首页> 外国专利> With component based parallel application checkpoint the excessive partitioning (overpartitioning) system and manner in order to augment

With component based parallel application checkpoint the excessive partitioning (overpartitioning) system and manner in order to augment

机译:借助基于组件的并行应用程序检查点,过度的分区(over- partitioning)系统和方式可以扩大

摘要

(57) Abstract In order the job which is run by the computer program, for checkpoint to be done more frequently, furthermore they are two manners in order to partition to the small piece. At the beginning, parallel task one where q has early partition or starts input data set, p splits with multiple data sets, into partition due to a some combination (partitioning device/cutting device), of the partitioning element the instance of the component program on each one of partition, the set of the multiple output files, the data set to which each set is partitioned p of the data runs one or creates is considered. This invention new excessive partitioning is applied to this kind of task in order like below to create the task which is done. (1) the partitioning device the q n*p splits entry to partition the excessive partitioning device displacement vis-a-vis integer excessive partitioning coefficient n. (2) the component program serial n is run at execute phase, p of the component program instance is run at optional point in time. At each phase, each instance of the component program, reads one excessive partition of the input data, creates one partition of the output data. (3) n each one of execute phase being last, the system is stationary, it is possible to do checkpoint configuration. The 1st example in order to create the intermediate file which excessively is partitioned, using the known partitioning device program, the communication channel, and the cutting device program, explicitly excessively partitions the input data. The 2nd example dynamically excessively partitions the input data by arranging the component program in order to read the continual subset of the original input data continually.
机译:(57)<摘要>为了由计算机程序运行的工作,为了更频繁地执行检查点,而且为了划分成小块,它们是两种方式。在开始时,并行任务一,其中q具有早期分区或开始输入数据集,p由于多个组合(分区设备/切割设备)而将多个数据集拆分为分区,该分区元素是组件程序的实例在每个分区上,考虑多个输出文件的集合,将每个集合分区到的数据集p的数据运行一个或创建一个。本发明将新的过度分区应用于这种任务,以便如下所示创建完成的任务。 (1)分区设备q n * p拆分条目以将过多分区设备位移相对于整数过量分区系数n进行分区。 (2)组件程序序列号n在执行阶段运行,组件程序实例的p在可选时间点运行。在每个阶段,组件程序的每个实例都读取输入数据的一个过多分区,创建输出数据的一个分区。 (3)在执行阶段的每个阶段最后,系统处于静止状态,可以进行检查点配置。为了创建使用已知的分区设备程序,通信通道和切割设备程序而被过度分区的中间文件,第一示例明确地对输入数据进行了过度分区。第二个示例通过安排组件程序来动态地过度划分输入数据,以便连续读取原始输入数据的连续子集。

著录项

相似文献

  • 专利
  • 外文文献
  • 中文文献
获取专利

客服邮箱:kefu@zhangqiaokeyan.com

京公网安备:11010802029741号 ICP备案号:京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有
  • 客服微信

  • 服务号