iPlane Nano: path prediction for peer-to-peer applications
NSDI'09 Proceedings of the 6th USENIX symposium on Networked systems design and implementation
Resolving IP aliases with prespecified timestamps
IMC '10 Proceedings of the 10th ACM SIGCOMM conference on Internet measurement
DataTraffic Monitoring and Analysis
| Hi-index | 0.00 |
Over the last few years, several new applications have emerged on the Internet that are distributed at a scale previously unseen. Examples include peer-to-peer filesharing, content distribution networks, and voice-over-IP. This new class of distributed applications can make intelligent choices among the several paths available to them to optimize their performance. However, as a result of the best-effort packet forwarding interface exported by the Internet, applications need to explicitly measure the network to discover information about any path. Not only does measurement at the time of communication impose a significant overhead on most applications, but it is also redundant to have every application reimplement an Internet measurement component. In this dissertation, I design, build, and evaluate an information plane for the Internet, called iPlane, that enables distributed applications to discover information about Internet paths without explicit measurement. iPlane efficiently performs measurements from end-hosts under its control to predict path properties on the Internet between arbitrary end-hosts. I pursue a structural approach in issuing and synthesizing measurements—instead of using only end-to-end measurements and thus treating the Internet as a blackbox, I discover the Internet's routing topology and then compose measurements of links and path segments. This structural approach enables iPlane to predict multiple path properties, such as latency and loss rate. My evaluation of iPlane shows that iPlane's predictions of paths and path properties are accurate, significantly better than previous approaches for some of the sub-problems that iPlane tackles. Also, I used information from iPlane to drive path selection in three representative distributed applications—content distribution, peer-to-peer filesharing, and voice-over-IP. In each case, the use of iPlane helped significantly improve application performance.