Abstract: In the aim of optimizing the I/O performance in the cloud storage system, the partitioning decision algorithm and the computational method of the best partitioning number based on the partitioning storage model of I/O performance were put forward. An algorithm called DADH was proposed to make partitioning decision and compute the best number of partitioning dynamically by taking each storage node's and client's actual performance into account, and to optimize data distribution according to the storage nodes' actual performance. Results show that DADH has the higher I/O performance than traditional data distribution algorithm with the advance from 18% to 44%.