Cocast:一种基于传感器的分布式网络距离预测任意播算法

Cocast: A Coordinate-fusion Based Distributed Network Distance Prediction Anycast Algorithm

针对Internet环境中基于网络坐标估计节点间网络距离的问题,提出了一个利用任意播机制分布式的网络坐标预测算法Cocast:所有用户节点均加入一个结构化Peer-to-Peer对等网;对等网中任意的节点均可以发送网络坐标任意播查询消息,该消息被转发到多个负责坐标计算的用户节点,这些用户节点分布式地预测消息发起节点坐标,最终消息发起节点获得一个综合的坐标位置。Cocast基于网络坐标任意播选择提供位置估计的用户节点,利用层次化网络嵌入预测坐标位置,借助位置融合传感器机制过滤恶意节点的影响。实验表明Cocast具有可扩展、快速收敛、渐增精度、抗恶意节点破坏等优点。

According to the network-coordinate based network distance estimation problem in an Intemet-wide environment, an anycast algorithm named Cocast for cooperative network coordinate estimation is presented. All peers join a structured Peer-to-Peer overlay, and any peer in the overlay, i.e., a requestor, may send a network coordinated anycast message, then the anycast message is forwarded to a set of peers in charge of network coordinate estimation, and each of these peers makes responses from a network coordinate to the requestor, finally the requestor gets a synthetical network coordinate. Based on three novel mechanisms, i.e., an anycast service to select peers for network coordinate estimation, a hierarchical network embedding mechanism to compute the network coordinates, and a coordinate-fusion sensor approach to filter malicious peers＇ fake-coordinate effects, Cocast owns a fast convergence speed in network coordinate estimation and is more stable. Simulation results confirm that Cocast is sealable, fast-convergent, incrementally accurate and resilient.