Ph.D. Research Proposal Exam: Zhihao Li
Monday, May 14, 2018
301 405 3681
ANNOUNCEMENT: Ph.D. Research Proposal Exam
Name: Zhihao Li
Professor Neil Spring (Chair)
Professor Richard La
Professor Tudor Dumitras
Date/Time:Monday, May 14th, 2018 at 2:00 PM
Location: AVW 3258
Title: Diagnosing and improving the performance of Internet anycast
IP anycast is widely used in critical Internet infrastructures, including many of root and top- level DNS servers, major open DNS resolvers, and content delivery networks (CDNs). What makes IP anycast such an attractive option for these globally replicated services is the desired properties that anycast would appear to achieve: reduced overall access latency for clients; im- proved scalability by distributing traffic across servers; and enhanced resilience to DDoS attacks. These desired properties, however, are not guaranteed. In anycast, a packet is directed to certain anycast replica through inter-domain routing, which lacks information to pick a route with bet- ter performance in terms of latency or load balance. Existing work painted a mixed picture of anycast performance: many clients of anycast are not served by their nearby anycast servers and experience large latency overheads; anycast sometimes does not balance load across sites effec- tively; the catchment of an anycast site is mostly stable, but sensitive to routing changes which happen quite frequently.
In this proposal, I use measurements collected from many real anycast deployments to quan- titatively demonstrate the inefficiency in performance of anycast. The results report that anycast performs poorly for most deployments I measured. Furthermore, I propose a longitudinal analy- sis that studies how routing changes cause large latency variation in anycast and characterize the causes of the routing changes.
With the goal of identifying the root causes for the performance deficit in anycast, I develop novel measurement techniques to compare AS-level routes from client to multiple anycast repli- cas. These techniques allow me to reaffirm that the major cause of the inefficiency in anycast is the performance-unawareness of inter-domain routing. With measurements from two any- cast deployments, I illustrate how much latency inflation among clients can be attributed to the policy-based performance-unaware decisions made by inter-domain routing. In addition, I pro- pose methods to quantify the effects of inter-domain routing on out-of-balance load distribution among anycast replicas.
In the last part of my work, I describe an incrementally deployable fix to the inefficiency of IP anycast. Prior work has proposed a particular deployment scheme for anycast to improve its performance: anycast servers should be deployed such that they all share the same upstream provider. However, this solution would require re-negotiating services that are not working un- der such a deployment. Moreover, to put the entire anycast service behind a single upstream provider introduces a single point of failure. I show that a static hint with embedded geographic information in BGP announcements helps to fix the inefficiency in anycast. Evaluation of such static hint in BGP route selection mechanisms is conducted through simulation with real network traces. The simulation results show that the fix is promising: in the anycast deployments I evalu- ated, the fix reduces latency inflation for almost all clients, and reduces latency by 50ms for 23% to 33% of the clients. In my upcoming work, I propose a method to evaluate the effectiveness of the static hints in BGP announcements with real-world anycast deployments.
Conducting the proposed analysis will help achieve broad and longitudinal evaluation of anycast performance for different services; identify the particularly amplified weakness of BGP routing for anycast routing; and provide suggestions to the community on improving anycast performance, which in turn help improve performance and reliability for many critical Internet infrastructures.