Research On Network-Level BitTorrent Traffic Distribution Measurement

Having enjoyed rapid development during recent years, peer-to-peer (P2P) has become the major internet traffic contributor. Being network oblivious, P2P applications generate large amounts of cross-ISP traffic, thus cause severe mismatch between P2P overlay and carrier network. However, traditional carrier networks provide"one-size-fit-all"bandwidth service, hence can hardly meet performance requirement of recently emerged application types such as P2P. Under this contradiction, it is of vital importance for Service Providers(SPs) to grasp network-level P2P traffic distribution through network measurement technique. At present, BitTorrent(BT) contributes the most part in P2P traffic. This paper focuses on measurement approach for network-level BT traffic distribution.Many previous studies have focused on link-level BT traffic identification and measurement. However, we cannot simply apply these link-level BT traffic measurement approaches to network-level BT traffic distribution measurement, the reason of which is that there's qualitative difference in problem scale between link-level measurement and network-level measurement. Network-level BT traffic distribution measurement system should bear following characteristics: scalability, reasonable cost, accuracy, real-time requirement and low impacts on network itself. In this paper, we propose a novel measurement system to estimate network-level BT traffic distribution through data fusion of sFlow and Tracker. The sFlow standard is designed specifically for scalable, network-wide traffic monitoring, and enjoys broad multi-vendor support nowadays. However, sFlow alone is longer sufficient to identify BT traffic. We tackle this problem by combining sFlow with information collected from Trackers'traffic. Tracker is the central server of BT system, thus it provides the most accurate judgement refence. Through data fusion of sFlow and Tracker, we obtain the result dataset. With different time granularity, we can achieve BT traffic trend with respect to time.To validate the efficacy and accuracy of our measurement system, we conduct measurement experiments in the environment of campus network. In the experiment, L7-filter is introduced to help establish ground truth. Evaluation results show that our proposed measurement system can reach a high accuracy in BT traffic distribution estimation.As network size increases, measurement system inevitably faces great challenge in mass measurement data analysis. This paper introduces MongoDB as optimization solution for mass measurement data analysis. MongoDB's performance is evaluated under several experiments, compared with traditional data analysis solutions.This paper is organized as follows. Chaper 1 gives brief introduction of research background and motivation. We propose BT traffic distribution measurement system based on data fusion in Chapter 2. The detailed system design is described in Chapter 3. Chapter 4 illustrates basic principle of data fusion, and presents experiments in campus network. Chapter 5 studies performance optimization solutions for mass measurement data analysis. Chapter 6 summarizes this paper and discusses areas for future consideration.