It really depends. Sometimes it's possible to slurp relatively small subsets in. Sometimes you've already got the data loaded in memory for other reasons. Somet

It really depends. Sometimes it's possible to slurp relatively small subsets in. Sometimes you've already got the data loaded in memory for other reasons. Sometimes (especially in long-running ETL stuff) you can come up with a way better query plan than a query-by-query approach could achieve. But yeah, usually, don't be clever and don't spaff the contents of the database across a network just to do a join.