Attribute relationships inside of Analysis Services are a critical piece to the design of your dimensions. The benefits of creating attribute relationships are multifold. It can reduce storage space, speed up processing time, improve query performance, and allow usage of member properties.
So how do attribute relationships work? The simplest answer is that it tells the formula engine how to roll up measure values. It defines a one-to-many relationship between attributes in a dimension.
For example, let’s say you’re working on a date dimension and you need to define a user defined hierarchy for your users because when they look at date they always look at it by Year – Quarter – Month –Date. Creating this user defined hierarchy saves the user time in dragging each field over individually and helps performance. If you want to follow along this uses the Adventure Works sample.
To create this hierarchy you must first drag fields over from the Data Source View pane to the Attributes pane to identify them as attributes. Next, you’ll drag them in the appropriate order (meaning highest level of the hierarchy goes at the top and lowest at the bottom) to the Hierarchies pane to define the user defined hierarchy. When creating this hierarchy you will likely receive a warning telling you that attribute relationships need to be defined to increase performance. Here is your big clue into how important attribute relationships are to Analysis Services.
The design of these relationships can be very different depending on the environment you’re working in. This is probably the biggest change in SSAS development between SQL Server 2005 and 2008.
If you’re working in 2005 your screen will look like the below screenshot. This is the completed design so what you must do to replicate this is expand each attribute in the Attributes pane. Once this is done you will see ‘L’ shaped attributes under the dimension key. This is what is used for defining attribute relationships in 2005. You must drag each ‘L’ shaped attribute from under the dimension’s key to each attribute that it relates to at a lower level. So, the ‘L’ shaped Year attribute goes under Quarter, Quarter goes under Month, and Month doesn’t go anywhere.
If you’re working in 2008 your screen will look like the below screenshot. This design is very different than 2005 because there’s another tab that is now used specifically for attribute relationships. Personally I think this newer method is a little easier to understand. This time after creating the hierarchy you just drag the lowest level on top of the next highest level. So drag Month on top of Quarter and then Quarter on top of Year. The top level of the hierarchy should always be on the far right. The relationship between Date and Month is already defined because every attribute has an implicit relationship with the key of the dimension.
Hope this gets you a good start on your dimension design. If you followed along using Adventure Works then the dimension will actually fail when processing. The next step is to define KeyColumns when necessary to ensure each attribute is uniquely understood within the user defined hierarchy.