Hacker News Reader

Comment by lordnacho | Hacker News Reader

If IP addresses are scarce, why does it not seem to cost me anything to get one? I've had a bunch of online servers that came with a static IP, and an ISP service that came with one, at various times. It seldom seemed to be explicitly priced, and when it was the price was not really something worth considering.

View on HN

Because address allocation is inefficient and it is impractical to address this inefficiency in IPv4.

ARIN is out but your ISP has extras. Why doesn't your ISP sell its extras, since a scarce resource should have monetary value?

Routing tables.

Selling individual IPs off piecemeal isn't practical because routers simply do not have enough RAM to handle the enormous routing tables that would resultantly be necessary in order for packets to find their destinations.

To be specific here, RAM is not the problem. It's often on-chip memory/TCAM. RAM is abundant. The place where the hardware actually installs the routes (the data plane/FIB) is often constrained.

Good clarification

I have a hard time seeing how IPv6 solves this problem. If we allocate them all individually, we have a world-shattering routing table problem. If we allocate them in gargantuan blocks, we have the same problem we have now with IPv4 -- we can't reallocate because routing becomes an impossible task. What am I missing? We're certainly not going to get the allocation right by doing it all at the beginning.

With both IPv4 and IPv6, you want to allocate addresses hierarchically: network providers get large blocks, and delegate subsets to their customers. The problem is that with IPv4, addresses are scarce enough that customers only get as many as they can justify now (or in the immediate future). When that allocation runs out, they need to get another non-contiguous block. And this problem reoccurs on every level of the hierarchy, so ISPs themselves end up with lots of different blocks. Even if you manage to get a large block to allow for your own future expansion, you're just increasing the pressure on your neighbors.

With IPv6, we have enough bits of address space that we can drastically over-provision each level of delegation -- by orders of magnitude, if necessary -- and still have plenty of room to spare. e.g. as a home customer, my ISP gives me a /56 with no questions asked, which means 256 subnets each with an effectively unlimited number of devices.

In theory, this means it should become very rare for a single physical or logical network to have non-contiguous addresses. Which in turn means that routing tables can use a single prefix to cover an entire provider, instead of having to know about all of the individual sub-allocations.

In your example, your ISP has allocated you well over half the bits in an address. It's true that this is only 1/2^56 of the total space, but you can't overprovision very many levels to that extent no matter how long the addresses are (using the same proportions, the level above "home customer" might have a /25, the level above "ISP" might have a /10, and you could probably have two more levels on top of that with /5 and /2, although I'm not sure there's a lot of value in having a /2 layer protecting the /5. But I don't do internet routing; what do I know?

There are some rival pressures here:

- IPv6 space is so abundant (and cheap to allocate) that you can have as much as you ask for. This obviously leads to massive overdemand, after which it's hard to reallocate while preserving the idea that everyone who owns space owns one contiguous block.

- You can counter that by restricting how much people can ask for. But at that point you're not using the full space; you're restricted to the amount that can be allocated, which is smaller.

- Not everyone will be on board with "I'm going to take as much space as I can get, just in case I end up needing a million times as much space as expected." But some of those less-demanding people really will end up with addressing needs in excess of what they anticipated. We could call this the Hofstadter's Law ("it always takes longer than you expect, even after accounting for Hofstadter's Law") of addressing.

We can't expect everyone to get things right the first time; we need a working reallocation system. If reallocation doesn't work, abundance alone isn't going to save us.

I think you've pretty much got it, but what makes it work is that you're overprovisioned at each level. You don't have to set up an addressing scheme that's guaranteed to be able to satisfy every conceivable request as a contiguous range. The extra buffer means that if you run out at one level, you can get another allocation from the next level up without having to advertise it globally.

Right now, if you need more IPv4 addresses, you have to buy a /22 or so from someone else. Because the address space is so hotly-contested, that block could come from almost anywhere, which means every core router on the internet needs to remember where it lives on the actual topology.

With IPv6, maybe once in a while a customer comes along who needs more than a /56. No big deal; you just give them another (maybe non-contiguous, if you're unlucky) allocation from the ISP's block. The ISP needs to remember that both allocations belong to the same customer, and route packets accordingly. But everyone else can just route based on the ISP's all-encompassing prefix. That's what makes the system fundamentally more scalable.

Really good point about the ISP just reaching into it's /32 allocation. Also, If there is the slightest chance that you will ever need more than a /56, (I.E. you are anything other than a vanilla home business account) the ISP will almost always give you a /48 right out of the gate anyways, and so you have on the very first day 65,536 subnetworks to work with. After that, everything you say is also true, if the customer starts to run low on those 65,536 subnetworks, the ISP can simply reach into it's /32, and then hand out a second /48 in the rare scenario that someone is using more than 65,536 subnetworks that need to be globally routed.

>In your example, your ISP has allocated you well over half the bits in an address.

The reason for that is because as of IPv6, DHCP is is no longer the standard way of getting an address. With IPv6 autoconfig each device assigns its own /64 address (either based on MAC, or randomly generated), thus subnets smaller than /64 are simply not practical for most use cases.

Having a /56 means you get 256 distinct subnets to assign as you please, just like you did with 192.168.0.0/16

> In your example, your ISP has allocated you well over half the bits in an address.

Eh, think of IPv6 as allocating from a 64 bit subnet space. So they're only allocating 1/8 of the address space.

> using the same proportions, the level above "home customer" might have a /25, the level above "ISP" might have a /10

More like /40 and /28

And once you are in the million-customer range you don't really need prefixes any more, but that gives room for more layers of generous prefixing and still having most of the address space spare.

Basically: you can't give everyone as much as they might ever ask for, but you can give everyone 10x as much as they seem to need and still have a dozen bits to spare

> What am I missing?

A few things.

There's only 4,294,967,296 addresses in IPv4, so if you hand out giant blocks, you run out quite fast, whereas with 340,282,366,920,938,463,463,374,607,431,768,211,456 you could hand out huge IPv4 /8 sized blocks from now until the heat death of the sun if you wanted to.

But there's more to it than that. Right now only 1/8th of the IPv6 address space is "assigned to the internet", so there's effectively 7/8ths being held in reserve for the foreseable future -- this is partially to accommodate realistic router memory limits in the near term. IPv6 also has a more efficient hierarchical routing scheme than CIDR (which was kind of bolted on to IPv4 after-the-fact), which helps keep router tables from consuming terabytes of RAM. And IPv6 is designed to be more easily renumberable when we do make mistakes with assignments, with a clear separation between subnet prefix all that requires is changing the routing prefix.

At the end of the day, with 340 trillion trillion trillion addresses, handing out too many isn't ever going to be a big worry in the foreseeable lifetime of...anything, really.

The statement about IPv6 [internet] having 2^128 addresses is completely and utterly untrue, in the same way as it would be untrue to claim that IPv4 [internet] have 2^48 TCP ports.

What IPv6 have address-space-wise, is 2^64 subnets, out of which 2^61 is actually allocated to be routable over public internet.

Then comes granularity of /48 allocations (= 2^16 subnets) to end users and /32 allocations (= 2^32 subnets) to ISPs. That gives just 2^45 end users and 2^29 ISPs at best. In practice, if handing a user a /48 would become a common practice, all medium and large IPSs would start accumulating multiple /32s, then routing concerns would come back, and largest ISPs will start aggregating to /28s or /24s or whatnot.

That is still a whole lot of addresses (and IPv4 still looks laughably small in comparison), but actual, practical numbers are way, way lower then "340 undecillion" marketing figure being plastered all over the place.

From what I've read, most end users will get /56 blocks and larger organizations will be the ones to get a /48, so the potential end users is much higher than you've estimated.

Blanket policy of /48 per end user comes from a (now obsolete) RFC3177. All RIRs now have less generous recommendations - both RIPE [0] and ARIN [1] seem to operate in /56 granularity nowadays, which is a very sensible step forward of course.

[0] https://www.ripe.net/publications/docs/ripe-650

[1] https://www.arin.net/policy/nrpm.html#six7

The fact that only 1/8th is assigned to "the Internet" right now is worrying. After all, we know how that worked out with the 240.0.0.0/4 range in IPv4 - it was reserved for future use, but was never able to be used because by the time it would have been useful, everything classed it as non-routable. It's a huge swath of addresses that just can't be used.

The only solution to that sort of thing is just to be bloody minded.

Make it known from the start that when the currently unallocated blocks are needed they will be used, and if that breaks things for people who have used them already that is their fault and their problem to fix.

To be clear, I don't mean that 240.0.0.0/4 isn't able to be globally routable because everybody's using it for their own private stuff. I mean that 240.0.0.0/4 isn't routable at all.

Even if everything and everyone fails to learn the lesson of 240.0.0.0, and ignores the standard, and hard-codes the assumption that 7/8's of IPv6 is unusable forevermore, 1/8 of 340 trillion trillion trillion is still unfathomably huge

> There's only 4,294,967,296 addresses in IPv4, so if you hand out giant blocks, you run out quite fast, whereas with 340,282,366,920,938,463,463,374,607,431,768,211,456 you could hand out huge IPv4 /8 sized blocks from now until the heat death of the sun if you wanted to.

Well, according to [1] the sun will stay white-hot effectively forever. But, given that it will continue in its current state for another 5 billion years, and that there are 2^96 IPv4-sized blocks in IPv6's address space, one could give out 2.5e21 IPv4-sized blocks every second from now until the end of the sun's lifespan.

So yeah, IPv6's address space is really, really, tremendously, amazingly, enormously big.

[1] http://astrosociety.org/edu/publications/tnl/39/sun2.html

This made me realize just how big a 128-bit key in cryptography really is.

Indeed! It's bonkers. My college security professor once described a 128-bit key to us by imagining that every electron estimated to be in the known universe (think on this concept for a moment) had it's own unique value.

For a 256-bit key, give each electron from above it's own universe, and assign values to all. That's the keyspace to be broken.

Bruce Schneier once did a little thought exercise where he demonstrated that, assuming one needed to shift one electron (the absolute minimal amount needed to conceive of a compute switching operation), the energy requirements needed to brute force the key would exhaust the energy potential of the solar system.

"These numbers have nothing to do with the technology of the devices; they are the maximums that thermodynamics will allow. And they strongly imply that brute-force attacks against 256-bit keys will be infeasible until computers are built from something other than matter and occupy something other than space."

Interesting to ponder.

(Source: https://www.schneier.com/blog/archives/2009/09/the_doghouse_... )

I suppose the idiom is normally supposed to be "heat death of the universe", but yeah, either way, the IPv6 address space is just galactically, unfathomably huge

I was trying to explain this to my boss the other week and after a little wolfram alpha came up with it being a lot less than the number of atoms in the universe but a lot more than the number of stars.

Seemed close enough.

"At the end of the day, with 340 trillion trillion trillion addresses, handing out too many isn't ever going to be a big worry in the foreseeable lifetime of...anything, really."

This is untrue, and a big over exaggeration.

Consider that pretty much any business (even a mom-and-pop shop) is allocated a /48 if they ask for it. Assuming one thousand /48 are allocated every second (ie. one thousand "business" IPv6 users coming online per second, which might be absolutely plausible 100+ years from now --we made IPv6 to be future proof right?), then we would run out of allocatable blocks in the global unicast IPv6 space (2000::/3) in 1,115 years. And we would run out of allocatable blocks in the entire IPv6 space in 8,919 years. So, yeah, I get annoyed when people mention the "death of the sun" or the "lifetime of the universe" when talking about IPv6, when in fact we could start running into exhaustion issues in the next ~1000 years.

Note that I am being pedantic here. I think any technology that seems to scale "only" for our needs up to ~1000 years from now is good enough :)

By my calculations, you could assign 1000 /48 subnets per second for 4.8 trillion years before exhausting that space.

http://www.wolframalpha.com/input/?i=%282%5E125%29%2F%282%5E...

You made an error. 2^48 should be replaced with 2^(128-48): http://www.wolframalpha.com/input/?i=%282%5E125%29%2F%282%5E...

That seems.. correct. My bad!

> At the end of the day, with 340 trillion trillion trillion addresses, handing out too many isn't ever going to be a big worry in the foreseeable lifetime of...anything, really.

Gates already made that mistake.

I know "this time is different", but if you start thinking of "internet of things", where literally every little thing in the world can be called (so needs its own name/number/ip/etc), then at some point we might run out of those numbers as well.

I'm not sure how much (2^128) / 7 000 000 000 is, but it seemed not alot of every human's property can have its own address.

> I'm not sure how much (2^128) / 7 000 000 000 is, but it seemed not alot of every human's property can have its own address.

Well, to put it into perspective, imagine that humans would colonize planets around all 400 billion stars in our galaxy (let's say that each star has 10 planets), with 7 billion people on each of them, and each person having 7 billion devices, each device could still have almost 2 million unique addresses available. The range by itself IS big enough for any foreseeable future. The problem may come from the fact that not all that range is available for use.

Every human can have trillions of addresses and not use that up.

Then every "device" you have connected with every other person's device might need its own public address.

And all of sudden, 640 kilobytes is not enough.

> And IPv6 is designed to be more easily renumberable

Can you point me to some information about this? I apologize if this is an easy question I am interested in ipv6 but only know enough to be confused more often than not.

> There's only 4,294,967,296 addresses in IPv4

In an internet-usable sense, there's fewer than that, too. The entire 127.0.0.0/8 block is for localhost/loopback, for example.

https://en.wikipedia.org/wiki/IPv4#Special-use_addresses

Worse: 224.0.0.0/4 and 240.0.0.0/4 are reserved, have always been reserved, and cannot in any practical terms be allocated, as there's huge amounts of hardware and software in production that's hard-coded to treat them as invalid. That's 1/8 of the entire IPv4 address space that's completely unusable.

224.0.0.0/4 is IPv4 multicast space. It's not reserved and often in active use. [0]

If there's software that treats IP traffic to or from 240.0.0.0/4 as invalid, then it's defective. I mean, I get that the broken software is a concern, but -if we care about carving out more IPv4 space- it's not something that we should be overly worried about. :)

[0] Indeed, mDNS/Bonjour/Avahi uses 224.0.0.251 for service discovery.

Perhaps I should clarify: That range isn't general-purpose IP space. And while multicast is useful, there's probably ways to implement it without sacrificing 1/16 of the entire address space.

> ...while multicast is useful, there's probably ways to implement it without sacrificing 1/16 of the entire address space.

It's far too late for that now. :( Perhaps if IPv4 multicast were more widely used on The Internet the large carve-out would look more sensible to you.

> That range isn't general-purpose IP space.

Yeah, that's a true statement. Just remember that "reserved" has a particular meaning when it comes to IP address ranges. :)

> At the end of the day, with 340 trillion trillion trillion addresses, handing out too many isn't ever going to be a big worry in the foreseeable lifetime of...anything, really.

Well, why can't I have them all?

You'll have to settle for a measly trillion of them

Certain DARPA-funded emerging technologies[1] will more than likely need just a wee bit more than "a measly trillion".

Before spending some time in a wireless sensor net research lab as an undergrad, I had a difficult time imagining the scale of an application that could justify such a massive network address space...probably because I was too focused on trying to visualize what it might look like without considering the possibility that it would be invisible to the human eye.

[1] http://www.eecs.berkeley.edu/~pister/SmartDust/SmartDustBAA9...

consider that there are something like 3.35 * 10^22 water molecules on earth

there are 3.4 * 10^38 addresses in the IPv6 space

we could individual address every water molecule on earth and still have an enormous amount of space left over

so even if DARPA is planning on replacing the oceans with dust (and somehow managing to not exhaust a number of critical resources in this drive to make a truly staggering volume of dust), it's still not an issue

My thought process isn't as terrestrially constrained...qualitatively science fiction by today's measure, to be frank.

I was thinking more along the lines of an architecture adopted today which survives many millennia into the future that accounts for exponential human population growth on a galactic scale, isn't limited by the natural resources of this planet, where increment() will be the only pragmatic address assignment operation, meaningful archaeologic insight might be attained by discovering a microscopic 128-bit invariant, and so on.

I mean, yes, IPv6 fails the improbable science-fiction wish-list test, for any number of reasons beyond address space

path MTU discovery algorithms aren't going to be efficient when packet transit times are measured in decades, for one thing

> If we allocate them in gargantuan blocks, we have the same problem we have now with IPv4

The problem with IPv4 is that when ISP can't get more, so they have to put some unlucky users behind a NAT. The company I worked for got /21, but we had way more than 2K simultaneously connected users. And, well, we failed the timing to get another allocation, and (I had left since then) is even more unlikely they'll get one those days.

IPv6 solves that by making all the blocks giant. With IPv6, when ISP is allocated /48 and allocates /64 per client - comparing to IPv4, it's like everyone gets a mighty /16 block right from the start, so even a tiniest company can grow to 65K users. And if they grow more, there is no problem in getting another subnet, those are not scarce, too.

With IPv6 the smallest allocation an ISP gets is a /32 (and it's pretty easy to get more as the ISP starts to fully utilize that /32). And even a small business can get a /48 from an ISP with almost no questions asked. That means, starting out, a company can grow to 65K networks (each network with effectively unlimited number of nodes on it). And, just as you said, getting another /48 won't be a problem.

Ouch. Yes. Sorry, you're absolutely right, it's /32, not /48. Was sleepy writing that comment and got numbers messed up in my head :)

The logic still stands, it's just that problem's solved by just making everything giant.

There are 2^128 IPv6 addresses. The number of "gargantuan blocks" possible with that many addresses is quite large. The problem of routing the allocated blocks is solved incrementally as needed by network expansion. In the event we need to route more than we can now routers are upgraded or topologies are changed.

Let's say we make a list of all IP allocations today. We could replace each of them with 16 million unlimited-size subnets worth of address space, and we'd still only be using .00004% of IPv6. And that's before we consolidate multiple blocks owned by the same user.

There are currently half a million active BGP entries. There's no reason to massively change that number.

Seems it'd be nice to "defragment" the IP address allocations, but totally infeasible since everyone's IPs would change and all the DNS records would point to the wrong places.

/24 can easily be moved. And if I have 65000 /24s with varying utilization (like Apple,) and could turn that into money by coalescing and adding NAT, that would drive change.

The price of an IP is usually included in the price of the service. Note that in some parts of Southeast Asia, you can pay for a cheaper "web access" which is not a full Internet connection with a public IP, but essentially puts you in a huge LAN behind a NAT.

Because while those IPs have been allocated it does not mean they're being currently used.

So, ARIN can't allocate any more to providers but many providers are not yet feeling the pinch yet so they're not passing the cost onto you.

Give it a time and you'll start seeing IPv4's coming at a premium and IPv6's being free.

edit

Also interesting, check out whose holding the /8's. [1] Some of them make sense like AT&T, but others don't...like MIT. Why's a university holding 224 ip addresses? Well, they just happen to be one of the first network holders. HP has 2 /8's which sounds pretty absurd (unless there's some reason I don't know that HP needs over 32 million ip addresses).

> Why's a university holding 224 ip addresses?

224? A /8 leaves 24 bits for the hosts. That's 16.7M addresses

Presumably that was supposed to be 2^24.

That would be correct.

One of HP's allocations is from DEC I believe.

Yeah, it is. But that's a hell of a lot of real estate for a single vendor whose not even providing significant network services.

Static or dynamic, the cost to the ISP is basically the same (perhaps a bit more for static if it's harder for them to manage the allocation); therefore, it's already built in to the cost of the service.

Have you tried getting additional IP addresses? One ISP charges me 50 cents/month per address for my colo box, while Comcast Business charges around $2/month (IIRC).

>Have you tried getting additional IP addresses?

Yeah. The price doesn't suggest scarcity. I guess the thing with the routing tables makes sense. I wonder if there's a way to rationalize that without it getting too messy. IIRC ipv6 thought of this and split up its address space in a roughly geographical way.

The price doesn't yet suggest scarcity because most ISPs have not yet had to start buying IP addresses because going to e.g. ARIN has been an option, and so they've not been pricing in acquisition costs.

It will rise, in proportion to the pain people experience in reducing their usage and/or switching to IPv6.

It does cost you, it's just calculated into the price of the service.

Try looking at for example a service like LowEndSpirit[1], which offers NATed VPS services for 3 euro a year(!) - it would be literally impossible to offer this service with a dedicated IPv4 address, and indeed the dedicated IPv4 VPS services typically start at $12-$15 per year.

Most of that $12-$15 is IP costs.

[1] http://lowendspirit.com/index.html

EDIT: Oh yes, no Markdown. Right.

Because IP addresses are NOT scarce. They are so plentiful and so cheap that a single residential user will get a /54 block containing 18 billion billion IPv6 addresses.

There is no catch. All you have to do is use IPv6 (which you are already using unless you have explicitly turned it off). Many people have IPv6 network traffic running on their networks without even realizing it because it is included by default with every operating system.

The article that was posted only refers to the old-fashioned simplistic IPv4 addresses and they are only still in use because too many graybeards want to stick with what they learned 30 years ago. Learning new technology is not that hard. Run with it.

Most consumer ISPs still do not support native IPv6 (at least here in Canada, pretty sure it's about the same in the US). OS support is one thing, but actually having network access is another.

at least here in Canada, pretty sure it's about the same in the US).

We're getting there. TWC is one of the largest ISP's in the US, IIRC, and they have said they have IPv6 support for about 90% of their residential network. Anecdotally, I started getting an IPv6 address from DHCP on my cable-modem connection about 2 months ago.

We're a long way from full IPv6 coverage, but now you finally start to get the sense that it's real, whereas in the past (even a year ago) it was easy to wonder if the ISPs and what-not were even really trying.

>90% of their residential network

I'd like to see where that number is coming from. I've been a TWC subscriber for 9 years, I currently subscribe to their fastest residential offering in my region as well as one of their commercial offerings in the same region.

Both modems are DOCSIS 3 and IPv6 capable, neither receive IPv6 addresses. I have an AICCU tunnel setup on my residential, though it only gets a fraction of the potential performance. (Rather annoyingly the tunnel's routing is also far from perfect: for e.g I can't reach Netflix via IPv6; but because the DNS resolves it doesn't attempt to downgrade to IPv4.)

Considering I'm only 2 hops from their Chicago POP I sincerely doubt they have anything approaching 90% IPv6 rollout.

I'd like to see where that number is coming from.

I don't know how they arrive at it, or how accurate it is, but here's what they claim:

http://www.timewarnercable.com/en/support/internet/topics/ip...

    Time Warner Cable was an inaugural participant in World IPv6 Launch.  
    TWC has rolled out IPv6 to over 90% of its residential network.

Both modems are DOCSIS 3 and IPv6 capable, neither receive IPv6 addresses.

FWIW, I have a relatively new Netgear router (as in, about a year old, maybe less) but even after TWC upgraded my cable modem to one of the newest ones, I still did not get an IPv6 address until I updated to the latest firmware on the Netgear router. Depending on what you have hanging off of your cable modem, I suppose there's a chance that you're not getting an IPv6 address because of some issue on that device, as opposed to TWC's network itself.

Comcast, the largest isp in the US, already claims to have ipv6 support in all markets since 2014. AT&T the next largest also claims full support. I would guess the majority of US residential internet users have ipv6 support from their ISP but they are just not using it because they have old modems or routers that dont support ipv6

I can confirm that in San Francisco, I've had a DHCPv6-PD assigned allocation for far longer than a year. When Comcast claimed to throw the switch to enable IPv6 for all of their residential customers a while back, something wonky happened to my IPv6 service (and stayed happened for several days) that required me to reboot my cable modem (which then proceeded to download a new config file and reboot several times) in order to resolve it.

I actually think this is less true than it used to be; the problem now is mainly on the server side. A substantial fraction of US residential customers can get IPv6 access as long as their network equipment supports it. But aside from a few prominent names -- like Google, Wikipedia, and Facebook -- the vast majority of the internet is only accessible if you have a v4 address.

From the latest statistics I could find, about 7-8% of Google visitors are connecting via IPv6 [1], but only about 1% of the Alexa top 10,000 sites are v6-enabled. [2]

[1]: https://www.google.com/intl/en/ipv6/statistics.html

[2]: https://www.vyncke.org/ipv6status/allgreen.php

that 7-8% is also worldwide, in the US it is 21% now

at least here in Canada

TekSavvy covers some of the bigger centers in Canada and has supported IPv6 for some time.

Reply on news.ycombinator.com