CNS Core:Small: Optimizing IP Anycast Performance at Scale

无论是网上银行、购物、电话会议还是游戏，用户都会注意到应用程序在互联网上是否有明显的延迟。网络研究社区和服务提供者已经花费了大量的时间和精力来最小化这种交互延迟，并且在减少它们方面已经取得了很大的进展。其中一项关键技术是IP任播。IP任播允许互联网用户访问存储在网络距离较近的服务器上的复制web内容。理论上，在不同地理位置部署许多副本服务器的服务提供商希望实现较低的用户延迟。然而，在实践中，情况往往并非如此：更多的副本服务器有时会导致延迟延长。该项目旨在为研究和网络运营社区提供知识和工具，以优化大规模IP任播系统的设计，使这些系统能够以最少数量的副本服务器实现低互联网延迟。具体来说，该项目将开发技术和工具，以实现大规模的任播优化。它将探索使用随机算法来减少任意播集水区预测所需的BGP实验数量，并研究结合互联网拓扑信息是否能有效地进一步减少BGP实验数量。它计划研究多路径路由和非通用BGP策略如何将错误引入集水区预测，以及这些做法是否可以提高集水区预测的准确性。它还将检查启发式知识（如站点数量和站点位置）和分而治之的技术（如进一步扩展anycast的分而治之）是否对性能优化有效。最后，该项目将关注负载平衡和弹性等延迟以外的性能指标。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Whether it is for online banking, shopping, teleconferencing or gaming, users notice if there is significant delay with an application on the Internet. The networking research community and Service Providers have spent significant time and effort to minimize this interaction latency, and have come a long way in reducing them. One key technique that has been devised is IP anycast. IP anycast allows Internet users to access replicated web content stored at a server close in network distance to them. In theory, a service provider that deploys many replica servers in diverse geographic locations expect to achieve low user latencies. However, in practice, this is often not the case: more replica servers sometimes lead to prolonged latencies. This project aims to provide the research and network operations community with the knowledge and tools to optimize the design of large-scale IP anycast systems such that these systems could achieve low Internet latencies with a minimal number of replica servers. Specifically, this project will develop techniques and tools to enable anycast optimization at scale. It will explore using randomized algorithms to reduce the number of BGP experiments needed for anycast catchment prediction and investigate whether incorporating Internet topology information is effective in further reducing the number of BGP experiments. It plans to study how multi-path routing and non-common BGP policies introduce errors into catchment prediction and whether these practices might improve the accuracy of catchment prediction. It will also examine whether heuristic knowledge such as the number of sites and the site locations and techniques such as divide-and-conquer to further scale anycast can be effective for performance optimization. Finally, the project will look at performance metrics other than latency such as load balancing and resilience.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.