Joins - done in NDB or MySQL?

[Gregor Melhorn]

I read quite some time now in forums and documentation about NDB clustering, but couldn't find an answer to the following:

If I'm right, then it's advisable to use several mysqld clients per ndb node, as a single mysqld would get a bottleneck regarding the speed possible with ndb. Somewhere on this forum I read that frequently a mysqld client is placed on the same machine with an application server, resulting in several machines with a mysqld and an app server each, all accessing one ndb node.

Problem: If I'm using large index based table joins with few results, will the "join operation" be done internally in NDB, or would this mean huge result sets being transfered over the net to my mysqld, which then does the join?

My present db structure means I have one table with about 100 000 rows and another with about 24 million rows. These are joined by joins using index, resulting in each row of the small table in the result set be joined with about 2000 rows in the large table.

As not all rows in the large table matching the criteria are needed since they won't get joined as there is no matching row in the smaller table, how are the join rows selected?

Does this set up mean that all possible rows are transfered to the mysqld client and part of them are discarded after the join? Or does the join operation only select matching rows in the large table, so there is no unwanted data transfer? Or is the join done at storage engine level?

best regards
Gregor