Introduction

  So far, the main purpose of IP-tables has been the efficient support for the optimisation of partitioned temporal join processing. In this chapter, we want to show that the scope and the applicability of IP-tables goes beyond that. In fact, IP-tables can be considered as a general metadata-structure  that can be used in a wide range of temporal query optimisation  techniques, especially in those that require (semi-) optimal partitioning of temporal data over a timestamp  attribute, such as physical data partitioning or balancing temporal index structures.

In this chapter, we concentrate on an optimisation issue which is not directly connected to temporal data partitioning but that can nevertheless be efficiently supported by IP-tables, namely the selectivity estimation of temporal conditions. Selectivity estimation is a powerful way to predict the result sizes for many operations. On the basis of these predictions, an optimiser can then take many performance-relevant decisions, such as

• Which algorithm should be used to perform the respective operation most efficiently? Often there is a wide range of algorithms available, with some being suitable for low selectivities, others for high selectivities. The join operation is a good example for this: in chapter 3 we saw a variety of join algorithms with the nested-loops approach being suitable only for high selectivities whereas sort-merge or hash algorithms were more efficient in the case of low selectivities.
• Which is an efficient order in which operations should be processed? A query that incorporates various operations is often translated into an operator tree. In this tree, operations with low selectivities (i.e. small result sizes) are moved near the leaves (i.e. these operations are performed first) and operations with high selectivities towards the root. This has the advantage that the initially-processed operations already discard large amounts of data and therefore reduce the sizes of the intermediate results. This can reduce the overall performance by a considerable amount.
• How many resources are required to process the query? From a system's point of view, the optimiser might want to consider the impact of the query on the system's overall performance: if the query is `heavy' (i.e. very resource-consuming), for example, then other, `lighter' queries might be granted priority.

We restrict ourselves to the discussion of selectivities of temporal join conditions as these have been the focus of major parts of this thesis. This restriction is also sufficient to prove the wide applicability of IP-tables. As defined in section 3.4, the selectivity factor  (or selectivity  , for short) of a join is

 

(64)

In section 11.2, we classify temporal join conditions. Actually this is a summary of what was discussed in section 4.1 but this time with a slightly different emphasis. In section 11.3, we derive equations that allow either the exact calculation, or a reasonably accurate estimation of the size of temporal join results and therefore also the selectivity of the corresponding temporal join condition.