PredictRoute: A Network Path Prediction Toolkit
PredictRoute: A Network Path Prediction Toolkit
复制标题
DOI：
10.1145/3543516.3460107
复制
发表时间：
2021-05
期刊：
影响因子：
--
通讯作者：
Rachee Singh;D. Tench;Phillipa Gill;A. Mcgregor
中科院分区：
文献类型：
--
作者：
Rachee Singh;D. Tench;Phillipa Gill;A. Mcgregor
准确预测互联网上任意主机之间的网络路径对于网络运营商、云提供商和学术研究人员来说至关重要。我们提出了PredictRoute，一个利用数据和控制平面的历史知识来预测Internet上主机之间的网络路径的系统。除了提供可用的traceroute和BGP路由表之外，PredictRoute还可以根据所选的BGP前缀对网络路径进行优化探索。PredictRoute用于探索网络路径的策略发现的自治系统（AS）跳数比目前实践中使用的其他知名策略多4倍。使用traceroutes语料库，PredictRoute训练针对Internet上前缀的路由的概率模型，以预测网络路径及其可能性。在75%的情况下，PredictRoute的as路径预测与实际路径最多相差1跳。我们通过REST API公开了PredictRoute的路径预测功能，以便将其包含在其他应用程序和研究中。我们还演示了PredictRoute在改进现实世界应用程序以规避互联网审查和保持在线匿名性方面的效用。
Accurate prediction of network paths between arbitrary hosts on the Internet is of vital importance for network operators, cloud providers, and academic researchers. We present PredictRoute, a system that predicts network paths between hosts on the Internet using historical knowledge of the data and control plane. In addition to feeding on freely available traceroutes and BGP routing tables, PredictRoute optimally explores network paths towards chosen BGP prefixes. PredictRoute's strategy for exploring network paths discovers 4X more autonomous system (AS) hops than other well-known strategies used in practice today. Using a corpus of traceroutes, PredictRoute trains probabilistic models of routing towards prefixes on the Internet to predict network paths and their likelihood. PredictRoute's AS-path predictions differ from the measured path by at most 1 hop, 75% of the time. We expose PredictRoute's path prediction capability via a REST API to facilitate its inclusion in other applications and studies. We additionally demonstrate the utility of PredictRoute in improving real-world applications for circumventing Internet censorship and preserving anonymity online.