An XML-based methodology for parametric temporal database model implementation

Abstract

Parametric data model is one of dimensional data models. It defines attributes as functions, modeling a real world object into a single tuple in a database. Such one-to-one correspondence between an object in the real world and a tuple provides various advantages in modeling dimensional data, avoiding self-joins which frequently appear in temporal data models which fragment an object into multiple tuples. Despite its modeling advantages, it is impractical to implement the parametric data model on top of conventional database systems because of the data model’s variable attribute sizes. However, such implementation challenge can be resolved by XML because XML is flexible for data boundaries. In this paper, we present an XML-based implementation methodology for the parametric temporal data model. In our implementation, we develop our own XML storage called CanStoreX (Canonical Storage for XML) and build the temporal database system on top of the CanStoreX.

Introduction

Conventional databases are only capable of storing and querying the current perception of reality and relationship among objects, such database systems exclude old data values when the data is updated. Unlike conventional databases, a temporal database is capable of storing evolution of data, thereby allowing users to examine complete object histories (Bhargava, 1989). Applications of temporal databases include finance, organizational record-keeping, and scientific applications (Jensen and Snodgrass, 1999).

Temporal databases are one of active research areas in the database community and tremendous research work has been done by many researchers. According to Jensen and Snodgrass’s survey (1999), more than 2000 research papers on temporal databases had been published. There are many different types of temporal data models and they have their own merits in their specific applications.

In the temporal database literature, we can find three types of timestamps—instants, intervals, and temporal elements. Based on timestamps, temporal data models are termed point-based models, interval-based models, and temporal element-based models, respectively.

It is important to emphasize that only temporal elements are legitimate domains of objects and events in the real world. However, a temporal element cannot be represented using a fixed length because it can have any number of intervals. Therefore, in many approaches, intervals or instants are used to timestamp fragments of objects that are stored in multiple fixed-length tuples. Typically, such timestamps are attached at tuple level rather than value level.

In our implementation, we consider the parametric temporal data model which uses temporal elements for timestamps. The parametric temporal data model defines attributes as functions of time. Because of this, it can model an object in the real world into a single tuple in databases. Such one-to-one correspondence between an object and a tuple can avoid self-joins which are inevitable in temporal data models which fragment an object into multiple tuples. However, such modeling advantage increases the implementation complexity because of variable attribute sizes. In particular, it is difficult to adapt the parametric temporal data model within relational database systems, which are the most popular approaches for temporal database implementation.

In order to meet the implementation challenge, we use XML and develop our own XML-based storage technology called CanStoreX (Canonical Storage for XML). CanStoreX is capable of paginating large XML data into small chunks of pages. Many artifacts, such as parse and expression trees are also represented in XML, which are human readable and more reliable by removing the reliance on traditional linked list-based trees. We also show how to utilize XML in developing a query execution engine. The primitives for the query language are implemented using the DOM parser. We use nine queries to validate the implemented system.

Our XML-based implementation approach is different from other approaches in that we do not use existing conventional database systems. It is a native temporal database system built on top of our own storage technology. Furthermore, our approach overcomes the implementation challenges of the parametric temporal data model, which has debated since the mid 1980s. Therefore, we can expect that our implementation paradigm provides an elegant solution for the implementation of complex data models which have difficulties to use existing database systems.

The organization of this paper is as follows. Section 2 briefly discusses various approaches to implementing temporal database systems. Section 3 discusses XML’s usability and some possible directions for the parametric temporal data model implementation. Section 4 discusses the parametric temporal data model. Section 5 introduces ParaSQL—the query language of the parametric temporal data model. Section 6 explains our XML-based implementation methodology. Section 7 discusses the system performance for a suite of ParaSQL queries. Section 8 summarizes our findings.