Research On Traffic Measurement In High-speed Ip Network

Traffic measurement serves as the basic for deeply understanding the essence of network and effectively comprehending the operation of network. It is also an important component of network applications including QoS management, network optimization and traffic engineering. With the rapid development of information technology, Internet has undergone great changes in the overall scale and architecture. Confronted with many challenges and various contradictions, traditional mechanisms of traffic measurement have become bottlenecks for further network developments.Combined with the fundamental technique research task of the New Generation Network with High Trustability project of the National High-Tech Research and Development Program of China (863 Program), this dissertation classifies and compares current researches on traffic measurement, and summarizes results achieved in this area. From the perspective of the extensibility supporting scheme, it comes up with researches into studying the design and implementation of front end processing algorithms for traffic management system on high-speed backbone networks. Its main work and contributions are outlined as follows:1. Aiming at the deficiency of traditional traffic measurement model, a novel scalable flow measurement model which can be deployed in high-speed network is presented. Introducing the idea of processing in batches, the model designs two-stage buffer structure to conduct buffering packets and measuring flows simultaneously. Flow sampling technique and synopsis data structure are considered as two key techniques based on the above model. Implementing the model to measure traffic can not only meet the demand of resources, but also improve the analytical speed.2. Based on the analysis of shortcomings possessed by the traditional static sampling mechanism, the dissertation proposes a novel sampling sketch called time stratified adaptive packet sampling based on traffic load. The sketch adopts the following methods: bounding sampling error within a pre-specified tolerance level, time stratified packet sampling and predicting traffic load adaptively. This easily-implemented packet sampling method gives the right tradeoff between resource consumption and accuracy for all traffic mixes. Experimental results demonstrate that the proposed method can achieve simplicity, adaptability and controllability of resource consumption without sacrificing accuracy.3. In order to circumvent the deficiency of the limitted computing and storage abilities in traditional traffic measurement, a innovative mechanism of synopsis data structure called identifying heavy hitters based on Multi-dimensional Counting Bloom Filter is proposed. Extending the standard structure of Counting Bloom Filter to multi-dimensional one, the mechanism can not only represent, query and count flows, but also sustain real-time multi-granularity measurement. Implementing renormalization and configuration of MDCBF, it can realize identification of heavy hitters and restriction of measurement errors. Experiments are conducted based on the data both randomly produced by computer and sampled from the real network trace. Results show that the proposed mechanism exhibits low computing complexity and finer space saving.4. Taking veracity and real time requirements into account, the dissertation divides the real-time traffic management system into front end and back end. This system architecture eliminates the coupling between front end and back end, thus a more efficient and flexible system implementation can be achieved. Based on such partitioned structure, the implemented technique of the front end algorithm using FPGA is discussed in detail. Simulation results validate this scheme.