基于图形处理器的Cuboid算法

近年来,基于图形处理器的通用计算获得了广泛关注,并在多个领域取得了进展.内存OLAP减少了磁盘I/O,但基于单核或多核CPU的计算能力及cache miss成为新的性能瓶颈,从而无法保证好的效率.而图形处理器由于其众多核和高带宽能够很好地适应OLAP计算特性.通过图形处理器来加速任一cuboid的计算,从而提高整个内存OLAP系统的性能.提出了基于图形处理器的分块并行算法,并对算法进行了优化及讨论了数据稀疏和数据分布倾斜等不同条件下的算法.算法通过扩展可以突破内存限制,组成磁盘、内存、显存三级流水线,适应海量数据计算;同时算法也可以作为计算整个cube的基础.通过实验比较,基于图形处理器的算法明显优于四核CPU算法.

GPU-Based Cuboid Algorithm

In recent years,general-purpose computing based on graphics processor units(GPU)has received wide attention and achieved progress in many fields.In-memory OLAP reduces disk I/O,but limited computing power of single-core or multi-core CPU and cache miss become the new performance bottleneck;thus in-memory OLAP can not guarantee a good efficiency.Because of many-core and high-bandwidth,GPU adapts to the characteristics of OLAP calculation well.GPU accelerats the calculation of any cuboid,thereby the performance of system is improved.GPU-based chunk parallel Cuboid algorithm is improved;it is optimized and discussed on sparse and skew data.The new algorithm can overcome the memory limit and extend to a disk,memory,GPU memory pipeline to fit massive data;at the same time,it can be the base for calculating the whole cube.Experiments show that GPU-based algorithm is more efficient than the quad-core CPU algorithms.