And finally run the second merge join

While intra-partition parallelism can speed up a variety of queries, it has another purpose as well: workload balancing. Even if we have created a per-fectly balanced system, imbalances will occur as queries are executed. For example, earlier we said date of sale was not a good candidate partitioning key, as sales volumes are very uneven from day to day. However, users still need to report on daily sales. This means data will be redistributed to group by day, and the temporary work files will become imbalanced. DB2 can rec-ognize these dynamic imbalances and compensate for them by assigning the appropriate number of parallel tasks to each partition.

Of course, some queries rely on indexes to satisfy their queries. Consider a star schema, where the fact table is accessed through a variety of indexes around the dimension tables. Traditionally, a DBMS would look at the most

Finally, DB2 supports parallelism at the I/O level. Modern RAID (redun-dant array of independent disks) controllers support parallel I/Os across their platters, and DB2 fully leverages this feature.

Partitioning Data Storage