机构地区: 中国科学院计算技术研究所
出 处: 《微电子学与计算机》 2003年第8期1-5,30,共6页
摘 要: 因为需要存取大量的数据,计算datacube要花费大量的时间和存储空间。本文研究了使用便宜的PC机群计算datacube的方法。我们使用多维数组作为存储结构,并将数组划分成若干个分片,对每个分片进行压缩以节约存储空间、减少存取时间和增量维护时间,分片被分布到不同的处理机。我们提出了一个新的流水线组织方法以及对分片建立索引的思想,大大减少了外排序的代价和存取磁盘的次数。实验结果表明我们的算法具有一定的伸缩性。 The computation of data cube is a very expensive op-eration because a lot of data have to be accessed.We investi-gate the approach of using low cost PC cluster to compute data cube.In our approach,multidimensional array is used to store data.We partition multidimensional array into fragments and distribute them among machines in cluster.Fragments are com-pressed to save storage space,access time and incremental maintenance time.We propose a novel method to organize pipeline and an idea of creating fragment index file.The cost of outer sort and the amount of accessing disk are dramatically re-duced.The experiment results show the algorithm is of scalability.
领 域: [自动化与计算机技术] [自动化与计算机技术]