Data Threaded Query Evaluation in Shared - EverythingEnvironments

In this paper, we present data threaded execution, a new strategy to exploit pipelining and intra-operator parallelism in a shared-everything environment. Data threaded execution does neither suuer from execution skew caused by workload estimation errors, nor from the discretization error of processor scheduling as it appears in conventional strategies. Further more, data threaded execution avoids any startup and shutdown execution delays. We evaluate our new strategy by simulation. Our simulation results show, that data threaded execution outperforms conventional strategies like segmented right-deep execution due to better utilization of parallel processing resources.