Customer Reviews

Temporal Data & the Relational Model (The Morgan Kaufmann Series in Data Management Systems)

5 star:		(4)
4 star:		(0)
3 star:		(0)
2 star:		(1)
1 star:		(0)

 

 

This book presents perhaps the first rigorous analysis of the issues surrounding the use of temporal data in a relational database. Temporal data may be thought of as information contained in relations (i.e. tables) which changes over time. This situation most often arises in data warehouse environments, but the authors rightly point out that current data (e.g. in an Operational Data Store) must be integrally related to the history of that data over time.

This book is not for the faint-hearted. Its approach is quite rigorous and abstract. In order to comprehend this book in any meaningful fashion, the reader will have to develop an understanding of predicate calculus, which is a specialized version of set theory and its logic. Previous exposure to "Foundation for Object/Relational Databases: The Third Manifesto," by Date and Darwen, is highly recommended. While their approach is necessary, it does entail endless dry proofs of the prerequisite material needed to develop a general theory of temporal data. The authors leave few issues not covered in an effort to present a fully rigorous analysis of the issues. The book practically begs for a companion book to present a simplified summary of their findings and to contrast them with current data warehouse practices.

The book begins with a review of relational concepts and an introduction to Tutorial D, which was first presented in "The Third Manifesto." The authors then develop a theory of time in the database, which is based on earlier work by Lorentzos. Their approach is to create a new data type for timestamped data and, more importantly, for intervals of time. After developing a relational model for these temporal constructs, they propose a generic design for current, historical and current and historical databases using the Supplier-Parts Database which readers will likely have been exposed to in previous writings by Date. They develop methods to implement integrity constraints, queries and updates, and make a valid comparison between stated times, which are beliefs about the data, and logged times, which are facts the database records about these beliefs. Since our beliefs about time can change, this is an important distinction.

The overall approach is to develop a model of data which is highly related to an audit trail, which should be viewed distinctly from the concept of a database transaction log. Unfortunately, in order to implement the model, an interval timestamp is required for every attribute, and historical data must be presented in the Sixth Normal Form, which they introduce as a decomposition of each historical record into a relation for each data attribute, with its associated timestamp.

While their approach is entirely correct, realization of their ideas would be a prohibitive task in current databases. A database that provided an automated construction of current and historical values would be necessary to implement their model in a consistent and economical fashion. Their solution is in many ways radically different than the typical "Facts and Dimensions" approach of star-schema data warehouses which are the most popular current tool for the analysis of historical data. The utility of the dimensional "slice and dice" approach to data analysis is beyond doubt, and the authors would need to provide some mechanism to allow for this type of analysis to be performed under their alternative approach in order for a commercial database vendor to consider supporting their propositions.

Still, there is much original insight to be gleaned from the author's analyses which will reward anyone who has the talent and perseverance to thoroughly read this book. I especially like the way in which the book provides an integrated means of analysis for current and historical data. Far too many other authors in the data warehouse literature have ignored at their own peril the undeniable link between transactional and analytic data. I can only hope that database vendors are sufficiently inspired by this book to implement their eminently reasonable and logically correct answers to the questions of temporal data.

My thanks go to the authors for this book. Lat year I introduced material on temporal databases to my undergraduate students, realising the importance of this topic. However I did not feel comfortable with the books that deal with temporal database or confident that the prevailing approach was sound. Here the authors make crystal clear every decision as to how to model things. The relational model once again provides a foundation which leads to confidence even though the full temporal datbase picture is not yet in sight. As I read the book I found I got engaged in the subject and started asking my own questions (which were often answered later on) - a sign of a good book!. Now I'm using it in earnest to revise my lecture notes for next session, and am appreciating the structure of the book: the order in which topics are introduced and the use of examples. This is not just (maybe not primarily) a student text book, but I would recommend reading it to anyone teaching temporal database.

There can be no denying, I believe, that insofar as current major Database Management Systems (DBMS) are concerned, temporal data represents a significant problem: it is simply not possible to declaratively constrain the integrity of temporal data in SQL databases (or non-temporal data, even). I am not alone I am sure in having had to deal with duplicate, triplicate, quadruplicate, etc., records covering, or abutting, or overlapping on the same temporal periods. Not to mention the hoops that must be jumped through to reliably manipulate those same data.

Date, Darwen, and Lorentzos have produced a formidable work here on applying some badly needed rigid logic to the whole sphere of temporal data within databases. And that rigid logic is afforded by the Relational Model. They consider three variants on the temporal data theme as vehicles for explanation and demonstration: 1. Semitemporal with current data only, 2. Temporal with current and historical data held within the same relvars (tables), 3. Temporal with current and historical data split into separate relvars.

Temporal data is a complex area, so this book has, inevitably, had to get 'down and dirty' with the detail, but the authors are clear and comprehensive throughout. A thorough familiarity with the Relational Model will help, as will any previous experience of their Relational language `Tutorial D', though they go though both in the first two introductory chapters.

We would indeed be much better off were the DBMS vendors to take serious note of the powerful logical arguments laid forth in this volume, and far from interpreting their reflections (criticisms) on NULLs, etc., as 'political', I would see them only as further reaffirmations of the principles that have led them to invest so much endeavour and thought into the problems and very real deficiencies of data modelling and integrity, specifically in regard to the Relational Model. And for that we should indeed be thankful.

The point of view of a French Database Architect and Administrator:

Un ouvrage magistral, traitant à fond et avec une extrême rigueur du sujet plus complexe qu'il n'y paraît des données temporelles (et plus généralement des données intervallaires de tout type), tant au plan de leur description que de leur manipulation et de leur intégrité. On ne le lit certes pas comme un roman, tant s'en faut, mais les concepteurs de bases de données se doivent de l'étudier, progresser sans se hâter et ne pas se décourager (ce qui exigera de leur part de faire pas mal d'exercices, crayon en main et tube d'aspirine à portée). Mais une fois l'étude de l'ouvrage menée à son terme, le concepteur aura tous les atouts pour véritablement traiter de la dimension temporelle des données, lors de l'élaboration des diagrammes de classes, modèles conceptuels et logiques de données ; il se rendra compte qu'il aura jusqu'ici procédé le plus souvent de façon trop intuitive et fruste, faisant que les développeurs auront dû ensuite pallier du moins mal possible, à l'occasion de l'organisation de leurs requêtes et programmes. Le développeur (averti) est lui aussi concerné par l'étude des opérateurs de manipulation des données temporelles, crayon en main à son tour et sans se précipiter pour ne pas lâcher prise. Mais on n'a rien sans rien, et de cet ouvrage on tirera des leçons qui seront autant de gages de succès dans le développement des projets dans lesquels la dimension temporelle des données est omniprésente. Il faut aussi ne pas oublier que la prise en compte des données temporelles dans un langage quel qu'il soit s'accompagne de bien des pièges, au point que la commission SQL de l'ISO a dû recaler les propositions d'extension du langage SQL qui lui ont été soumises en la matière (SQL/Temporal).

This book was a big disappointment. The topic of temporal databases is one that should be of great interest and widespread application. However, this book won't advance the science or the application in my opinion.

The authors seem to continually have an ax to grind, to the extent that it detracts from the good ideas they do have. I don't recall having seen so many political opinions in a technical book in a long time: it appears that no other author or researcher has ever done anything right! It seems to me that they want to rename a bunch of database concepts just for the sake of having their own names on them.

Most readers are probably better off to keep their money.