大型多维数据集合中并行I/O数据存储新策略研究

  • 摘要: 对大型多维数据在分布式系统的多个磁盘之间进行并行I/O分布存储进行了研究,将已有的数据循环拆分法从适用于二维数据扩展到多维数据中,提出了一种新的启发式多维数据循环策略,即基于访问步长值Hi与访问长度M互质的启发式策略(HPPHM),实验结果表明了新算法在并行度和鲁棒性等性能方面都具有优越性.

     

    Abstract: Its focused on the multidimensional dataset that is allocated in the multi-disks among the distributed processing system by using the parallel I/O technology. A new multidimensional data cyclic declustering policy named HPPHM is proposed aiming to data retrieve based on scope by extending existing cyclic policy to multidimensional dataset from adapting to two-dimension. The experimental result has demonstrated the efficiency of new strategy not only in parallel degree but also in robust.

     

/

返回文章
返回