Hacker News Reader

Hacker News

Comment by freediver | Hacker News Reader

Full thread

freediver

I did test 1.1.1.1 and found it to be pretty slow on long tail domains (obviously everyone is caching popular ones).

I bascially ran a 'dig' with multiple DNS providers and CloudFlare was slowest among the bunch for long-tail domains.

Here are the details: https://twitter.com/vladquant/status/1428761979808669704

CloudFlare never responded to this tweet.

View on HN

That particular domain is sluggish from the UK, but other domains, but my route53 hosted domains - including ones never before used (wildcard subdomain) - are all fine - around 5ms.

When I tried, Quad9 (9.9.9.9) resolved enigma.rs in 5.2 s.

I'm happy with Cloudflare, even if it's slightly slower.

Interesting, but what are long tail domain?

Domains that have less frequent lookups so the chance of getting a cached response is lower.

Why aren't they just called infrequently used domains then?

>Why aren't they just called infrequently used domains then?

You could call them "infrequent" but "long-tail" is also a common description to convey a Power Law distribution: https://en.wikipedia.org/wiki/Long_tail

I think in this case about DNS caching, "long tail" is better than "infrequent". In the wikipedia graph, some of the domain lookups in yellow may be "frequent" (absolute sense) but simultaneously but much less popular (long tail) such that they don't stay in DNS lookup caches.

DNS is essentially a cache. I've never once in my life heard of infrequently accessed cache items as "long-tail". This is definitely a dumb phrase that should be avoided.

>I've never once in my life heard of infrequently accessed cache items as "long-tail".

The parent poster wrote "long tail _domains_" and not cache items: https://news.ycombinator.com/item?id=29036188

You also used the word "domains" when you asked about "infrequently used domains" and that's the context I was responding to. I didn't say that cache items are labeled "long tail".

I've never heard them referred to as anything else, so YMMV

I think the idea is that you take a list of all domains and then count the number of DNS lookups that are done for each over the course of some time period (e.g. 1 year). Then sort them from high to low number of lookups. At the start of the list you'll probably find only a few domains with many billions of lookups. As you go down the list you'll find many domains with very few lookups, the "long tail".

It's a bit confusing because normally "long tail" refers to a histogram or probability distribution where a large portion of the population is far from the central part. I don't think that works in this case, unless I'm confused about what to put on the x-axis. (Because if the x-axis is lookup frequency, the domains being referred to here would be in a peak close to 0, not in the tail.)

edit: maybe the x-axis of a histogram could be "mean time between lookups" instead of frequency? That would put the popular domains in the peak near 0, and the unpopular ones farther out in the "tail".

> what to put on the x-axis

Perhaps 1/frequency.

Reply on news.ycombinator.com