The Oracle8 Server Concepts manual says about the execution plan used in
star transformation queries:
"After the relevant fact table rows have been retrieved ... they are
joined with the dimension tables to produce the answer to the query."
I find that this happens regardless of whether columns from all
dimensions are in the select list of the query. Even if only one column
from a dimension table is in the select list, ie the select statement
groups by this column and sums up a column in the fact table, the
execution plan includes steps to join the relevant fact table rows with
<all> the dimensions. Since the where clause has already been processed
at that stage (using joins with the dimensions and ANDing the bitmap
indexes), why is this necessary? Anyone have any ideas? It seems to slow
down the query processing, since the explain plan output shows that
considerable cost is associated with these steps, being hash joins
between the dimensions and relevant fact table rows.
TIA
Francois
Received on Wed Jan 06 1999 - 12:00:06 CST
