Databases Reference
In-Depth Information
Figure 8.1 Physical clustering of data: (a) data blocks clustered by date of month, and (b) data
blocks unclustered.
difficult to design even for experienced database designers and industry experts. As
noted by Jagadish, Lakshmanan, and Srivastava [1999], the design of optimal cluster-
ing in this context has combinatorial complexity, having an extremely large search
space. For hierarchical dimensions (such as City
→
→
Country) the search space is
typically doubly exponential in the number of dimension hierarchy levels! Range
dimensions (dimensions that are typically queried by range, such as salary or date)
more often have a continuous domain, and further exacerbate the problem since there
is a near infinite set of ways in which they can be clustered since the range can be sub-
divided by various degrees of granularity. For example, looking at a date dimension,
the data could be clustered by day, or by week, or by month, etc. Similarly, a salary
dimension could be clustered by dollar, or by every ten thousand dollars, and of course
by any granularity in between.
State
