I made some tests with SSAS 2005, because, when I used dimenions with
unary operator, the performances was very bad.


I created a cube :
2 regular Dimensions
Dim 1 : 3 200 lines
Dim 2 : 150 lines

Facts : 400 000 lines

ProClarity is the client taht I use to query the cube.



Find below the details of the tests


Test 1 :
----------
No UNARY operator
Full process of dimensions and cube : Server memory (MSMDSRV) = 64 Kb,
virtual memory : 61 Kb
Open cube : server memory = 64 Kb, virtual memory : 61 Kb. Time = less
than 2 seconds
Drill to bottom level of the 2 dimensions : server memory = 195 Kb,
virtual memory : 192 Kb
Queries are very fast.


Test 2 :
----------
UNARY at the bottom level of Dim 1.
Full process of dimensions and cube : Server memory (MSMDSRV) = 70 Kb,
virtual memory : 68 Kb
Open cube : server memory = 70 Kb, virtual memory : 68 Kb. Time = less
than 2 seconds
Drill to bottom level of the 2 dimensions : server memory = 195 Kb,
virtual memory : 192 Kb
Queries are very fast.


Test 3 :
----------
UNARY at the bottom level of Dim 2.
Full process of dimensions and cube : Server memory (MSMDSRV) = 64 Kb,
virtual memory : 61 Kb
Open cube : server memory = 68 Kb, virtual memory : 65 Kb. Time = less
than 2 seconds
Drill to bottom level of the 2 dimensions : server memory = 195 Kb,
virtual memory : 192 Kb
Queries are very fast.


Test 4 :
----------
UNARY at the bottom level of Dim 1.
UNARY at the bottom level of Dim 2.
Full process of dimensions and cube : Server memory (MSMDSRV) = 64 Kb,
virtual memory : 61 Kb
Open cube : server memory = 834 Kb, virtual memory : 831 Kb. Time :
more than 1 minute, just to open the cube.
Drill to bottom level of the 2 dimensions : server memory = 962 Kb,
virtual memory : 959 Kb; Very fast
But if I navigate thru the level of the Dim 2, the memory used increase
: server memory = 1104 Kb, virtual memory : 1102 Kb



Test 5 :
----------
No UNARY on Dim 1.
Dim 2 is replaced by a parent-child dimension with 180 members. There
is a UNARY operator at the bottom level.
Full process of dimensions and cube : Server memory (MSMDSRV) = 65 Kb,
virtual memory : 62 Kb
Open cube : server memory = 68 Kb, virtual memory : 65 Kb. Time = less
than 2 seconds
Drill to bottom level of the 2 dimensions : server memory = 193 Kb,
virtual memory : 190 Kb
Very fast


Test 6 :
----------
UNARY at the bottom level of Dim 1.
Dim 2 is replaced by a parent-child dimension with 180 members. There
is a UNARY operator at the bottom level.
Full process of dimensions and cube : Server memory (MSMDSRV) = 68 Kb,
virtual memory : 66 Kb
Open cube : server memory = 834 Kb, virtual memory : 831 Kb. Time :
more than 1 minute, just to open the cube.
Drill to bottom level of the 2 dimensions : server memory = 226 Kb,
virtual memory : 223 Kb. very fast
But navigate thru the level of the Dim 2 (parent-child) is very slow :
more than 50 seconds for each drill.


Conclusion :
------------------
If you use just one dimension with unary operator. It's OK? You have
good performances. But if you use 2 or more dimensions with unary
operator, the performance become very bad, especially if the dimensions
are regular : the memory used is very large.


If somebody have information or solution, it can help me

Thanks

Check out http://www.entellicache.com We developed this technology
over 5 years of implementation in very large environments. Works very
well. Different architecture from cubes, but is generally more
flexible and produces better results. Works with Cognos, Reporting
Services or anything that speaks OLEDB. Does not do MDX.