cylondata/cylon

Implement broadcast join

nirandaperera opened this issue · 2 comments

nirandaperera commented

Implement broadcast join. This will be important where one relation is significantly smaller than the other. Then, it is efficient to broadcast/ all-gather the smaller table rather than shuffling.
All the required building blocks are already available. Should support all join variations (left, right, inner, outer).

Steps:

  1. Broadcast/ all-gather right table (this should be the small one)
  2. Do a local join
chathurawidanage commented

Note: In a left join, you can only broadcast the right table. In a right join, you can only broadcast the left table regardless of the smallest table.

  • 👍1
nirandaperera commented

Note: In a left join, you can only broadcast the right table. In a right join, you can only broadcast the left table regardless of the smallest table.

Precisely!