Physical Data Warehouse Design - Data Warehouse Systems: Design and Implementation - page 260

smaller joins that occur between each of the partitions, producing significant
performance gains, which can be even improved taking advantage of parallel
execution.
7.7.2 Managing Partitioned Databases
Partitioning also improves the job of database and data warehouse adminis-
trators, since tables and indexes are partitioned into smaller, more manage-
able pieces of data. In this way, maintenance operations can be performed
on these particular portions of tables. For example, a database administrator
may back up just a single partition of a table instead of the whole one. In addi-
tion, partitioned database tables and indexes induce high data availability.
For example, if some partitions of a table become unavailable, it is possible
that most of the other partitions of the table remain on-line and available, in
particular if partitions are allocated to various different devices. In this way,
applications can continue to execute queries and transactions that do not
need to access the unavailable partitions. Even during normal operation, since
each partition can be stored in separate tablespaces, backup and recovery
operations can be performed over individual partitions, independent from
each other. Thus, the active parts of the database can be made available
sooner than in the case of an unpartitioned table.
7.7.3 Partitioning Strategies
There are three most common partitioning strategies in database systems:
range partitioning, hash partitioning, and list partitioning.
The most usual type of partitioning is range partitioning , which maps
records to partitions based on ranges of values of the partitioning key. The
temporal dimension is a natural candidate for range partitioning, although
other attributes can be used. For example, if a table contains a date column
defined as the partitioning key, the January 2012 partition will contain rows
with key values from January 1, 2012, to January 31, 2012.
Hash partitioning maps records to partitions based on a hashing
algorithm applied to the partitioning key. The hashing algorithm distributes
rows among partitions in a uniform fashion, yielding, ideally, partitions of the
same size. This is typically used when partitions are distributed in several
devices and, in general, when data are not partitioned based on time since it
is more likely to yield even record distribution across partitions.
Finally, list partitioning enables to explicitly control how rows are
mapped to partitions specifying a list of values for the partitioning key. In
this way, data can be organized in an ad hoc fashion.