An MPI-based Framework for Proessing Spatial Vector Data on Heterogeneous Distributed Systems
An MPI-based Framework for Proessing Spatial Vector Data on Heterogeneous Distributed Systems
复制标题
基于 MPI 的异构分布式系统空间矢量数据处理框架
DOI：
--
复制
发表时间：
2016
期刊：
影响因子：
--
通讯作者：
Taiki Shimbo
中科院分区：
文献类型：
--
作者：
Kouichi Araki;Taiki Shimbo
地理信息系统（Geographic Information System，GIS）被广泛应用于地貌分析、灾害地图绘制、疏散路线规划等领域，由于空间矢量数据包含大量顶点数据，需要大量的空间处理，因此部分GIS采用异构分布式系统，包括异构机器和云基础设施。然而，空间分析人员和研究人员很难有效地进行空间处理，因为他们需要考虑负载平衡。此外，学习并行编程，如消息传递接口（MPI），也是必需的。在本文中，为了减轻这种负担，我们提出了一个基于MPI的框架，在异构分布式系统中执行空间矢量数据的空间处理。我们的框架包括一个执行时间预测器，隐藏MPI编程的封装器和一个包装库。我们的实验结果表明，我们的框架是12.9倍的速度比顺序处理在我们的GIS组成的亚马逊EC2和本地集群，而我们的库的源代码步骤的数量几乎是相同的顺序版本。
Geographic information system (GIS) is utilized in geomorphic analysis, hazard mapping, evacuation route planning and so on. Some GISs employ heterogeneous distributed systems consisting of dissimilar machines and cloud infrastructures because spatial vector data, which has the large number of vertex data, requires heavy spatial processing. However, it is difficult for spatial analysts and researchers to efficiently perform the spatial processing by such GISs because they need to consider load balance. Additionally, learning parallel programming, such as message passing interface (MPI), also is required. In this paper, to alleviate such burdens, we present an MPI-based framework that performs the spatial processing for the spatial vector data in the heterogeneous distributed systems. Our framework consists of an execution time predictor, schedulers and a wrapper library for hiding MPI programming. Our experimental results show that our framework is 12.9 times faster than sequential processing in our GIS consisting Amazon EC2 and a local cluster while the number of source code steps with our library is almost identical to that of the sequential version.