Performance question

[Andy]

Hello there,

maybe a stupid question, but my colleague and I are in discussion about the performance using subsets or not.

While opening the cube, which elements will be calculated?
All dimension elements from every dimension or only the elements from the dimension from the chosen subsets?
Is that maybe also a point where to raise the performance?

PS:
My quesion has nothing to do with feeders. We only want to if there is a differents in performance using subsets or not.

Thanks in advance.

[David Usherwood]

You don't open the cube, you open a view of the cube. When you click on the cube, you open the view called 'Default' which someone may have adjusted from TM1's first cut default view. And the view may have named subsets for some or all dimensions, or there may be explicit selections, if not you'll get the default subsets.

TM1's calculation algorithm will calculate and cache what it needs to deliver the content of the view - including elements in the paging dimensions which aren't directly visible on your screen.

So subsets are only relevant insofar as they are attached to the view you are opening.

HTH

[Andy]

Hi David,

thanks for the fast respond and the detailed answer.

So, you can only increase performance using subsets for dimensions witch are not hierarchical.
In a hierarchical dimension the n-level is neccessary to calculate the c-level.

If you use use a subset in a flat dimension with for example 3 elements instead of all (for example 20 elements) then only the 3 elements will be calculate in cache.

Thanks again.

[Alan Kirk]

If you use use a subset in a flat dimension with for example 3 elements instead of all (for example 20 elements) then only the 3 elements will be calculate in cache.

You said that it wasn't a question about feeders, but you haven't indicated whether it's a question about rules-calculated values or not.

For standard, non-rules calculated elements, then there is nothing to calculate or cache. The values are loaded into memory at server startup and are simply displayed from the memory contents.

If they're rules-calculated values then yes, obviously calculating three elements will be faster than calculating all 20 (on the proviso that the three are not dependent on some of the other 17), but in reality unless you are talking about some very badly written and/or over-complex rules I seriously doubt that you would ever be able to tell the difference.

For that matter that also applies to consolidations. Yes, not having to (initially) calculate consolidations will be faster than having to calculate them, but given the efficiency of the consolidation calculation engine whether that difference would be realistically measurable is another matter.

[Andy]

Thanks Alan.
That is the answer I needed.