4 ms·

Anycast and servers all over the globe don't help with latency. You also need to keep refreshing DNS records in background and have massive cache. But the memor

Anycast and servers all over the globe don't help with latency. You also need to keep refreshing DNS records in background and have massive cache. But the memory is limited and you can't refresh everything for everyone. So only a percentage of queries can get responses from cache. If your own resolver could refresh records in background, you would get more responses from cache than sharing it with everyone. You also won't be affected by things like server load and connectivity issues, so you'll get overall better latency with queries mostly served from own cache matching your personal profile.

belorn 8y ago
If you use your ISP's recursive resolver it will likely be a virtual machine on their cloud infrastructure, the very same system that they provide as a content delivery system for the anycast service. The latency for a DNS query will therefore often be the exact same latency for the caching resolver as if you asked the anycast DNS server directly, because it's part of the exact same system, possibly running on the exact same machine. There is no need to keep refreshing DNS records in background and have a massive cache. The content delivery system does that all day long. Because this is a service that ISPs want to sell, there is also a huge incentive for them to get large companies to pay for the CDN rather than expect them to run background jobs with large memory costs for free. But even if there were added latencies for some subset of sites, we are talking about milliseconds. With DoH there has been a simultaneous claim that with HTTPS they can push multiple responses for a single request (something which the DNS protocol supports but has never been used in any implementation), which means that now we can reduce the latency to a single round trip. The difference between a single round trip to Cloudflare's CDN recursive resolver, and a single round trip to the official anycast DNS resolver with DoH, is marginal if any. If the site is located on Cloudflare the whole question is also moot.

belorn 8y ago

If you use your ISP's recursive resolver it will likely be a virtual machine on their cloud infrastructure, the very same system that they provide as a content delivery system for the anycast service. The latency for a DNS query will therefore often be the exact same latency for the caching resolver as if you asked the anycast DNS server directly, because it's part of the exact same system, possibly running on the exact same machine. There is no need to keep refreshing DNS records in background and have a massive cache. The content delivery system does that all day long. Because this is a service that ISPs want to sell, there is also a huge incentive for them to get large companies to pay for the CDN rather than expect them to run background jobs with large memory costs for free. But even if there were added latencies for some subset of sites, we are talking about milliseconds. With DoH there has been a simultaneous claim that with HTTPS they can push multiple responses for a single request (something which the DNS protocol supports but has never been used in any implementation), which means that now we can reduce the latency to a single round trip. The difference between a single round trip to Cloudflare's CDN recursive resolver, and a single round trip to the official anycast DNS resolver with DoH, is marginal if any. If the site is located on Cloudflare the whole question is also moot.