Enhancing functionality of relational database systems
Abstract
This thesis explores ways to enhance functionality and performance of relational database systems. Specifically, it presents the design and implementation for supporting abstract data types (or classes), composite objects and indexing methods for nested objects in a relational database system, and compares the performance of such a system with relational and object-oriented database systems. To explore functionality enhancements to a relational DBMS, we supported abstract data types, composite objects, persistent pointers, and replication of composite objects, on top of the RAID distributed relational database system prototype. The resulting system called O-Raid was developed so that it retains all the features of relational data base systems, while being augmented with features of a general purpose object-oriented programming language. This approach allows co-existence of relations and objects and reuse of the relational database system software. To study the performance, we ran the OO1 engineering benchmark on the extended relational DBMS, O-Raid, and its underlying relational DBMS, Raid. We identified the benefits and overhead in the extended relational DBMS. Our studies showed that the extended relational database system, O-Raid, incurred only a small amount of overhead (under 15%) for insert and select queries. The traversal of links in memory for the extended relational DBMS O-Raid took less time than the traversal of data in Raid which was through expensive relational join operations. We re-evaluated three index schemes (nested index, path index, and multi-index) applicable to queries on nested attributes. Among these, we found that a multi-index scheme was best supported in the object-oriented or extended relational DBMS environment. Multi-index schemes not only provided a better balance between retrieval and update costs than did the nested index or path indices, but they also scaled well for update when the number of indices increased. We proposed a multi-index design that reused the single-table index structures already present in a DBMS. Our performance study extended the previous models by permitting attributes to be multi-valued as well as single-valued. We also suggested that a combination of nested index and multi-index schemes offered a feasible solution to the support of queries on nested objects.
Degree
Ph.D.
Advisors
Bhargava, Purdue University.
Subject Area
Computer science
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.