To mine data stream efficiency, a new slide window, combines basic window and time window, adopts a scheme of window based on sector. It could apply to the real data streams and change the size of the windows, and save time and space that no necessary to use. And the algorithm based on frequent itemsets mining-DS-stream could self-adaptively adjust the size of time windows, considering the large number and chronotropic character of the data stream. In the algorithm, transaction list group is adopted as synopsis data structure. The experiments results in Eclipse indicate that the DS-stream algorithm is more effective in term of temporal and spatial performance.