated with them.

0
 
DerekRobertsAuthor Commented:
The above is pretty much what I've been thinking (dreading).  Any innovative ideas are always welcome but I suspect in this case the real requirement is for more powerful hardware to accommodate the above scenario with regard to snow-flaking.

Fortunately we don't have to deal with the issues surrounding slowly changing dimensions.
0
Question has a verified solution.

Are you are experiencing a similar issue? Get a personalized answer when you ask a related question.

Have a better answer? Share it in a comment.

All Courses

From novice to tech pro — start learning today.