Hacker News Reader

Hacker News

Comment by bertolo1988 | Hacker News Reader

Full thread

bertolo1988

Why not IPv16 and fix the problem for ever?

View on HN

IPv27, with 3192-bit addresses, for when you're running TCP/IP on quarks.

Will it fix it? This seems like a scaling issue we're going to have to deal with forever.

IPv6 has a 2^128 address pool.

edit: theoretically, we'll only ever use a fraction of that, both by design and because that's a huge number

Well, because there is no equivalent to CIDR with IPv6, what you really need to focus on is the 2^64 networks available. Still a big number.

That's what I meant in my edit :)

And yet I still have /127, /126, /120 and /112 IPv6 subnets running on my network.

Just because the IETF says SLAAC or bust doesn't mean it's true.

Genuinely curious - Why? We were considering doing so at one point for PtP links (equiv to a /30 or a /31 if you area really bleeding edge) - but then decided that there wasn't any reason to not just use /64s everywhere.

I take your point though - it certainly is theoretically (and in practice) possible to doing something other than a /64 - I've just never seen it in the wild.

I have heard of one good reason: with a /64, someone sending packets to random addresses within that /64 will fill up the neighbor cache, while with a smaller network like a /126, that can't happen.

You may be underestimating how many IPv6 addresses there are! 2^128 is huge. Like, if we assigned one trillion new addresses every second since the birth of the universe, we’d still have enough left over to do the same in another 800 million parallel universes. Truly astronomical numbers.

Obviously not all of these addresses will be used; in practice we aren’t going to run out any time soon.

For reference, 2^128 = 340282366920938463463374607431768211456

340,282,366,920,938,463,463,374,607,431,768,211,456

While you're right, it is huge. This mistake has been made in computing time and time again. You think you've got adequate scale for your solution but you didn't anticipate X which introduces new growth that out paces your design.

Assuming the internet has been growing exponentially since the beginning and will continue to do so indefinitely.

IPv4 has 32 bits and lasted us 34 years so far. Let's call it 32 years. Address space would need a nice and tidy 1 bit extra per year. That would mean IPv6 would last until around the year 2110.

But will the internet keep growing exponentially until then? How many people will even be alive by then? According to https://en.wikipedia.org/wiki/Projections_of_population_grow..., predictions used by the UN range from 7 to 18 billion people. Let's round it up to 20 billion people. How many IPv6 addresses could we give out to each person? According to https://www.wolframalpha.com/input/?i=2%5E128+%2F+20*10%5E9, 1.7 * 10^28, which is a lot. But okay, we have a lot of computers per person and not all computers are associated with individuals. So what if we do something crazy, like give out IPv6 addresses to the (human) cells in everybody's body? https://www.wolframalpha.com/input/?i=1.7+*+10%5E28+%2F+cell..., that would be about 1.7 * 10^14. What if we give each cell some IPv6 addresses per second (pretending all of our cells live for 80 years)? https://www.wolframalpha.com/input/?i=1.7+*+10%5E14+%2F+(80+...

Here we land on 67 thousand addresses, per second, for each human cell in the bodies of 20 billion people, for the span of 80 years.

That's how many IPv6 addresses we'll need to use before we run out.

We have made this mistake before, but 128 bits really is a lot.

Of course this assumes we don't assign half of the address space to Apple, Facebook, Google and their friends, or something silly like that, like it was done with IPv4 (with the equivalent "friends" at the time).

I read in one of the IPv6 RFCs (don't recall which one at the moment), that initially only 1/8 of the possible address space will be allocated. If for some reason that space is filled up, there's still another 7/8 which can be allocated with a more conservative policy.

End users should receive /64 subnet and servers should receive /48 subnet. Given that many users might want to run some servers, it's just 65536 times more than IPv4. Good enough, but not future proof enough if you ask me.

Current recommendations is end users of ISP's should get a /56

> fix the problem forever

Situation: There are 14 competing standards

"14?! Ridiculous! We need to develop one universal standard that covers everyone's use cases." "Yeah!"

Situation: There are 15 competing standards

Src: https://xkcd.com/927/

If you're getting downvoted and can't work out why, that XKCD is massively overused and shoehorned into conversations that have no bearing on what's being discussed.

The whole '15 competing standards' meme only really makes sense when there are 15 competing standards bodies. In the case of IP addresses, there is only one standards body, the IETF. There is one legacy standard (IPv4), one new standard (IPv6), and the only question is how best to ease the transition between the two. I see no problem if that transition relies on an interim protocol.

Win some, lose some. In hindsight it was pretty shoehorned in.

If only we didn't have multiple competing transitional mechanisms defined as standards already.

You're suggesting there are networking standards that compete with IPv4/IPv6? What are those standards?

There's like 4 (?) different IPv4/v6 integration systems. None of which worked the last time I fiddled with some of them.

Which would suggest there's room for an integration system that does work. In any case, the goal is a transition to IPv6, as long as the integration systems all share that goal there can be interoperability between them.

Reply on news.ycombinator.com