Next: Load Scheduling
Up: Partitioning Techniques
Previous: Partitioning Techniques
- Eg. Consider two 10000 tuple relations in which the join attribute 1 appears
1000 times in each relation and all other attribute values appear only once.
- Dividing over 10 processors using equal sized ranged partitioning will cause
both relations to map their 1000 tuples each with 1 as the join attribute
value to P0 thus causing it generate 1000000 tuples.
- This is an example of severe
join product skew.
- The solution lies in choosing many more partitions than there are
processors. This is known as Virtual Processor Partitioning.
- In the Eg we could
use 100 virtual processors each with 100 buckets (subranges). Now, there would
be 100 1's in each of 10 virtual processors.
- Map these virtual processors to
the 10 (different) real processors in a load balanced manner.
DBMS
1999-03-11