Abstract: Join is one of the most time-consuming and most frequently used operations in database query processing,so it has great importance to improve the speed of join operation.The array-based manycore processor is one of the most important classes of manycore processors and has great parallelism,and can be used to accelerate parallel computing.In this paper,we designed and optimized an efficient algorithm called multi-layer partition hash join algorithm on array-based manycore architecture.Our algorithm achieves high performance by multi-layer partitioning which reduces the times of main memory accessing,and by partition rearrangement which eliminates the influence of data skew.Experimental results show that the performance of multi-layer partition hash join algorithm on DFMC(Deeply-Fused Many Core) processors is 8.0x times higher than that of the fastest hash join on CPU-GPU coupled-architecture,which de-monstrates that array-based manycore processors have advantages in data query processing.

Key words: Array-based manycore,Hash join,Data skew,Parallel algorithm