SGER: Staged Query Processing

点击翻译按钮获取中文摘要

In today's data management environment, query processing is spread across multiple components, with each middleware component doing some of these tasks, and with some tasks done several times, which we term staged query processing. The potential presence of multiple stages restricts the information available to the middleware component, with particular impact on query optimization. The conventional approach, of cost-based query optimization, in which numerous query plans are enumerated and the cheapest one selected, is not appropriate when detailed cost metrics are not available and when the later stages perform their own, complex optimization. A new approach, termed cost-based graph rewriting, is used to transform a query in one query language into an efficient output query in perhaps a different language. A novel idiomatic cost estimator provides cost parameters to the rewriter, enabling the rewriter to determine the processing that should take place within the component, and the processing that should be performed by later stages. This estimator is more coarse-grained than conventional estimators; in particular, it does not cost individual relational operators in a query plan. The goal of this exploratory research project is to develop an initial prototype utilizing these concepts for a temporal middleware component translating TSQL2 into SQL, and to evaluate the efficacy of this approach by targeting at least two different commercial DBMSs and demonstrating that the middleware component can reflect the differences of these DBMSs in the generated code.