Efficient OLAP algorithms on GPU-accelerated Hadoop clusters

In the time of big data, on-line analytical processing (OLAP) is an important method to process massive data. In order to realize a system with the capacity of both high storage and high computing power, Hadoop and GPU are both applied in OLAP. In general, three cores of OLAP determines the efficiency of OLAP analysis, which are aggregation of multi-dimensional data, pre-calculation of multi-dimensional data set (Cube) and connection of dimension table and fact table. For the purpose of boosting efficiency, this paper presents optimizing algorithms for each core. Beginning with aggregation on single machine, this paper firstly designs the GPU-based aggregation algorithm. Then, GPU-based Cube algorithm is introduced to accelerate pre-calculation, using inverted index to shrink computation amount. Finally, with new-designed dimension table connecting algorithm and query algorithm, GPU-based OLAP analysis algorithm is presented. Along with corresponding experiments and results, each algorithm shows their ability of boosting efficiency, optimizing GPU-based OLAP analysis on Hadoop.