DISTINCTCOUNT combined with ADDCOLUMNS + FILTER

Question

Hello,
I've been going nuts trying to optimize fhe following expression in PowerPivot :
CALCULATE(DISTINCTCOUNT([Identifiant Personne Physique]);(FILTER(
ADDCOLUMNS(
'Contrats';"Nb Contrats P-1 T";
SUMX(DISTINCT(Contrats[Identifiant Personne Physique]);
SUMX('Contrats';VALUE(Contrats[Base dénombrement P-1])))
);[Nb Contrats P-1 T] = 0 && [Base dénombrement Flux Entrée] = "1")))
The idea is to count distinctly the persons (without removing the context filters) whose SUM([Nb Contrats P-1 T]) is equal to 0 and flag [Base dénombrement Flux Entrée] = 1.
It does the job but I have 2 issues :
1. It is extremly slow when I change the values of the slicers (20 millions rows in the model). Does anyone know who to improve the performances of that query ?
2. The measure returns a NULL values when the field [Identifiant Personne Physique] is not used as a dimension in my Pivot tables. What did I miss to make it possible ?
Thanks a lot for any help,
Fred

Answer 1

The slowness you are seeing is due to your trebly nested iterators.
FILTER() and SUMX() both iterate over rows one by one. Further, FILTER() is forced to single-threaded mode no matter what, and I cannot recall right now if SUMX() does the same.
Working from the inside out, SUMX('Contrats'....) is being evaluated for every single row in DISTINCT(Contrats[Identifiant Personne Physique]), which in turn is being evaluated for every row in 'Contrats'. This whole expression is then fed to FILTER() which
means that we step row by row through Contrats and perform the test you have defined.
Without knowing more about your model, it will be difficult to help tune this measure. If you have a sample data set you can share, with the working measure included, that would be very helpful.
As a first pass, it seems you may be able to remove the innermost SUMX() and replace it with a simple SUM() (it may have to be wrapped in a CALCULATE(). This would look like:
SUMX(
DISTINCT( Contrats[Identifiant Personne Physique] )
; CALCULATE( SUM( Contrats[Base denombrement P-1] ) )
I suggest this because SUMX( <table>; VALUE( <column> ) ) is telling the engine to step row by row through <table> and add the number on the current row in <column> to a running total and then return the total.
Again, based on my naive first impression, looking at this it seems it may be possible to simply remove the outer SUMX() as well leaving you with just the CALCULATE( SUM( ... ) ) from above.
Again, this is based on a naive first look, and I expect you have a good reason to have eliminated the suggestions above.
I don't think anyone (I know I, at least) will be able to provide much help without a sample as I expressed above.