一种基于被动路标的网络距离预测方法

A Novel Passive-Landmark Based Network Distance Prediction Method

网络拓扑信息的引导能够显著提高大规模分布式应用程序的性能,然而直接测量节点之间拓扑信息产生的开销远大于其收益.提出一种新的基于被动路标的节点间网络距离预估方法PLNDP,使用Lipschitz变换将普通节点到路标节点的网络延迟映射到度量空间Rn,再利用距离函数计算映射后的网络坐标之间的距离,从而预测节点之间的网络距离.PLNDP中路标节点不需要主动探测,可利用Internet上已部署的高性能服务器为之,极大降低部署成本.引入有效路标和修正因子,提高了预测的准确性.实验结果表明,与经典方法GNP和Vivaldi相比,PLNDP在多个性能参数方面具有明显的优势.

With the direction of network topology information,the performance of large scale distributed applications could be enhanced greatly.However,if the topology information between nodes is obtained by directly measure,the cost of the probing packets may be more than the gain from the performance improvement.This paper proposes a novel passive landmark based network distance prediction method-PLNDP.The vector of transmission delay from normal node to landmarks is embedded into the metric space Rn by the Lipschitz transformation.After getting the network coordinates,normal nodes use the distance function to compute the distance between coordinates.Then the network distances between nodes is predicted by the distance between nodes＇ coordinates.Unlike other network coordinates system,landmarks in PLNDP only need to respond to probes passively,while not measuring distances to other landmarks actively.Existing high performance public servers,such as DNS servers and Web servers,can be used as landmarks.So the cost of deployment can be reduced greatly.In order to improve the prediction accuracy,valid landmarks and correctional factor are used in the distance function.Experiment results show that,for several different accuracy metrics,PLNDP is better than classical network distance prediction methods GNP and Vivaldi,especially when some landmarks have been failed.