No-Jump-into-Latency in China's Internet No-Jump-into-Latency in China's Internet! Toward Last-Mile Hop Count Based IP Geo-localization Chong Xiang1, Xinyu Wang1, Qingrong Chen2, Minhui Xue3, Zhaoyu Gao4, Haojin Zhu1, Cailian Chen1, Qiuhua Fan5 1 Shanghai Jiao Tong University 2 University of Illinois Urbana-Champaign 3 The University of Adelaide 4 University of Massachusetts Amherst 5 RTBAisa
Discussion & Conclusion 目录 Contents 1 Motivation 2 Algorithm 3 Evaluation 4 Discussion & Conclusion
Discussion & Conclusion 目录 Contents 1 Motivation 2 Algorithm 3 Evaluation 4 Discussion & Conclusion
IP Geo-Localization Targeted Advertising Digital Right Management
IP Geo-Localization Existing methods: map RTT to Physical distance[1] [1] Weinberg, Zachary, et al. "How to Catch when Proxies Lie: Verifying the Physical Locations of Network Proxies with Active Geolocation." Proceedings of the Internet Measurement Conference 2018. ACM, 2018.
Is RTT-based method good? Physical distance versus RTT latency across 10 province in China Physical distance versus RTT latency in Shanghai
Is RTT-based method good? Weak correlation and unstableness The path between network end points can often be circuitous RTT is also likely inflated by router queuing and processing delays Expensive cost Multiple ping servers and ping requests are required To obtain the RTT with least inflation To fit a complex mapping function
Hop Count Based Approach End points bounded to the same last-hop router The last few hops of the traceroute toward a target IP should be stable and reliable Is the physical distance between these two end points well bounded? If so, we can map hop count to the upper bound of physical distance!
Hop Count Based Approach The physical distance is well bounded! Reasoning: the last router only serves a limited physical area (e.g., a building or a block) CDF analysis of the physical distances between the landmarks bounded to the same last hop router
Advantages of Hop Count Better stableness and correlation Last few hops is unlikely to change significantly Last few hops are less circuitous; thus, simple mapping (e.g., linear function) is possible Cost efficiency One active vantage points with a set of passive landmarks Only requires one traceroute request
Discussion & Conclusion 目录 Contents 1 Motivation 2 Algorithm 3 Evaluation 4 Discussion & Conclusion
1. Derive the Network Topology Traceroute to derive the network topology One traceroute request is enough
2. Fit the Mapping Function Identity the last common router (instead of the last-hop router) Calculate the hop count distance Estimate coverage radius for provincial routers Divide the physical distance by hop count distance and take the arithmetic mean Only take nearby landmarks into account
3. Geo-Localize the Target IP Calculate the physical distance with hop count distance and estimated provincial router radii Localize the target IP to the position of its nearest landmark
Discussion & Conclusion 目录 Contents 1 Motivation 2 Algorithm 3 Evaluation 4 Discussion & Conclusion
Experiment Setup Dataset from RTBAsia: 244,344 IPs with corresponding GPS coordinates Choose 10 provinces for evaluation Compare with Base-IP[2] 5 ping requests to obtain minimum RTT 1 traceroute to derive network path Plot of IP locations across China [2] Yong Wang, Daniel Burgener, Marcel Flores, Aleksandar Kuzmanovic, and Cheng Huang. 2011. Towards Street-Level Client-Independent IP Geolocation. In NSDI, Vol. 11. 27–27.
Evaluation Result Our hop count based method significantly outperforms RTT based IP geo-localization! CDF analysis of the absolute distance error in ten provinces from our method and Base-IP
Evaluation Result Performance is insensitive to the location of the ping server! Further support our assumption. Since we only use the hop lying within a province. Should be insensitive CDF analysis of the influence of the vantage point lo cation
Evaluation Result Can benefit from multiple ping server more significantly than RTT based approach! CDF analysis of the incorporation of multiple ping servers
Discussion & Conclusion 目录 Contents 1 Motivation 2 Algorithm 3 Evaluation 4 Discussion & Conclusion
Band Remote Access Server BRAS is the last visible router to external observer IPs in a restricted area have a great variety of RTTs IPs under the same BRAS (50 km2 in Beijing) RTT distribution
Conclusion We shows RTT a not-so-good measure for the physical distance and give possible reasonings We proposes a hop count based IP geo-localization approach which fits a reliable mapping from the hop count to the physical distance We extensively evaluate the proposed method and show one traceroute request from one ping server is enough to outperform existing expensive RTT based methods We hope the insights of this work can inspire a new way of IP geo-localization
Thanks! Chong Xiang, NSEC Lab, Shanghai Jiao Tong University xiangchong97@gmail.com