The promising results of parallel sequential scan encouraged us to include parallelism in other operators of PostgreSQL as well. Continuing the fervour, we witnessed several patches proposed for new parallel operators in the last few months. Extending parallelism for the access methods, parallel bitmap heap scan, parallel index, and parallel index-only scans were proposed. Next, to utilise the parallelism at the higher levels of the plan tree came parallel hash join and parallel merge join. Finally, to not miss the benefits of parallelism at the topmost nodes involving aggregates and sorts, the idea of gather-merge was suggested.
We will start this presentation with a brief overview of the design ideas of these new operators. We will move forward with the performance analysis of these operators on the standard decision support benchmark queries of TPC-H. We will present the performance of these operators for the scale factor 20, plus, to evaluate their performance for the ‘Big data’ scenarios we will also present the results on 300 scale factor. Next, we will show the effect of different parameter settings on these performance numbers. Next, the cumulative effects of these operators will be analysed. Finally, the presentation will be concluded with a discussion on the scope of further parallelism in PostgreSQL.