Temporal Data Warehousing: Approaches and Techniques

Introduction

At the core of most business intelligence applications, data warehousing systems are specialized in supporting decision making. They have been rapidly spreading within the industrial world over the last decade, due to their undeniable contribution to increasing the effectiveness and efficiency of the decisional processes within business and scientific domains. This wide diffusion was supported by remarkable research results aimed at improving querying performance, at refining the quality of data, and at outlining the design process, as well as by the quick advancement of commercial tools.

In the remainder of the paper, for the sake of terminological consistency, we will refer to a classic architecture for data warehousing systems, illustrated in Figure 1, that relies on three levels:

Figure 1.

Three-levels architecture for a data warehousing system

• 1.

  The data sources, that store the data used for feeding the data warehousing systems. They are mainly corporate operational databases, hosted by either relational or legacy platforms, but in some cases they may also include external web data, flat files, spreadsheet files, etc.

• 2.

  The data warehouse (also called reconciled data level, operational data store or enterprise data warehouse), a normalized operational database that stores detailed, integrated, clean and consistent data extracted from data sources and properly processed by means of ETL tools.

• 3.

  The data marts, where data taken from the data warehouse are summarized into relevant information for decision making, in the form of multidimensional cubes, to be typically queried by OLAP and reporting front-ends.

Cubes are structured according to the multidimensional model, whose key concepts are fact, measure and dimension. A fact is a focus of interest for the decisional process; its occurrences correspond to events that dynamically occur within the business world. Each event is quantitatively described by a set of numerical measures. In the multidimensional model, events are arranged within an n-dimensional space whose axes, called dimensions of analysis, define different perspectives for their identification. Dimensions commonly are discrete, alphanumerical attributes that determine the minimum granularity for analyzing facts. Each dimension is the root of a (roll-up) hierarchy that includes a set of levels, each providing a way of selecting and aggregating events. Each level can be described by a set of properties.

As a consequence of the fact that the decisional process typically relies on computing historical trends and on comparing snapshots of the enterprise taken at different moments, one of the main characterizations of data warehousing systems is that of storing historical, non volatile data. Thus, time and its management acquire a huge importance. In this paper we discuss the variety of issues, often grouped under term temporal data warehousing, implied by the need for accurately describing how information changes over time. These issues, arising by the never ending evolution of the application domains, are even more pressing today, as several mature implementations of data warehousing systems are fully operational within medium to large business contexts. Note that, in comparison with operational databases, temporal issues are more critical in data warehousing systems since queries frequently span long periods of time; thus, it is very common that they are required to cross the boundaries of different versions of data and/or schema. Besides, the criticality of the problem is obviously higher for systems that have been established for a long time, since unhandled evolutions will determine a stronger gap between the reality and its representation within the database, which will soon become obsolete and useless (Golfarelli et al, 2006).