表格处理分区卷影副本大小

本文介绍了表格处理分区卷影副本大小的处理方法，对大家解决问题具有一定的参考价值，需要的朋友们下面随着小编来一起学习吧！ 问题描述

如果我有一个5GB的立方体，我知道该规则通常是完全发生过程的2倍，所以阴影副本可以在那里直到处​​理完成。我的问题是，如果为了简单起见我将多维数据集分区为大约1GB的块，如果I 单独处理一个分区，那么我需要6GB还是需要10GB内存？

If I have a cube that is 5GB, I know the rule is typically 2x that for a process full to happen so the shadow copy can be there until the processing is done. My question is if I have the cube partitioned into about 1GB chunks for simplicity, if I process just 1 of the partitions individually and at a time, would I need 6GB or still need the 10GB RAM?

推荐答案

典型规则更像是2.5x，因为您需要原始副本+卷影副本+工作空间。工作空间的大小很难准确知道，因为它取决于您的数据的配置文件和并行操作的数量，但是正在处理的对象的额外50％是粗略的经验法则。

The typical rule is more like 2.5x as you need the original copy + shadow copy + working space. The amount of working space is hard to know precisely as it depends on the profile of your data and the number of parallel operations, but an extra 50% of the objects being processed is the rough rule of thumb.

要回答您的问题，您只需要6Gb（或6.5Gb以允许工作空间）。 SSAS仅创建作为当前事务的一部分进行修改的对象的卷影副本。因此，如果您只处理一个1Gb分区，然后提交事务，则只需要该分区的卷影副本。在提交阶段，SSAS然后通过它的主版本映射并更改它的指针，以便它们使用卷影副本作为新的数据源，然后一旦版本映射更新，新结构就会硬化到磁盘和旧版本的分区已被删除。

To answer your question you would only need 6Gb (or 6.5Gb to allow for the working space). SSAS only creates shadow copies of objects that are modified as part of the current transaction. So if you process only a single 1Gb partition and then commit that transaction you'll only need a shadow copy of that one partition. During the commit phase SSAS then goes through it's master version map and changes it's pointers so that they use the shadow copy as the new source of data, then once the version map is updated the new structure is hardened to disk and the old version of the partition is dropped.