Quality Management in Database Systems-A Thesis Proposal

In this proposal, we argue that quality parameters such as delay and uncertainty are more important performance metrics than processing time, which is the only concern in query optimization in traditional DBMS. We explore specific problems on how to maintain quality in multimedia and stream databases, with a focus on the latter. Data stream processing have attracted a great deal of attention from the database community. The continuous feature of both data and queries in data stream management systems (DSMSs) place great demand on system resources. However, queries can be processed with different levels of quality such as timeliness, reliability, and uncertainty. In this proposal, we discuss the problem of how to maintain quality in query processing in DSMSs. We focus on two types of quality metrics: general Quality-of-Service parameters that apply to all types of queries and uncertainty parameter for probabilistic queries. A widely-used approach to maintain QoS (especially tuple delays) in DSMS query processing is load shedding, i.e., dropping data. Current load shedding solutions utilize simple, intuitive ideas in determining the time and amount of load to be discarded and do not work well in the presence of system/environmental disturbances. We propose a solution based on feedback control theory with significantly improved long-term performance. For probabilistic queries, an important optimization is to have the least uncertainty in query results under resource contraints. We propose to approach this problem by exploiting the temporal dependence in stream data.