Research Of Flash Dissemination Oriented Latency And Bandwidth Sensitive Technology

Flash dissemination, which requires to disseminate the data in a high-throughput, low latency manner, is a key enabling technique in many emerging fields, such as emergency response, information battlefields environment, multimedia applications, news dissemination, financial securities, etc. Due to the success of the recent overlay techniques, flash dissemination can be practical, by taking advantage of flexible deployment, self-organization, high scalability and robustness of overlays. However, the lack of physical network topology information in overlay construction can easily result in low efficiency and bad transmission in the process of data dissemination. Latency and bandwidth, as two important parameters of network, greatly affect the efficiency of data dissemination. To support efficient flash dissemination on the overlays, three complementary problems are studied: latency predicton, bandwidth measurements, and the latency and bandwidth sensitive technologies for optimal routing and path selection. Firstly, considering the requirements of convergency, accuracy, security etc. in the latency prediction process, a heuristic network coordinate computing method based on clustering CHCC is proposed to optimize the latency prediction. In CHCC, to optimize the scalability and accuracy, a modified Binning clustering algorithm EBinning combined with the optimal landmark selection strategy Hybrid is used. To improve the security of coordinates, a malicious nodes filtering algorithm MNDA is raised. To enhance the convergency and accuracy, a heuristic coordinate updating algorithm HCUA is proposed to update the coordinates periodically. Finally, a coordinate stabling algorithm CSA is raised to improve the stability of the coordinates. Simulation results show that the convergency, stability, security and accuracy in CHCC are better than classical method Vivaldi.Secondly, to solve the problem of the end-to-end path available bandwidth measurement in Internet environment, a fast and self-adaptive method FSabe is proposed. The PRM model is adopted in FSabe, so the restrictions that the bandwidth capacity needs to be known and the narrow link should be the tight link are removed. In FSabe, the send rate is adaptively adjusted depending on the arrival rate at the target and the burst of the cross traffic, which improves the convergence of the measurement. The length of the probe train is adaptively adjusted according to the bursty of the cross traffic and the path load, which improves the practicability. The appropriate size of the probe packet and the improved method that judging the change direction of OWD are also studied, which enhances the accuracy and the interference immunity. Simulation results indicate that the accuracy, time cost and interference to the path of FSabe is better than the PRM-based method Pathload.Thirdly, to improve the low efficiency in routing and to remove the performance bottleneck of the path in flash dissemination, a location-aware unstructured overlay topology construction algorithm Lautc as well as a latency and bandwidth sensitive path selection algorithm LBPS is proposed. In Lautc, unstructured topology and distributed construction manner are adopted, which greatly improves the dynamic adaptability, resilience, and scability. Random walking manner combining with low-latency neighbor selection strategy are employed, which reduces latency stretch and greatly improves the efficiency of data location. LBPS is built on the basis of location-aware unstructured overlay topology, which improves the efficient of data location. In LBPS, a semi-distributed management pattern of bandwidth information is adopted to improve scalability and practicality. To improve the efficiency of the dissemination, paths are selected in terms of latency, bandwidth capacity and available capacity, but at the same the delay is reduced as much as possible. Simulation results show that the latency stretch of the topology constructed by Lautc is lower than the Binning method and its efficiency of data location is much higher than Gnutella. Moreover, the results of the available bandwidth of the default paths and the optimized paths show that the dissemination paths selected by LBPS are greatly optimized and the efficiency of flash dissemination is remarkably improved.