User-Defined Table Operators

Abstract

Parallel database technology makes it possible to handle ever-increasing data volumes. However, this is not sufficient to process complex queries on large databases fastly. For queries that must apply complex algorithms to the data and especially for those that correlate data from several tables, it is essential to enable an efficient and completely parallel evaluation of these algorithms within the DBMS. For example, as we have shown in chapter 5, new tailored join algorithms can increase the performance for certain operations like spatial joins, etc. by orders of magnitude. But, as we have already pointed out there, it is not yet possible for third-party developers to implement efficient user-defined join algorithms in current commercial ORDBMS. In fact, one cannot implement any new database operators. UDF cannot be used to implement new operators, as they are invoked by built-in database operators. The limitation of UDTF is obvious: although they can produce an entire output table, they can only have scalar arguments. Hence, UDTF are helpful in accessing external data sources [21] etc., but cannot be used to implement new database operators like new join algorithms.