Optimising data storage for performance in a modern binary relational database based on a triple store

This thesis introduces a new approach to understanding the issues relating to the efficient implementation of a binary relational database built upon a triple store. The place of the binary relational database is established with reference to other database models, and a detailed description of a new triple store implementation is presented, together with a definition of the architecture. The use of a model, which reflects the performance of the triple store database, is described, and the results of performance investigations are presented. In the first, the use of more than one sort order in the triple store database is analyzed, and the use of two sort orders is found to be optimal. In the second, the effect of compression in the triple store is considered, and compared with other approaches to compressing the non-index portion of a database management system. In conclusion, the model successfully predicts the effect of using two sort orders, and this was confirmed upon subsequent incorporation into the database. It is also found that significant performance gains can be made by the use of compression in the triple store. It is shown that by extending the compression algorithm even greater gains could be made. In addition, it is found that by keeping the design of the database as simple and pure as possible, a foundation for a variety of higher level views can be achieved, leading to the possibility of the triple store being used as the foundation for new databases.