* Sample Build relation
* Use samples to compute approx
partitioning vector
* Redistribute Build relation using
approx partitioning vector
* Build in-memory hash table.
Overflow tuples put into buckets
such that each bucket fits in memory
* Redistribute Probe relation
using the same partitioning vector
* For each tuple of the probe
relation probe the in-memory hash table
outputting a join result for each match
* Use weighted partitions and
Subset-Replicate
* No. of Virtual Procs
= Multiple of No. Real procs