Abstract: Parallel data streaming algorithm was proposed according to the weak real-time performance, low detection precision and estimation accuracy for detection of long duration flow. The different threads access the shared data structure in the parallel algorithm of long duration flow detection based on shared data structure, but it generates excessive synchronous overhead. On the basis of the analytical result on the parallel algorithm of long duration flow detection with shared data structure, the different threads own local data structure in the parallel algorithm of long duration flow detection based on independent data structure, where it doesn’t need synchronization and generates minor overhead. Theoretical analysis and experimental results show that the parallel algorithm of long duration flow detection based on independent data structure has good time efficiency, high detection precision and estimation accuracy of long duration flow.

Key words: traffic measurement; flow duration; data streaming; parallel algorithm