When partitioning is employed, a table changes from a physical object to a virtual concept. There isn't really a table anymore, just a set of partitions. Since all of the partitions must share the same attribute and constraint definitions, however, it is possible to deal with the set of partitions as if they were a single table. The storage parameters, such as extent sizes and tablespace placement, are the only attributes that may differ among the partitions. This situation can facilitate some interesting storage scenarios, such as hosting infrequently accessed partitions on a CD jukebox while the heavily-hit data partitions reside on disk. You can also take advantage of Oracle's segmented buffer cache to keep the most active partitions in the keep buffer so they are always in memory, while the rest of the partitions can be targeted for the recycle or default buffers. Additionally, individual partitions may be taken offline without affecting the availability of the rest of the partitions, giving administrators a great deal of flexibility.
Depending on the partitioning scheme employed, you must choose one or more columns of a table to be the partition key. The values of the columns in the partition key determine the partition that hosts a particular row. Oracle also uses the partition key information in concert with your WHERE clauses to determine which partitions to search during SELECT, UPDATE, and DELETE operations (see Section 10.6 later in the chapter for more information).