Using partitions may decrease search time when searching through a lot of data from minutes to seconds, from hours to minutes. Storing information partitioned per month/year and retrieving it knowing which month/year you want from millions of records is more efficient than an index on all those months and years. Unfortunately changing to partitions afterwards almost always needs application changes, (I actually never tried an sql_profile…).
When looking for a solution using partitioning, I came across another problem. They were using a function in the one and only predicate in the query and the fact is that when a database uses partitioning for storing data, using a function on the partitioned column will the optimizer searching all data, not the partition. Understandable, because with the function you are searching for derived values, not the actual values in the partition. Oracle must search all the partitions and can not ‘prune’ the partitions. Determining and using a partition is called Partition Pruning.
The solution to be used here is called Virtual Column Partitioning and as a coincidence and nice extra they didn’t even had to change the application!
[Read the rest of this entry…]