Hacker News Reader

Comment by omeid2 | Hacker News Reader

Something that isn't talked much about IPv6 and I believe has subtle and indirect influence on it's adoption is that it is far far far less human-readable than IPv4.

You can't find me a single sysadmin or casual user who would look at an IPv6 and say, nice, I prefer that one over an IPv4 or find it easier to type or communicate.

You can argue about merits of IPv6 and very reasonably so all day long, but this fact remains.

I don't think governments or other institutions can strong-arm IPv4 out of existence. I don't know what the long term solution would look like, but I strongly doubt it will be IPv6-only. Ever.

View on HN

I'm a sysadmin and find IPv6 easier to use. You can easily make short ipv6 address just by having a larger prefix and then omitting zeroes (with `::`) in your host-part that you don't need.

Take `2001:db8::1`. Here `2001:db:8` would be your network address and `1` your host-id. For example, a router. `2001:db8:2` could be a server and so on.

I don't imagine it's easy to get a global /32, so do you mean for purely local networks? In which case why not use 1::1?

Its easy to get a global /44 - /48, which is not that much more.

The prefix I get from my ISP is a complete 56 bits, more or less. Thus even if I use ::1 on the inside, it's very long and tedious to type. ULAs are better but are still supposed to be 48 bits[1].

[1]: https://en.wikipedia.org/wiki/Unique_local_address

"IPv6 addresses are ugly" is only an argument for as long as ipv4 addresses don't cost anything. as soon as they put a price on it, nobody's going to care.

Arguing that "we can price IPv4 out" only reinforces my point; of course, ignoring the fact that it is a pipe-dream. But the fact remains, no one really prefers IPv6 over IPv4.

I do.

Just like users prefer domain names over ip's. But they are necessary.

I started to prefer it after I foundout that with IPv6 every Docker container can have its own IP address and this can be used to solve the hairpin NAT problem.

That happened a long time ago.

And yet still, nobody cares.

AWS starting to charge per-address next week is making a whole lot of people start caring.

There is over 1 billion users with less than 5% on IPv6 in China alone. Some 200 million users in Indonesia with less than 15% on IPv6.

Hundreds of million of users in Africa that are growing at a very fast pace, still on IPv4.

In Europe outside Germany and Belgium, the IPv6 usage is between 5-20% max. Millions of people.

There is also millions of industrial systems, ATMs (heck, some of them still run XP!), and whatnot that will not get IPv6 short of complete replacement.

You can't price IPv4 out with that kind of numbers combined with the fact that it is an extremely open market.

IPv6 adoption is low now, but the answer isn't to just give up and stick with IPv4 forever.

As the extra cost of IPv4 keeps rising it will sooner or later outprice the cost of moving to IPv6.

Similar as with fossil fuel cars -- super convenient, there are petrol stations everywhere. But as the price started rising (assisted by very harsh taxation) suddenly the idea of switching to electric didn't seem so bad.

Where are you getting the numbers from? According to APNIC, >30% of China is on IPv6 right now.

There is a very sizable IPv6 population in Asia.

https://stats.labs.apnic.net/ipv6

CISCO and Google, I am not sure what kind of sampling APNIC uses but that does seem pretty off.

https://6lab.cisco.com/stats/cible.php?country=CN&option=pre...

https://www.google.com/intl/en/ipv6/statistics.html#tab=per-...

Note that Google is restricted in China. Their data is not necessarily accurate.

The original idea was that configuration would be so automatic that humans would never see the addresses. Instead of an admin assigning a name and IP address to each host, the admin would only assign the name. The host would figure out its local topography from Neighbor Discovery then tell the dns where it's connected. I'm not sure how well that worked out in practice.

It didn't work in practice. In many situations names don't work. One example would be games with local lan support, and you use IP addresses to make sure you are connecting to the right server. Other would be tiny embedded devices, that are not powerful enough to scan the network and need a hard-coded or user configurable IP address to connect to the right place.

> that are not powerful enough to scan the network

just look at the ND table. Nobody is ever going to scan a single /64 of v6, ever.

Then why do I still have to see the addresses?

...

how would configuring DNS even work without having to see the addresses, and how would SSH'ing to a device on the LAN work without seeing the addresses

NetBIOS/WINS protocol could discover names -> IPs by broadcast, more modernly Zeroconf/Bonjour can do it, and if there's some organization then DHCP servers can register hostnames in DNS when handing out addresses.

That requires every device to be given a LAN-unique name...

...which a local IP address is, conveniently

ipv6 is so design by commitee that it's ... not worth the trouble ... as a hobbyist. You just described a scenario where you need 3 services for the machines in the same local network to talk to each other.

I'm not willing to spend the time to set it up on my internal network until I start to lose connectivity to some places, sorry.

To be clear, I'm not defending it, just describing it. I was at the IETF meetings where all this was being discussed and remember thinking it would never work.

When people tell me ipv6 is already widely used, I ask them what name I should use when I want to ssh to their ipv6 device, their phone for example. People don't notice ipv6 isn't working because the Internet has evolved into a world of two classes of people, the rich and powerful like Amazon with addressable IP endpoints, and the second class citizens like you and me who can only connect to the rich and powerful, not to each other.

Technically I have a business connection going into my home with a public fixed ipv4 and probably a fixed ipv6, if only i could be bothered to configure pppd and whatever 5 other daemons i need to install.

Problem is, that's not my only fiber. I have two from different ISPs (Romania, so they're cheap). With ipv4 i only need to change the default route to the other router to choose which ISP i use. With ipv6 where my internal machines get an address from the ISP, i don't know where to start.

And I definitely don't want to spend a couple thousand on an "enterprise" multi home router that will do it for me...

> Something that isn't talked much about IPv6 and I believe has subtle and indirect influence on it's adoption is that it is far far far less human-readable than IPv4.

No shit Sherlock: an IPv4 is 4 bytes. An IPv6 requires 16. This is four times as much data, it has to be less readable, that’s just the price of a bigger address space.

Now one could have argued that 8 bytes, heck, even 6 bytes, would have been enough. I guess it would have, but having a /48, or even a /64 range, per user, is quite convenient: we can ditch the NAT (we probably won’t, but we can).

But even then, 8, or even 6 bytes, remain harder to read than 4.

The number of bytes isn't what makes it hard, "habanero eardrum" is 16 bytes and much easier to remember, there probably is a way to make ipv6 addresses easier to use, but hard to do now.

[prefix]::1 [prefix]::2 ... [prefix]::ffff

Or if you want mnemonics on your /64:

[prefix]::dead:beef:1 [prefix]::b00b:1 [prefix]::b00b:2

etc

"habanero eardrum" is actually less than 4 bytes -- let's say 12 bits for a common word (eardrum) + 16 bits for an uncommon word (habanero).

If it were 16 bytes, the other options from the same 16-bytes space would be, for example, "\xa3\x80W%\xa3\x82\xa1\xea\x10\xf9\x8b\x07'\xf93J" or "\xf0\xef\x9b\x8f[0\xe5\xb9,\x0b\xd4^\xb00\xed\x00" (chosen by a fair /dev/urandom roll).

If you want to convey 16 bytes using similar encoding, you need to use about 10 to 12 human words -- see for example Bitcoin wallet 12-word seed phrases or https://xkcd.com/936/ (xkcd correct horse battery staple).

For IPv6, if your addresses are not SLAAC, but DHCP or manually assigned, you can go with half of it - the "random" part is the 64bit network prefix.

We just need to agree on a 2^16 word dictionary, each 16-bit segment gets it's own word. Then we can write addresses such as correct:horse:battery:staple::one .

As a private user, as long as I remember the prefix assigned to my connection I much prefer having a stable [prefix]::dead:beef or ::f007:1 and ::f007:2 etc compared to the ever-changing 85.nnn.nnn.nnn v4 address.

Or simply just /think/ of it as a /112 subnet with [prefix]::0001-ffff available - still room for 65535 addresses in that one, more than I'll ever need.

No need to rely on long autogenerated SLAAC addressses if you don't want to, you can have whatever you want after your prefix and you don't have to pad it out with garbage random-ish values. A sysadmin can get creative :)

Of course your ISP will need to have proper v6 support.

How often are you communicating IP addresses? Is that not the point of domains, because humans are rather terrible at remembering sequences of numbers?

Sure, if DNS was always reliable 100% of the time.

Also, in our desktop/laptops, we set the DNS servers to an IP address, not to a hostname.

I believe it's more about using :: as separator instead of dot. There must be a reason why no one wanted to copy it from C++.

> Something that isn't talked much about IPv6 and I believe has subtle and indirect influence on it's adoption is that it is far far far less human-readable than IPv4

What? This comes up in like 95% of ipv6 threads on HN (see also: ‘they should have just added an extra quad to v4 addresses’)

    My grand father used the dots in the address!
    My father used the dots in the address!
    I used the dots in the address!
    Henceforth every human being should use the dots in the address till heat death of the universe!

I'm certain the issue here isn't the dots.

123.123.238.217.110.100.1.1 would still have been perfectly readable.

fdd2:1228:3372:1sdf meanwhile is pretty unreadable even at IPv4 length.

ipv6 is 16 bytes, so more like

123.123.238.217.110.100.12.255.123.123.238.217.110.100.242.176

True, but we probably don’t actually need enough bytes to address all the atoms on the surface of the earth a hundred times over.

Good! If we did then we should've made it bigger.

We don't know exactly how many address bits we're ultimately going to end up needing. Our only options are to go for "too big" or "too small". Giving how much effort it takes to deploy a new IP protocol, we should err on the side of having too many addresses rather than too few.

I don’t understand… why do you consider the way longer one to be more readable than the shorter one? I find the colon string a lot easier to read, a lot easier to compare at a glance, and a lot easier to memorize.

I thought the whole purpose of DNS was to avoid having to deal with IP addresses?

And how do you configure your DNS and network subnets?

Hoe does it affect DNS adblocking?

ipv6 should render any use of static blocklists moot, but i would not hold my breath for blocklist vendors to go out of business

Reply on news.ycombinator.com