Fitting number of dimensions

In cubes how many dimensions are one to many? Say in cognos cubes. And the reasoning?

It isn't as much an issue with the number of dimensions, but rather capacity and load times.

I haven't worked with Cognos in the last few years, but if things haven't changed much, a Powerplay 'cube' is stored as a file of discrete values, somewhat like a fact table, and the actual cube is built in memory. The size of the data file tends to grow proportional to the amount of detail, not expotentially based on the number of dimensions. I do not know the structure of the in-memory cube, however, if it is a multidimensional data structure of some sort, the size would be more related to the number of dimensions, cardinality of the dimensions, and data sparcity. So a reasonable sized cube is a function of available memory on the client machine and how long they are willing to wait for the cube to load. What works for you may not work for someone else.

I also think that with cognos cubes, the hierarchies have to be appropriately designed. Lets say you have 2 levels in a hierachy. Say the top level has 4 values. Each of the Top Level values drills down to 5,000 values. When the user goes to drill down from the top level to the next level, the cube is going to try to deliver a huge amount of data and potential overwhelm everything.

You can go with a lot of shallow dimensions or a few deep dimensions. And big jumps in the volume of values at a level will cause problems.

TM1 might be handling this a little better.