A novel algorithm for transforming row-oriented databases into column-oriented databases

Abstract

With the development of cloud computing and distributed system, more and more applications are started to migrate to the cloud to use its computing power and scalability. These systems require interaction of heterogeneous data and applications that involves databases that store the data of the common domain under different representations. The interactions are result of schema mappings and transformation that is also called data migration. The above interaction of different systems and integration is known as information integration. Two well-known approaches to information integration are data exchange and data integration. In the first approach, data is extracted from multiple heterogeneous sources, restructured into a common format, and finally transformed into a target schema. In second approach, several local databases are queried by users through a single, integrated global schema. Mappings are determined to know, how the queries that users pose on the global schema are to be reconstructed in terms of the local schemas. In this thesis, a novel approach is proposed that transforms a relational database into a column-oriented database. Out of various available column-oriented databases like HBase, Cassandra etc., we have considered HBase for the study, which is an open-source distributed database and similar to BigTable which is the database proposed and used by Google. For relational database, we have considered PostgreSQL. The transformation procedure is divided into two phases. First relational schema is transformed into HBase schema based on the data model of HBase. Using four rules, HBase schema is extracted from PostgreSQL schema which could be further utilized to develop an HBase application. In the second phase, relationships between elements of two schemas are expressed as a set of nested schema mappings, which could be further used to create a set of nested queries that transform the PostgreSQL database into the HBase database representation automatically.