Database Reference
In-Depth Information
aged. We say that a structure class
A
strongly depends on another
structure class
B
if a change in selection of
B
often results in a change
in that of
A
. Otherwise, we say that
A
“weakly” depends on
B
.It
can be argued that weak dependencies are likely to exist because a new
physical structure is normally introduced to help areas in which existing
structures do not apply or do not perform well. Figure 9.5 summarizes a
dependency table for various types of physical design structures. Mutual
strong dependencies are dicult to
break
and thus are better handled
by using a combined approach. If only
B
strongly depends on
A
,wecan
iteratively search
A
followed by
B
, so that
B
is properly influenced by
A
.
Weakly coupled components can be scheduled separately in any order.
In principle, it is even possible to go through a series of iterations of all
physical structures until some convergence condition is met. In this way,
the hybrid approach can break the implementation of different physical
structures into smaller components while capturing the most important
interdependencies among them.
9.5 Summary
Each physical design structure added to a DBMS requires new algorithms to
automatically recommend configurations that leverage the new functionality.
In this chapter we discussed techniques that can be used to automatically
recommend:
•
Horizontal and vertical partitioning
•
Data cubes
•
Multidimensional clustering
9.6 Additional Reading
We next provide some references that can be used as a starting point to
obtain more detailed information on the different techniques and algorithms
discussed in this chapter. Rao et al. describe in detail different techniques to
recommend hash-based horizontal partitioning on parallel DBMSs.
7
Agrawal
et al. focus on integrated recommendation of indexes, materialized views,
and horizontal and vertical partitioning.
2
In this chapter we briefly com-
mented on a basic approach to select subcubes to materialize. More details
on this approach can be found in the work of Shukla et al.
8
We note, how-
ever, that several subsequent techniques improve and extend the original ap-
proach in different ways (e.g., considering indexing
3
and multicube systems
4
).
Search WWH ::
Custom Search