Subset-Replicate

• In the presence of highly skewed data it may be necessary to map a single data value to multiple partitions. Eg. In {1,1,1,1,1,1,1,1,2,3} an equal - sized partitioning would map {1,1,1,1,1} to P0 and {1,1,1,2,3} to P1.
• In this case one must be careful to ensure that all possible tuples of the join are generated.
• It is sufficient to send all tuples with the repeated join attribute from one relation to all the sites to which that attribute value is mapped and to send each tuple of the other relation to exactly one of the sites with the repeated values. This technique is called subset-replicate.
  Eg. If R was the relation above, all Si's that include 1 will be sent to both subsets of relation R.