Parallelism in DB Operations (cont)
Parallel hash join
- distribute partitions to different processors
- partition 0 of R goes to same node as partition 0 of S
- join phase can be done in parallel on each processor
- then results need to be merged
- very effective for equijoin
Fragment-and-replicate join
- outer relation R is partitioned (using any partition scheme)
- inner relation S is copied to all nodes
- each node computes join with R partition and S
|