An adaptive statistical sampling technique for computer network traffic

The rapid growth of real-time applications transmitted over multimedia networks, makes measurement of their generated traffic increasingly important. These measurements allow the quality of service (QoS) provided by the network for the transmission of the applications to be assessed. However, most real-time applications generate an extensive amount of traffic data. Analysing these data in real-time is computationally intensive. Therefore, in order to reduce the amount of processed data, sampling needs to be performed. In fixed rate sampling, the sample rate is unaffected by the packet transmission rate. However, it is advantageous to adapt the sample rate in relation to packet transmission rate. In this study a novel statistical adaptive sampling method has been developed. The method adaptively adjusts the time interval between two consecutive sampled sections (called pre-and post sampling sections). This time interval is decreased when the two sections significantly differ statistically and it is increased when their net statistic is within a predefined threshold. The operation of the developed sampling method was evaluated using a simulated computer network. The results demonstrated the effectiveness of the method in various scenarios, however more work is in progress to make the method more robust.