【摘要】 随着大规模的分布式网络应用对网络主机间距离信息的需求,深入研究了基于虚拟坐标的IP网络距离预测机制。在基于固定landmarks的网络坐标系统中,普通主机通过测量到所有landmarks的距离来计算自己的坐标,其中landmarks的个数将影响到系统的距离预测准确度和测量开销网络坐标系统(network coordinate system, NCS)综合了Vivaldi和GNP的优势,具有良好的网络距离预测性能,然而它没有讨论landmarks的选取问题因此,提出一种改进型网络距离预测算法—INCS,该算法首先选定网络中的一组主机作为landmarks,然后按照landmarks之间的相互距离对其进行聚类,最终从每个聚类中按照所提出的策略选择一个节点作为普通主机的实际参考节点。仿真结果表明:与NCS算法相比较,INCS算法在牺牲0～24%(根据90%相对误差)预测准确度的情况下,能够大大减小系统的测量开销;同时INCS可以使参考节点的选择灵活化,分散参考节点处的负载,从而提高系统的稳定性。更多还原

【Abstract】 With the increasing need of the network distance between hosts for many large scale applications,the network distance prediction schemes based on virtual coordinates for IP networks were studied.In the previous network coordinate systems with fixed landmarks,the ordinary host measures its distance to all the landmarks to compute its own coordinate, where the number of the landmarks will have a great impact on the network distance prediction accuracy and measurement overhead.The algorithm of network coordinate system(NCS) integrates the advantage of Vivaldi and GNP,and shows good performance of network distance prediction,which,however,does not include any discussion about landmark selection. The scheme of improved network coordinate system(INCS) based on NCS was proposed.In INCS,some hosts were firstly selected as the landmarks,and clusters were formed based on the inter-landmark distances.Then,one node was chosen from each cluster according to the proposed rules to be the reference node of the ordinary hosts.The simulation results show that compared with NCS,can greatly reduce the measurement overhead while sacrificing accuracy within a range of 0～24% according to the 90 percentile relative error.Meanwhile,INCS makes the choosing of reference nodes more flexible,and balances the load on the reference nodes,which can improve the robustness of the system.更多还原