I have successfully tested two analytic views, each selecting a fact table (joined to attribute views). Function and performance were good, as expected. These fact tables are identical except certain measures exit in one fact table but not the other. To provide a single view of the data, I created a calculation view that joins the two analytic views. With queries that are relatively non-selective, e.g.,,
WHERE calendar_year_name = 'CY2012'
I hit out of memory errors. The peak memory usage in the server is consistent with these errors.
Queries that are more selective (e.g. below) execute without error and but take considerable longer than the queries against analytic views.
WHERE calendar_year_name = 'CY2012'
AND department_name = 'Computers'
AND category_name = 'Total Personal Computers'
AND region_name = 'South America'
I wish to avoid joining the two fact tables in the ETL process. Is joining analytic views in a calculation view the best solution? The impression I get is that data is being copied from one layer/engine to another during this process (this the out of memory errors). Is that impression reasonable.
One other detail that might be useful In the calculation view joins I've created joins between all attribute columns. This appears to be necessary in order to get those columns in the output of the join. The tables really only need to join on a few columns (the primary keys of each fact table). Are my joins (on all attributes) the correct practice? Would the number of joins have any bearing on the out of memory errors?
Thanks.