You can roughly compute the resulting data set size with the
("length of record 1" + "length of record 2") * (NObs1 + Nobs2 - NMatchingObs)
You could get closer with two test runs. One with all the row and limit to the joining columns (to find the total number of observations) and one with just a few rows from each table (to find the length of one row).
However, if you are having space difficulties, you also need to look at the amount of work space that SQL will use. For that, you need the SAS companion for the OS and the SAS manual for SQL. If you are on a mainframe, SQL may use an external sort procedure depending on what your local admins set up.
Personally, I find the OS compression to be superior to SAS compression if that is available to you.
Doc Muhlbaier
Duke