关键词: 数据分区, 可扩展性, 元组聚类

Abstract: Data partitioning is an effective method to advance scalability of databases. In transaction-intensive systems, reasonable methods of data partition can reduce the number of distributed transactions and the response time of transactions. This paper proposes a novel approach based on tuple clustering to partition data incrementally. Al-gorithm complexity is reduced by the method of tuple cluster and the method of data filter based on partition-aware. Firstly, the related tuples are clustered by using time window model, and then the clustered node graph is con-structed. Secondly, some nodes in the graph are deleted according to partition-aware strategy. Lastly, partitions are obtained by dividing the graph into subgraphs. Compared with existing methods, the new method spends less time on data partitioning, and obtains less distributed transactions and faster speed on query for partitioning databases.

Key words: data partitioning, scalability, tuple clustering