Hacker News Reader

Comment by gregmac | Hacker News Reader

Also don't use for anything real, because you should not be teaching users that https://1234-5678-9adc-def0-1234-5678-9abc-def0.has-a.name is the proper, trusted domain for your service. Because it was supposed to be https://1234-5678-9abc-def0-1234-5678-9abc-def0.has-a.name and I just tricked you.

View on HN

The fundamental problem with this whole scheme is that it is still just trying to memorize the entire IP address. The point of a name is to make it something that feels natural for actual people.

Maybe it would make more sense to choose the 65,536 most common English words (or whatever language you want) and break the address up into 8 words to form a crazy looking phrase. This is still difficult to memorize but not as bad as just stuffing 128 bits of hex down your throat. You could even allow for the collapsing 0s by making entry 0 be "and" and adding a bit of logic that does the collapse.

For fun I wrote a tiny script that does this and tried it with their example domain:

1234:5678:9abc:def0:1234:5678:9abc:def0 -> balcony gaining pawn toothill balcony gaining pawn toothill

Or Google's address from that page:

2a00:1450:4009:811::200e -> chinker bauchle dorter amor and bromidic

So maybe this wasn't the best idea, but at least they're a bit more amusing than the hex noise in the article.

Anyway, if anybody else wants to play around with it I have a tiny demo: http://jubei.ceyah.org/cgi-bin/ipv6toenglish

This scheme isn't designed to make the addresses memorable though. It's designed as a workaround for cases where hostnames are supported but v6 literals aren't, for example in UNC paths to Windows network shares.

I don't think they're proposing this for anything other than a prototyping solution along the lines of this:

http://xip.io/

There are cases where you simply want to use a name during development because it generalizes better for the eventual production case, but you don't care what the name is, you don't care if anyone can memorize it, you just need it to be a name and not an IP address.

I'm guessing you are referring to the "config needs a hostname" problem. Been there myself.

Kinda like https://what3words.com/. Question is, how long would the string of words have to be to enable such a scheme for the entire IPv6 address space?

8 words. 128 bit address space and 65k English words gives you 16 bits per word, so you need 8 words to cover the entire space.

My dictionary excludes all proper nouns and all words longer than 8 characters. Only about 60% of the 8 character words were used, chosen completely at random. Even so, given a full IPv6 address the resulting "phrase" is a mouthful. Shortened addresses are a little easier to deal with.

Dictionary size of what3words: (57*10^12)^(1/3) ~= 38485 [0]

Number of IPv6 addresses: ~340 undecillion [1]

Length of string of words: ~8.4 [2]

Probably closer to 8 as what3words likely doesn't cover the full 3-word-address-space of the English language.

[0] https://en.wikipedia.org/wiki/What3words#cite_note-16 [1] https://en.wikipedia.org/wiki/IPv6#cite_ref-rfc2460_2-2 [2] https://www.wolframalpha.com/input/?i=log+base+38485+of+340%...

I call dibs on Buffalo.buffalo.Buffalo.buffalo.buffalo.buffalo.Buffalo.buffalo

https://github.com/zamicol/ipv6-word

Once you really get into IPv6 you tend to find that the addresses are not actually that bad or at least not as bad as you might expect. For starters, your prefix stays the same and probably looks like this 2001:wwww:xxxx::/48 ISP is wwww and your allocation is xxxx. There are lots of ways you can break up your allocation: yyzz. For me yy is the first dot1Q and zz is the second tag. yy = 00 is for single tagged 802.1Q VLANs. I have quite a lot of "spare" space for expansion, VPNs or whatever. So using the https://tools.ietf.org/html/rfc3849 documentation prefix 2001:DB8::/32 - that's my "ISP". Let's say I am :2d1:, I give my routers ::1 ie the first address and therefore on my default VLAN with no tags, I get an address of:

2001:db8:2d1::1 for the router on that VLAN (10.10.0.1)

My DNS servers are on a single tagged vlan 10 (they are 10.10.10.11 and 10.10.10.12):

2001:db8:2d1:a::11 and 2001:db8:2d1:a::12

I wont be using that site or anything like it. DNS does the trick for me along with documentation, just as it does for IPv4. I toyed with ULA addresses (a bit like RFC 1918) but dropped it because it adds complexity. You only need a few addresses to bootstrap a network and a change of ISP will have more hassles to deal with than a prefix change.

Were you given a /48 for your residential IP, or is that for a business? Residential AT&T in Georgia assigned a /64 for me.

Typically it's pretty easy to get an ISP to cough up a /48 if you ask.

/48 for work (actually I have six of the bloody things!) and /56 for home which is reasonable. /64 is ridiculous unless they will give you more on request but then why not /56 so you can plan. We really are not going to run out of those and even /48 for all is pretty reasonable. At work I have an additional /64 just for WAN but I understand /112 and the like are popular for that as well.

I suggest you have a play with the /64 just for fun and to get the hang of things but the future has a shed load of IoT in it and all networks will need breaking up into subnets for security, including residential networks. Even if only to separate guest wifi.

In the UK we have the relative luxury of being able to choose from a fairly long list of ISPs. Some of them nearly get IPv6. Funnily enough dear old BT (business) is one of the few that gets it all correct out of the box, when you order a leased line.

You have my sympathies.

/64 is a lot of addresses. That's 2^32 internets.

But it’s only a single subnet or point-to-point link, you can’t do anything interesting with that space

When you see an IPv6 address mentally block out the last 64 bits. A /64 in IPv6 land is the equivelent of a single public IPv4 address for most practical purposes.

I would prefer to "A /64 in IPv6 land is the equivalent of a single public IPv4/24 for most practical purposes."

Back in the 80s or possibly 90s that might have worked, but an experienced admin today will look at a /24 and go "I can support 16 sites with that", whereas a /64 is really the smallest allocation you should see with IPv6, like a single IP address in IPv4.

My old ISP gave me only a /64. That's absolutely useless for doing anything more than there most basic of configuration. This seemed to be because of limitations in their own infrastructure as they were using 6rd. I ended up switching ISP's to one that gave me a /48.

I seem to recall general recommendations that ISP's assign at least /56 to customers?

One of the biggest things that keeps me using IPv4 is that I can memorize all the IPv4 addresses for most of my servers. That alone is a huge mental hurdle.

That just proves you haven't understood IPv6 yet. ULA addresses make it much easier, e.g. fd00::1 or fd00:2 (though this is bad practice, should be fdxy:zvwx:xyzw, xyvwz being random).

I thought this would be an issue when I deployed IPv6 on my home network, but it turns out that I memorized the prefix pretty quick and the host part you mostly end up ignoring.

$ host sprint.net

sprint.net has address 208.24.22.50

sprint.net has IPv6 address 2600::

wonder which one is easiest to remember? =)

Same/similar working mechanism is built-in into Windows since Vista: 1234-5678--abcd.ipv6-literal.net This doesn't even need functioning DNS and can work offline (add s<devnumber> for link-local addresses).

I actually had to look three times before I spotted it. Good one!

In case anyone wants to know the difference it's that the third "section" on the first URL is "9adc" and the third "section" on the second URL is "9abc".

I had to diff it! Even if there was no difference, the time it would've taken to be sure really illustrates the problem.

It would be interesting if there was even an informal URL extension that was able to specify alternative name resolution protocols. Maybe overload scheme in the url otherns-https://name

Neither of those loaded. Not tricked.

Reply on news.ycombinator.com