Adaptive Sampling Based Network Traffice Measurement System For High-speed Mulit-link Logical Channel Environment

With the high-speed developing of computer network technology, the research on network traffic measurement, especially the passive measurement meets the new challenges laid by the increasing transport capability and complex network interconnection in modern network backbone's practical deploying. While many measurement systems developed for early low speed network environment are not fit for today's measurement need to detect more intricate traffic behavior under more complicated network link interconnection and huge transport traffic volume. So how to do traffic measurement efficiently under the large-scale high speed backbones has become a hot research topic for all the researchers in this field.Taking sampling based passive measurement for high-speed network as the research background, this paper carry out research on the efficient way to do adaptive sampling based measuring under high speed multi-link logical channel environment. On the analyzing of this kind of network environment's characters and the drawback of traditional measurement systems'design, this paper proposes a novel design of adaptive sampling based passive measurement system for the multi-link logical channel.Based on the analysis of multi-link logical link's characters, this paper firstly introduces a data flow model of passive measurement fit for multi-link logical channel's measurement. And further, this paper gives out the measurement system's framework design, the detailed definitions of all functions of the framework and the description of the organization of them.Then this paper represents a design of network flow measurement of link traffic, which combines adaptive packet sampling and adaptive flow sampling to control the consumption of CPU and memory resources individually. In the designing of adaptive packet sampling, this paper invents a dynamic packet counting based algorithm to computing packet transport speed and a threshold-detecting based algorithm to control the sample rate adapting. And then this paper gives out an improved packet sample-hold based algorithm to control flow sampling by packet random sampling and a statistic model of the sampling control. In the discussion of flow properties'estimation and error analysis, this paper introduces an algorithm to estimate the original packet sum and bytes sum in a flow record and calculate the estimate error.Finally, this paper describes a network flow measurement system implement based on the architecture proposed in this paper. Systematical tests had been conducted on this prototype to evaluate the performance and the efficiency of the adapting algorithms. As shown by the test results, the design of the network flow measurement system for the high-speed multi-link logical channel has achieved the expected targets on its function and performance.