Incorporating Other Physical Structures - Automated Physical Database Design and Tuning - page 157

Chapter 9
Incorporating Other Physical
Structures
Over time, new releases of database management system (DBMS) products
introduced additional types of access path methods to deal with scenarios
that did not perform as eciently as expected. Examples of such new fea-
tures include various forms of data partitioning, materialized views, cubes,
and hash, bitmap, partial, Extensible Markup Language (XML), columnar,
spatial, and multidimensional indexes. Each new such structure comes along
with its own benefits and overheads, and it is generally challenging to under-
stand when and how to leverage its functionality. For that reason, after each
new such structure is introduced, there are attempts to generalize the physical
design problem to automatically recommend ecient configurations for input
workloads. Analogous to the structure of Chapters 4 to 6, there are three
main questions associated with recommending a new kind of physical design
structure. First, we need to define what is the space of alternatives that we
should consider and whether these would interact with other features in the
DBMS. Second, we require a mechanism to evaluate the hypothetical benefit
of a given configuration in the search space. Finally, we need ecient proce-
dures to traverse the search space to quickly obtain the desired configuration.
In this chapter we explore some recent work on automatically recommending
some physical design structures that deal with scenarios involving large data
sets and complex query workloads.
9.1 Data Partitioning
Data partitioning is an important aspect of physical design and can have a
big impact on performance. Partitioning can additionally be used in parallel
database systems, where partitions are stored in different autonomous servers
and can even improve server manageability (if tables and indexes are parti-
tioned in the same way, common operations such as backup/restore can be
drastically simplified). In this section we summarize some work in the area of
automated recommendation of partitioning under different scenarios.
157