routed /48

[xezlec]

Oh, and another thing, putting extra space at the top means you can always allocate it downward to people as needed, but allocating it at the bottom means you can't easily get it back if needed.  So it's dumb that we're putting all the padding at the bottom of the address and none at the top (especially since the top is where we're seeing all the growth in the industry).

[jgeorge]

Basically, /48s are what the IANA recommends to assign to end users with more than one LAN, or businesses, per site.

And yet we're giving each residential home user a /48?

Nope, nobody ever said that. As far as address allocations go, a "residential home user" is not the end consumer of IP address space. Their ISP is the end user of that address allocation.

Joe

[xezlec]

Basically, /48s are what the IANA recommends to assign to end users with more than one LAN, or businesses, per site.

And yet we're giving each residential home user a /48?

Nope, nobody ever said that. As far as address allocations go, a "residential home user" is not the end consumer of IP address space. Their ISP is the end user of that address allocation.

Joe

It depends how you read the recommendation in the RFC.  The way I read it, an ISP gets a /32 and a /48 is only for a single office or building.  Hurricane Electric offers /48 tunnels to pretty much anyone who wants one, right?

[cholzhauer]

The way I read it, an ISP gets a /32 and a /48 is only for a single office or building.  Hurricane Electric offers /48 tunnels to pretty much anyone who wants one, right?

Any one can have a /48 that wants one.  Technically speaking, isn't a house a building?  With that logic, they get a /48 too

[jimb]

A /48 per customer isn't set in stone, but there is enough IPv6 space for the foreseeable future.  I could see ISPs giving out /56s or even /60s to end customers.  Or just individual /64s as needed.  It's really up to them, there address and routing plan, etc.

But /48 per site wont "break the bank" for the foreseeable future.  Considering only the current slice of the IPv6 space in use today, 2000::/3, there are ~536 million /32s.  If you gave ISPs, super large datacenters, govt agencies, etc,  1000 /32s each, there'd be ~536,000 of them to give out.  Think about that.  You'd have to give out ~536,000 1000 /32 netspace chunks of just the current in-use /3 to exhaust the entire /3.  Are there even that many ISPs, dataceneters, govt agencies in the world?  And each would get 1000 /32s, which have enough space for ~65,536,000 /48s to give to customers.  And oh yeah, after that there's seven more /3s to use.

By the year 3000, we won't be using IPv6 anymore, since other factors (the "unknown unknowns" you speak of) will have likely made IPv6 obsolete.  In the year 3000 the world won't even be recognizable to us, even if the human race is still around.  :P  That's 1000 years.  Think about the world we have now, vs. the world in the year 1011.  

[xezlec]

A /48 per customer isn't set in stone, but there is enough IPv6 space for the foreseeable future.  I could see ISPs giving out /56s or even /60s to end customers.  Or just individual /64s as needed.  It's really up to them, there address and routing plan, etc.

So could I.  That's why I'm saying they should do that.  This thread started because Hurricane Electric doesn't do that and one server ran out of /48s.

But /48 per site wont "break the bank" for the foreseeable future.

I was trying to show numerically that it may get tight, if you look at my numbers.  And it's just unnecessary.  A little common sense now will prevent even the possibility of that, without any negative consequences.  There are no ill effects from allocating /56s instead of /48s.

Considering only the current slice of the IPv6 space in use today, 2000::/3, there are ~536 million /32s.  If you gave ISPs, super large datacenters, govt agencies, etc,  1000 /32s each, there'd be ~536,000 of them to give out.  Think about that.  You'd have to give out ~536,000 1000 /32 netspace chunks of just the current in-use /3 to exhaust the entire /3.  Are there even that many ISPs, dataceneters, govt agencies in the world?  And each would get 1000 /32s, which have enough space for ~65,536,000 /48s to give to customers.  And oh yeah, after that there's seven more /3s to use.

And you'd have an even more fragmented network than we have today.  You have to allow some slack space to group things together.  You can't just assume 100% utilization.  That's crazy.  IPv4 only gets about 14%, and a longer address space will be much worse.  And no, as a matter of fact, 500k datacenters in the world would only be about one for every 30,000 people, in a world with 15 billion people.  That obviously wouldn't be enough (to say nothing of ISPs and government agencies).  If you'll just look at the numbers I quoted, or take a look at the exponential rate of internet increase around the world, you'll see that this is not an unrealistic concern.  And it's so easy to fix!

By the year 3000, we won't be using IPv6 anymore, since other factors (the "unknown unknowns" you speak of) will have likely made IPv6 obsolete.

I was trying to head off any such speculation.  My claim is that the addresses will run out much sooner than that.  The year 3000 thing was to point out that the /48 policy doesn't have any obvious purpose even if we do hopelessly try to think that far ahead.

Nothing you mentioned presents a good reason to allocate each of us a /48.  You're just trying to say it doesn't hurt anything.  But then why do it if there's no reason to?

[cholzhauer]

i have to agree with jim.  we need to move away from ipv4 thinking in the way we allocate ip addresses.

[broquea]

We didn't run out of /48s. We set a maximum number per tunnel-server, the more popular ones got larger allocations. We're looking at a different way of allocating master blocks to the tservs so more /48s can be made available, however there is no lack of available /48s out of our /32.

[xezlec]

i have to agree with jim.  we need to move away from ipv4 thinking in the way we allocate ip addresses.

I'm not sure what you mean by "ipv4 thinking" but I would very much like to know why we need to move away from allocating addresses in a responsible, common-sense way.  Please explain the benefit of allocating /48s instead of /56s.

We didn't run out of /48s. We set a maximum number per tunnel-server, the more popular ones got larger allocations. We're looking at a different way of allocating master blocks to the tservs so more /48s can be made available, however there is no lack of available /48s out of our /32.

I didn't mean to suggest that you ran out entirely.  But the fact that you need to look for a new way of allocating them to the servers seems unnecessary.  If you just hand out /56s, this problem won't exist, and there is no downside at all.  Isn't that easier than looking at a different way of allocating master blocks?  And it's much more responsible in the long run.  I have shown numerically that the current practice is ultimately unsustainable.

[broquea]

Policy is /48s for those that need multiple subnets.

[PigLover]

Zezlec, when you begin with a flawed argument you are likely to end up with wrong conclusions, to wit:

...A /48 for every user (which, in virtually every case, will be populated by a single device) means that essentially all of the addressing of the internet must be crammed into 48 bits.  That's only slightly longer than an IPv4 address...

A /48 is not "just slightly larger" than the current internet.  It is 16 bits larger - >65,000 times larger.  I don't know about you, but for me, that is not "slightly".  That's enough to allocate the entire internet address space separately to every country - no - every province of every country - and still have about 60,000 times more address space leftover than the total we have today.  Your concept of scale is completely off.

Furthermore, you are comparing network addressing in v6 to host addressing in v4.  Since the smallest non-subnetted network in v4 is a class C, a /48 is actually 24 bits larger, making the available space a few million times larger than what we had even with what you argue is 'inefficient' allocation.

/48s for every multi-LAN user are fine.

[jschweitzer]

xezlec - after reading this entire thread, my opinion is that you are trying to "realize" or "quanitfy" the IPv6 address space in a way that you can understand.  ive seen many people try to do this.

the address space of IPv4 is easy for the human brain to understand - 4.2x billion addresses.  theyre all allocated now.  But the IPv6 address space is simply not quantifiable by the human brain.  no one truely has any idea just how big 340 undecillion addresses is.

i think everyone agrees that addresses should be allocated in a thoughtful manner, but you dont fully realize just how many addresses there really are.  fighting over /48s or /56s just doesnt matter.  btw, where did you ever get the /56 idea?  ive never once seen anyone suggest a /56 before...

edit: ok, i just learned about /56s  ;)

[cholzhauer]

GoGo6 offers a /56 with their services.  IIRC, you get 256 /64's instead of 64k worth of /64s in a /48

[jimb]

GoGo6 offers a /56 with their services.  IIRC, you get 256 /64's instead of 64k worth of /64s in a /48

IIRC?

NUM_SUBNETS = 2^(LAN_PREFIX_LENGTH - PREFIX_LENGTH)
2^(64 - 56)  = 256
2^(64 - 48) = 65,536

;D

[cholzhauer]

If I recall correctly

Thanks for backing my answer up with proof :)