Data Modifications and Versioning in Trio

Abstract

The field of uncertain databases has recently attracted considerable interest. Many motivating applications for uncertainty rely fundamentally on improving the quality of data over time, through modifications, as additional information becomes available, e.g., after analysis (as in scientific data management systems), with user feedback (as in {\em pay-as-you-go} data integration). Incorporating data modifications, while still serving applications' needs to access and query older data (as in hypothetical databases), necessitates light\-weight versioning for such applications. This paper presents the first DBMS for uncertain data that supports data modifications and versioning. Our work is in the context of {\em Trio}, a project at Stanford for managing data {\em uncertainty} and {\em lineage}. We introduce SQL-based language constructs for data modifications and lightweight versioning in Trio's query language. We present an extended Trio data model, {\em ULDB$^v$}, and show how primitive modifications are applied to it, yielding versioned relations with uncertainty and lineage. We show that Trio's lineage feature enables an elegant approach to query answering in ULDB$^v$. We give efficient algorithms for propagating data-modifications to materialized views. We have incorporated the data modification and versioning capabilities in the Trio system, and validate our techniques through experiments.