Review of "Quantifying the Causes of Path Inflation"

From: Michelle Liu (liujing@u.washington.edu)
Date: Tue Nov 02 2004 - 22:46:05 PST

  • Next message: Shobhit Raj Mathur: "review of paper 17"

    Review of "Quantifying the Causes of Path Inflation"

    Jing Liu

     

        This paper analyzes how topology and policy combine to contribute to path inflation at three levels in the Internet: intra-domain, peering, and inter-domain.

        Through a trace-driven analysis, the effects of the six causes of path inflation: topology and policy at all three levels are isolated. There are a set of conclusions drawn. First, intra-domain traffic engineering has only a minimal impact on path inflation. Load-balancing routing policy may lead to longer paths. Considering topology, if the tier-1 ISPs having more POPs over a larger geographic area, such as crossing different continents, it is more difficult for them to connect all pairs with a good set of links. Thus, path inflation might happen. Second, peering topology does not inflate paths significantly. The inflation will be less if crossing boundaries between two ISPs are in the same continents than in different continents. Peering policies, such as early-exit and late-exit and load-balancing often cause path inflation and these policies sometimes lead to highly inflated paths. Moreover, the inflation decreases quickly as the number of peering points increases. Third, inter-domain routing has a significant impact on path inflation. However, it is not because of the policies, such as no-valley and prefer-customer arising out of commercial concerns, but using shortest AS-path as the distance metric.

        The paper is well organized and the methodology is stated very much in detail. For instance, the paper talks a lot about how to choose ISPs, data collection, extracting topology and why to do it in this way. Furthermore, the research before has been limited to the inter-domain case. This paper adds two key missing pieces to routing policy inference work: inference of intra-domain traffic engineering and inference of peering policy between a pair of ISPs. Finally, this paper also finds that it is not as we used to think that policies arising out of commercial concerns mainly contribute to the path inflation, but because shortest AS path is the default.

        This paper provides a quite thorough study on path inflation of the Internet. Since some reasons of path inflation have been found, it will be naturally to consider how to reduce path inflation and which mechanisms could be used in the future work.

        


  • Next message: Shobhit Raj Mathur: "review of paper 17"

    This archive was generated by hypermail 2.1.6 : Tue Nov 02 2004 - 22:46:22 PST