Optimization of multi-join queries in shared-nothing systems

Abstract

This paper proposes a semi-greedy framework that generates an optimal pipelined plan for multi-join queries. The generated plan comprises several pipeline, each performing several joins. The framework determines the 'optimal' number of joins to be performed in each pipeline. The decisions are made based on the cost estimation of the entire processing plan. Two optimization algorithms are extended under the framework. Our study show that the new algorithms outperform their counterparts that are not extended.