Inferring link weights using endtoend measurements.pdf英文资料.pdfVIP

Inferring Link Weights using End-to-End Measurements Ratul Mahajan Neil Spring David Wetherall Tom Anderson University of Washington Abstract—We describe a novel constraint-based approach straint system deal with noise in the data, such as longer to approximate ISP link weights using only end-to-end mea- paths being taken due to transient failures. surements. Common routing protocols such as OSPF and We applied our approach to six diverse ISP maps col- IS-IS choose least-cost paths using link weights, so inferred lected by Rocketfuel [19], including Ebone (Europe), weights provide a simple, concise, and useful model of intra- domain routing. Our approach extends router-level ISP Sprint (USA) and Telstra (Australia), using the traceroute maps, which include only connectivity, with link weights data collected as part of topology measurement itself. We that are consistent with routing. Our inferred weights agree find that the inferred weights are an excellent model for well with observed routing: while our inferred weights fully observed routing: across the six ISPs, the sets of paths characterize the set of shortest paths between 84-99% of the between 84-99% of the router-pairs were fully character- router-pairs, alternative models based on hop count and la- ized by our weights. We also find that simpler alternatives tency do so for only 47-81% of the pairs. such as minimum hop count or latency are poor models of intra-domain routing: they described paths between only I. INTRODUCTION 47-81% of the router-pairs.