Hacker News Reader

Comment by brandon272 | Hacker News Reader

Round trip time for me to 80.80.80.80 is double what it is to 8.8.8.8. So I'm not convinced on the "fast" claim.

Google has the advantage to anycast their DNS server IP addresses. Use a ping service that will ping an IP from around the globe and you'll see that 8.8.8.8 has sub 20ms RTT time for pings reguardless of country. So you will likely always hit the closest Google datacenter when using 8.8.8.8.

80.80.80.80 seems to be served out of a single datacenter. Based on ping times I'm seeing, it looks like they are hosted in Amsterdam. So it will be slow for most of Asia and the Americas.

For this same reason you'll also get better results from CDN services with Google DNS. 8.8.8.8 will lookup a geo record from (almost) your actual location. 80.80.80.80 will look it up from the USA.

Any CDNs that are using geo-dns will be way slower than they should be outside the US as you'll have been given a US IP to load the content from.

Modern CDNs don't use DNS based routing. If a CDN is doing that then they're not really focused on being fast or reliable.

Google actually implements a workaround to this issue.its called EDNS client subnet. Freenom does not.

Ah nice, I'll have to remember that one.

That's not some advantage randomly conferred onto Google, though. That's how you build a fast, global DNS service. Freenom is as fast as their servers allow in north western Europe (assuming they are indeed hosted in Amsterdam), but it's just not fast in the rest of the world.

I've always wondered though if you could make a better, more anonymous naming service entirely. I believe GNU has one, I think IPFS has another.

These are, after all, just giving a more human understandable way of proving addressing information. On Unix and Unix-like systems there is a resolver, and I would be fascinated to know how hard it would be to literally swap out or choose an entirely new name-to-IP address resolution mechanism at runtime.

The website says they are using anycast as well.

It looks like they are technically anycasting, but doing it totally wrong.

From multiple test locations I am seeing lots of network tromboning. Queries from the US going to Europe and vice versa.

I'm getting <1ms from a server in Amsterdam, some hundreds (3-5) of ms from the UK, with some even timing out.

RTT is >5x higher for me. (~140ms vs ~25ms) I'm in Minneapolis, and the traceroute to 80.80.80.80 ends in Prague.

That's surprising: With that IP address, you'd expect it to be ten times as slow.

Same here. 118ms vs 29ms (average round trip time).

same for me: dig "Query time" almost double for 80.80.80.80 (vs. 8.8.8.8)

From Denmark, 80.80.80.80 is actually slightly faster than 8.8.8.8, but the difference is tiny.

80.80.80.80 is always 12ms. Appears to be in Amsterdam.

8.8.8.8 is between 13-60ms, usually 12-25ms. Somewhere in Sweden.

Reply on news.ycombinator.com