As an important infrastructure of big data, HBase has been wildly used in lots of businesses and provides high performance in massive data reading and writing. However, HBase is still undergoing continuous improvement for meeting the special demands in lots of individual application fields. The issues involve cache replacement algorithm, region split and compact strategy, load balance and so on, and a relatively uniform scheme is still needed. In this paper, we propose a RowKey generation method based on consistent hashing and introduce a prepartition scheme to distribute data to RegionServers uniformly. We also give the consideration to reading performance while design the improved writing mechanism for several data types. Experiments show that the performance of reading and writing has been significantly improved comparing with original HBase, which benefits by mapping data to the virtual nodes in the designed hashing ring.