> However, the provided command actually runs 300 queries in a really short timeframe (specified by -c 300)
Our servers see 2500-25000 requests per second. So, my intention was to test the server, not the client (original post points out issues with the server).
> which is very different from real-world use cases
Not really. A decent DoH/DoT/ODoH client would pool TLS/HTTP connections (on top of mutliplexing HTTP/2 on TCP) & have its own caches (probably hydrate the top domains in the background, even).
> As sky.rethinkdns.com is hosted on Cloudflare, there shouldn't be any notable network differences between Rethink DNS and Cloudflare DNS.
Cloudflare likely co-locates its DNS resolvers with all major ISPs, which might explain the speed. For instance, ping 1.1.1.1 reports ridiculous numbers (<3ms) on some networks.
> With Cloudflare DNS, it takes only 35 ms, even when using DNS over HTTPS
In my experiments, for (non-random) repeat queries, the fastest Cloudflare response is about 20% faster with "-c 1" than Max and 30% faster than Sky.
You should try if any other non-BigTech public DNS posts comparable numbers for "-c 1": Like ControlD, AdGuard, NextDNS, CleanBrowsing, DNSFilter, Quad9 etc. In my experiments, these aren't significantly better than Max or Sky. And imo, concurrent "random" (cache-busting) queries are better test of a remote resolver's capabilities.
For Max specifically, caches speed up response times significantly, even if Max is deployed to only 30+ locations (as compared to Cloudflare's 300+) running 40+ low-powered servers on Fly.io's network. For Sky, a lot is under the control of Cloudflare: How they route & shape our traffic and just where they run our Workers. For the most part though, Sky is comparatively pretty fast. If you use Rethink (enable Configure -> DNS -> DNS Booster to turn on optimizations), you should be able to see per-query round-trip time in Configure -> Logs -> DNS.
Our servers see 2500-25000 requests per second. So, my intention was to test the server, not the client (original post points out issues with the server).
> which is very different from real-world use cases
Not really. A decent DoH/DoT/ODoH client would pool TLS/HTTP connections (on top of mutliplexing HTTP/2 on TCP) & have its own caches (probably hydrate the top domains in the background, even).
> As sky.rethinkdns.com is hosted on Cloudflare, there shouldn't be any notable network differences between Rethink DNS and Cloudflare DNS.
Cloudflare likely co-locates its DNS resolvers with all major ISPs, which might explain the speed. For instance, ping 1.1.1.1 reports ridiculous numbers (<3ms) on some networks.
> With Cloudflare DNS, it takes only 35 ms, even when using DNS over HTTPS
In my experiments, for (non-random) repeat queries, the fastest Cloudflare response is about 20% faster with "-c 1" than Max and 30% faster than Sky.
You should try if any other non-BigTech public DNS posts comparable numbers for "-c 1": Like ControlD, AdGuard, NextDNS, CleanBrowsing, DNSFilter, Quad9 etc. In my experiments, these aren't significantly better than Max or Sky. And imo, concurrent "random" (cache-busting) queries are better test of a remote resolver's capabilities.
For Max specifically, caches speed up response times significantly, even if Max is deployed to only 30+ locations (as compared to Cloudflare's 300+) running 40+ low-powered servers on Fly.io's network. For Sky, a lot is under the control of Cloudflare: How they route & shape our traffic and just where they run our Workers. For the most part though, Sky is comparatively pretty fast. If you use Rethink (enable Configure -> DNS -> DNS Booster to turn on optimizations), you should be able to see per-query round-trip time in Configure -> Logs -> DNS.