关键词: 数据划分 大规模并行处理 一致划分 生命期 数据流

Abstract:

Data partitioning is the most important technique in MPP. And we take advantage of the methods such as renaming, array expanding, loop exchanging and loop splitting to achieve the consistent data partitioning while the information of the array llfe cycle is the basis for those methods. Using the rcgion extension method for the attribute of data flow, we have designed an automatic analysis tool for obtaining the information of the array life cycle, wh ich is the basis for achie ving the best data partitioning.

Key words: data partitioning, MPP, consistent data partitioning, life cycle, data flow