Parallel of User-Defined Functions

Abstract

Our main contribution in this chapter is to show how a broad class of user-defined functions can be processed in parallel. This class includes both, user-defined scalar functions and user-defined aggregate functions. To this aim we propose a framework covering both the necessary interfaces that allow the appropriate registration of userdefined aggregate functions with the ORDBMS and their parallel processing. Parallel computing of user-defined aggregate functions is especially useful for application domains like decision support (e.g. based on a data warehouse that stores traditional as well as non-traditional data, like spatial, text or image data), as decision support queries often must compute complex aggregates. For example, in the TPC-D Benchmark 15 out of the 17 queries contain aggregate operations [99]. In addition, if scalar functions with a global context are processed in parallel, caution is needed in order to get semantically correct results. Our framework can help in this case, too. Furthermore, we show that some aggregate functions can easily be implemented, if their input is sorted, and they can thus profit from parallel sorting.