Including the load balancing issue in the optimization of multi-way join queries for shared-nothing database computers

Dynamic load balancing issues in multiway join operations is discussed. In particular, four multiway join query optimizers have been implemented and a simulator developed to investigate the effect of skewed tuple distribution on these techniques. Dynamic load balancing is found to very critical to the performance of shared-nothing (SN) systems, particularly when the system is large. Since load balancing becomes more difficult for larger systems, one should exploit interjoin parallelism in order to limit the number of processing nodes used for each join operation. The skew issue must be considered during query optimization to achieve the best performance. The load balancing techniques proposed can easily be extended to support interquery parallelism in a multiuser environment.<>