@luke_e , To answer your questions, the answer is, sadly, it depends. With that said, sparsity is not a necessarily a bad thing. Take a look at these two articles written by David Smith which dispels the sparsity myths (The Truth About Sparsity). You are correct, many people use subsets to decrease the overall module/line item cell count, but there could be diminishing returns if you have to do those same or similar calculations on numerous subsets. Remember the C in DISCO (calculation modules). It is best to use more space for calculation modules (where the logic is clear/understandable/maintainable) where the calcs are done one time versus multiple times all over the place. Simple is often better. Also, with the use of SYS modules (and booleans), you can create logic with early exits which will help the models overall performance. On to your specific questions: Correct, utilizing a SYS module can help with this where a subset is not really needed. Correct, it depends but really you need to look at what the subsets are really used for? If for calculation purposes, then examine the logic and see if it is repeated for other Leases or other subsets. If the subset is used dashboards and helping users pick the appropriate member, that is a different story. Evaluating if the cell count is 0. Short answer is no. What people often forget and what I believe you are alluding to, is when you create a line item and put logic in it, that formula is not kicked off one time, but is really kicked of the number of cells of the list item. For example, if I have a line item with a formula parent(item(Product)) and that line item contains 1,000,000 cells, that means parent(item(Product)) is getting kicked off 1,000,000 times. There is no need for that which is where the SYS modules come into play. It is much better to have a SYS module with the parent already defined in a "flat" module (only dimensionalized by Product) and just reference that line item. It is much better to reference a line item 1 million times than it is to calculate it 1 million times. it depends on the list size. For this example, let's use the formula FindItem(). If the list is 100,000 members and the cell count of line item is 1,000,000, the formula is reading a list of 100,000 members 1 million times. So again, this goes back to #3. Early exists are your friend here as well as booleans. Again, subsidiary views aren't bad from a performance standpoint, but they are bad in that the logic is now hidden away in a module. For example, let's say Johnny and Mary are both model builders and are working on similar, if not the same logic. If Johnny puts the logic in subsidiary view, Mary will not know where to look, so she will likely recreate it. So now, the model is doubling up the logic which takes time. On the other hand, if Johnny had created a module for the calculation, then Mary would be able to able to see it, add to it if her logic was slightly different, and just reference it. Again, this not only speaks to performance of the model, but it cleans up the model so it is easier to maintain. Hopefully I have answered your questions and this helps. Rob
... View more
