An object-oriented framework for the parallel join operation

Abstract

We propose an object oriented framework for one of the most frequent and costly operations in parallel database systems: the parallel join. The framework independently captures a great variety of parameters, such as different load balancing procedures and different synchronization disciplines. The framework addresses DBMS flexibility, configuration and extensibility issues, via the instantiation of known algorithms and facilities for the introduction of new ones. The framework can also be used to compare algorithms and to determine the execution scenario an algorithm is best suited for. Related algorithms are grouped in families, suggesting a taxonomy.