Evaluating Public DNS Services in the Wake of Increasing Centralization of DNS

Recent studies have shown centralization in the Domain Name System (DNS) around public DNS services, which are hosted on centrally managed infrastructure and advertise higher reliability, improved security, and faster response times for name resolutions. However, many of the recently emerged public DNS services have not yet been extensively studied regarding popularity and performance. In light of this, we use 10.6k RIPE Atlas probes and find that 28.3% of the probes (and the their host network by extension) use at least one public DNS service, with Google being the most popular public DNS service among these probes. We further quantify the response time benefits of such public DNS services using ≈2.5k RIPE Atlas probes deployed in home networks (1k of which are IPv6 capable): Overall, we provision around 12.7M DNS requests based on a set of 23 domains and ten centralized public DNS services both over IPv4 and IPv6. For comparison, we additionally resolve the same set of domains using the probes' local resolvers, which are typically managed by the ISP and exhibit lower response times in general. We observe that even though IP and AS paths to local resolvers are generally shorter, some public DNS services (e.g., Cloudflare), achieve faster responses over both IPv4 and IPv6. Across all continents, Cloudflare, Google, and OpenDNS exhibit the lowest response times out of all public resolvers for successful DNS measurements. Probes in Europe (EU) and North America (NA) experience comparable latencies to public and local resolvers, thereby diminishing claimed latency benefits of public resolvers. We also observe inflated path lengths to and response times (over both address families) from most public resolvers for probes in Africa (AF) and South America (SA). Based on our observations, we provide recommendations and discuss situations in which switching to public DNS services may be beneficial.