Query Processing in Embedded Control Programs

Abstract

An embedded control program can be viewed as a small main-memory database system tailored to suit the needs of a particular application. For performance reasons, the program will usually de- fine concrete low level data structures to encode the database, which in turn must be understood by anyone who needs to develop or modify the program. This is in contrast with the data independence that can be achieved by using a database system. However, because of space and performance requirements, the use of current database technology is not likely to be feasible in this setting. To explore one obstacle to this, we have developed a query optimizer that compiles queries on a conceptual schema to native Java or C code that navigates low level data structures. Of crucial significance is that an arbitrary collection of such structures, perhaps already devised for an earlier version of the control program, can be given as a part of the input to the optimizer. We present an overview of the underlying algorithms that are used to accomplish this. The algorithms are based on a novel resource bounded plan generation mechanism in which integrity constraints abstracting the definition of stored views are applied to source queries to extend the search space of possible query plans. We also report on some preliminary experimental results that suggest generated code navigates concrete data structures with an efficiency comparable to code written directly by expert programmers.