Abstract

Management of uncertain data in numerous real life applications has attracted the attention of database and artificial intelligent research communities. This has resulted in development of new database management systems (DBMS) in which uncertainty is treated as first class citizens. We follow a different approach in this thesis and develop a system (to which we refer as DBMS with Uncertainty, or UDBMS) which is capable of representing and manipulating uncertain data at the application level on top of a standard relational DBMS. Compared to the first approach which treats uncertainty as its first class citizens, the proposed approach may be considered as “light weight” because it is built upon existing database technologies. As the underlying uncertainty formalism, we consider the Information Source Tracking (IST) method, which is essentially probabilistic. We extend the standard SQL language with uncertainty (to which we refer as USQL), to express queries and transactions in our context. The query processing and optimization techniques are extended accordingly to take into account the presence of uncertainty. To evaluate the performance of UDBMS, we conducted extensive experiments using USQL queries and IST relations obtained by extending the standard TPC-H benchmark queries and generated data. We compare and discuss the two approaches mentioned for uncertainty management. Our results indicate that the performance of the proposed UDBMS is reasonably good when the relations involved can be loaded completely into the main memory.