Fast MDX query for one row per-hierarchy on shared dimension - union

Question

Working on a SSAS multidimensional cube that includes a UserGroups dimension made up of multiple hierarchies. Each hierarchy classifies users as belonging to a particular group or not. Users can belong to more than one group.
Trying to generate a table with one row for each hierarchy in the dimension. The best approach that I have come up with is the following:
WITH
MEMBER [Measures].[Group Name] AS
CASE
WHEN [UserGroups].[Group A].CurrentMember.MemberValue <> 'All' THEN "Group A"
WHEN [UserGroups].[Group B].CurrentMember.MemberValue <> 'All' THEN "Group B"
-- More user groups...
WHEN [UserGroups].[Group T].CurrentMember.MemberValue <> 'All' THEN "Group T"
ELSE "Error"
END
SELECT
{[Members].[Group Name], [Measures].[User Count], [Measures].[Revenue]} ON 0,
Order(
UNION(
[UserGroups].[Group A].&[True] *
{[UserGroups].[Group B].[All]} *
-- More user groups...
{[UserGroups].[Group T].[All]},
[UserGroups].[Group A].[All] *
{[UserGroups].[Group B].&[True]} *
-- More user groups...
{[UserGroups].[Group T].[All]},
-- A bunch more blocks so that there is one for each row.
[UserGroups].[Group A].[All] *
{[UserGroups].[Group B].[All]} *
-- More user groups...
{[UserGroups].[Group T].&[True]}
),
)
ON 1
FROM [Finances];
This succeeds in generating the expected table:
Group Name User Count Revenue
All All ... True Group T 90 1800
... ... ... ... ... ... ...
All True ... All Group B 20 400
True All ... All Group A 10 100
However, the query is very slow. Any given row can be calculated in isolation in about 30 seconds. When combining rows using the UNION statement, each additional row seems to result in an exponential increase in computation time rather than the desired linear increase in computation time.
Why is this increase in time observed?
Is there a way to instruct the underlying engine to compute each row in isolation?
Is there simply a better way to issue this query?

Answer 1

I guess you've already experimented with rearranging the query - if you take UNION(...) out of the SELECT and make it a named SET in a WITH clause does that help?
WITH
SET [RowSet] AS
UNION(
[UserGroups].[Group A].&[True] *
{[UserGroups].[Group B].[All]} *
-- More user groups...
{[UserGroups].[Group T].[All]},
[UserGroups].[Group A].[All] *
{[UserGroups].[Group B].&[True]} *
-- More user groups...
{[UserGroups].[Group T].[All]},
-- A bunch more blocks so that there is one for each row.
[UserGroups].[Group A].[All] *
{[UserGroups].[Group B].[All]} *
-- More user groups...
{[UserGroups].[Group T].&[True]}
)
MEMBER [Measures].[Group Name] AS
CASE
WHEN [UserGroups].[Group A].CurrentMember.MemberValue <> 'All' THEN "Group A"
WHEN [UserGroups].[Group B].CurrentMember.MemberValue <> 'All' THEN "Group B"
-- More user groups...
WHEN [UserGroups].[Group T].CurrentMember.MemberValue <> 'All' THEN "Group T"
ELSE "Error"
END
SELECT
{[Members].[Group Name], [Measures].[User Count], [Measures].[Revenue]} ON 0,
Order(
[RowSet],
)
ON 1
FROM [Finances];
What is the purpose of the ORDER function? I cannot see what you are ordering by?
Again just a shot in the dark but I think sometimes I've noticed differences in efficiency between using UNION and using the + operator - so another option:
WITH
SET [RowSet] AS
{
[UserGroups].[Group A].&[True] *
{[UserGroups].[Group B].[All]} *
-- More user groups...
{[UserGroups].[Group T].[All]}
}
+
{
[UserGroups].[Group A].[All] *
{[UserGroups].[Group B].&[True]} *
-- More user groups...
{[UserGroups].[Group T].[All]}
}
-- A bunch more blocks so that there is one for each row.
+
{
[UserGroups].[Group A].[All] *
{[UserGroups].[Group B].[All]} *
-- More user groups...
{[UserGroups].[Group T].&[True]}
}
MEMBER [Measures].[Group Name] AS
...
...
Apologies I don't have definites - the above are just options that might get lucky!! Best to discuss in as much detail as you can with Greg as he will be able to give you solid advise on possible amendments to the cube design.