Characterization of DNS Servers for Latency Estimation Metrics
It is easy to “ping” a remote computer from one’s own to determine the time delay between them, but estimating the latency between two remote hosts is much less trivial. This problem has various applications: social networking and online gaming are a few significant examples. Clever algorithms have been developed to approach the problem, though recent changes in the Internet are rendering them increasingly inaccurate. One such algorithm, Turbo King, makes use of DNS servers geographically near two remote hosts to estimate the latency between them. Unfortunately, with the emergence of IPv6, certain DNS server implementations have added additional queries and steps to their name resolution process. Because these extra delays are unpredictable, they add error to Turbo King’s latency estimation, rendering the algorithm imprecise. We have reevaluated Turbo King by classifying DNS servers by behavior and analyzing delays for potential patterns. With this information, we have patched certain cases to allow Turbo King to work correctly; in other cases we have been able to exclude IPs from the set of those usable by Turbo King. Now the algorithm’s range of usefulness has been expanded, possibly with enough coverage to measure the entire web.
Johnson, Joseph (2018). Characterization of DNS Servers for Latency Estimation Metrics. Undergraduate Research Scholars Program. Available electronically from