We have gone down the multiple cube design path. We a cube for each logical area in each database. These cubes can be of many different measure groups, but will all be of similar dimensionality. All of the measure have got a set of common
dimensions, so we have created a "Analysis" cube which only has the common dimensions in it with linked measure groups to all of the base cubes.
While we were working on some new additions, one of the developers tried to create a cube using loopupcube. Basically he added the required dimensions with a small measure group (since one is required to build the cube). All the measures were
created using the lookupcube function with the required dimensionality.
We have run some tests on this cube and an identical cube that is using Linked measure groups.
In all cases, the cube that uses the LookupCube function is a lot faster on the cold cache, initial query (more than 50%)
On each subsequent execution, the Linked cube is much faster (31ms for Linked, 750ms for LookupCube)
I know everyone recommends that you do not use Lookupcube, but the much faster initial query time is attractive.
Anyone else seen this or tried this?
View Complete Post