I understand that once rules are added you are supposed to use skipcheck and feeders to make the cube more efficient to run
However, there does seem a high chance of errors/(frustration!) when using feeders.
Is there a limit on cube/cell size when it the difference in performance is material between skipcheck on or off
What I'm getting at is not including skipcheck simplifies the cubes and rules but slows the calcs down, but by how much?. If it's a few seconds then it may not be worth the extra coding required
As an EP consultant, I know we have recommendation for cube sizes when designing. Are there similar guidelines for TM1?
Cheers
David
Hi David,
The two most important criteria are the size of the cube (nr of cells, based on number of dimensions and the number of elements in the dimensions) and most of all density (the percentage of populated/non-zero cells) of the cube.
The influence of Skipcheck/Feeders is most noticable (for usability even required) on big, sparse cubes. If you have a small 2- or 3-dimensional cube with a lot of populated cells (f.i. lookup cubes) the use of Skipcheck/Feeders will not bring much.
Michel
Thanks for the reply Michel
Is there any kind of diagnostic tool you can run? to :
a) check sparsity and
b) check ram /speed - you could then run with/without feeders and see the impact
Cheers
David
Quote from: DavidSmith on 14 Jul 2010 04:43:59 PM
Is there any kind of diagnostic tool you can run? to :
a) check sparsity and
b) check ram /speed - you could then run with/without feeders and see the impact
A/ Yes. To calculate theoretical maximum cube size (number of possible leaf cells) multiply together the number of N elements in each dimension that comprise the cube. To calculate sparsity you first need to know the number of populated leaf cells, to get this turn the performance monitor on, then sparsity = number of leaf populated cells / total leaf cells
B/ The only way to know the performance difference is to do the experiment and calculate some rules with and without skipcheck and feeders. (Note: to get a good idea the vie being calculated should be quite consolidated or high level otherwise you're really not doing an appropriate test.)
Feeders are
essential for optimising performance. Rules with feeders can calculate very quickly, rules without skipcheck can calculate very slowly. (As you are essentially placing a similar constraint on TM1 that EP has of having no sparsity control.) As a rough rule of thumb I would say that feeders are essential for any cube with more than 3 dimensions but it is all dependent on dimension size(s) and rule complexity and what is deemed acceptable response time.