软件DSM机群上并行大规模地理图像处理系统ParGIP
PARGIP:A Prototype System for Parallel Large-Scale Geographical Image Processing on Software DSM Cluster
-
摘要: 基于共享虚拟存储 (sharedvirtualmemory ,SVM )PC机群的大规模并行地理图像处理原型系统ParGIP(parallelgeographicalimageprocessing)采用Client Server计算模型 ,通过软件分布式共享存储 ①(softwaredistributedsharedmemo ry ,softwareDSM )中间层将PC机群组织成一个逻辑上共享内存的并行计算平台 ,地理图像处理可以充分利用ParGIP提供的大共享内存和并行处理能力来提高性能 ,缩短处理周期 ,从而解决传统单机串行方式下地理图像处理中内存匮乏和计算能力不足的问题 ParGIP还进一步将机群中各个结点上分布的磁盘组织起来 ,提供地理影像库所需的海量存储空间和并行I/O能力 测试结果表明 :ParGIP的 8机并行I/O带宽达到 10 2 6MB/s ,典型的图像处理算法获得了接近线性的加速比Abstract: ParGIP uses the client-server computing model to support large-scaled geographical image processing Based on JIAJIA, a software distributed shared memory （DSM） system, geographical image processing takes advantage of huge cluster memory and executes in parallel Based on NFS protocol, ParGIP organizes distributed disks to store massive geographical image data sets It also makes use of parallel file I/O technique to obtain high I/O bandwidth Algorithms-based processing makes client submit processing tasks to server step by step and get results interactively Performance evaluation shows ParGIP can provide good parallel computing properties and high I/O bandwidth for geographical image processing
下载: