栅格数据处理中邻域型算法的并行优化方法

Parallel optimization methods for raster data processing algorithms of neighborhood-scope

随着并行计算的成熟,众多数据密集型的栅格处理算法亟需利用并行计算来缩减执行时间。针对其中一类邻域型算法,构建了用于估计是时间代价的串行/并行时域模型,分析了各个组成的代价影响因素,提出了降低数据I/O代价的并行I/O方法和降低数据通信代价的光圈预测方法。实验证明,所提的两个优化方法可以使邻域型栅格处理算法的并行程序更加充分地利用并行计算资源,进而在一般并行化的基础上进一步提升其并行性能。

As parallel computing has become mature and practical, data intensive raster data processing algorithms are desiderating parallel computing technologies to reduce the running time. The objectives of this research focuses on the parallelization of neighborhood-scope algorithms. the sequential/parallel temporal model was developed, the affecting factors of each component of the temporal model were analyzed, and two optimization methods were proposed, which can further promote the parallel performance of neighborhood-scope algorithms： the Parallel I/O method that can reduce the data I/O cost; and the Halo Prediction method that can reduce the data communication cost. Experiments verified the effectiveness and efficiency of the proposed optimization methods, which can further promote the parallel performance by making the parallel algorithmic program fully take advantage of parallel computing resources.