routed /48

[awisskirchen]

Hello,

I like to test a routed /48-Network, but I can't allocate them.

This is the message that I see: Failed: Maxmimum # of 48s for this tserv allocated

Where is the problem ?

Thank you.

Regards
  Andreas

[cholzhauer]

Maxmimum # of 48s for this tserv allocated

I would guess that the server can't allocate any more /48's. 

I think you should email ipv6@he.net and alert them to the problem you're having

[awisskirchen]

mhh okay, but I tested with different server.

[broquea]

That tunnel-server is out of available /48 allocations. We will be adding more in the future, however there is no ETA right now.  You can get a /48 for testing purposes by choosing a different tunnel server.

What did you want to "test" a /48 for? The only real difference between it and the routed /64 we already provide, is additional /64 subnets for different LAN segments. If you aren't using auto-configuration then you can use smaller blocks with DHCPv6 or even just static addressing.

[abelbeck]

That tunnel-server is out of available /48 allocations. We will be adding more in the future, however there is no ETA right now.  You can get a /48 for testing purposes by choosing a different tunnel server.

I too would like to use autoconfig with multiple subnets, DHCPv6 is not supported with OS X yet.

Any update on the ETA for more /48's at Dallas? How about /56's is that a possibility?

Thank you for this excellent service.
Lonnie

[SimRett]

I'd also happily take a /48 at Frankfurt...

[hisken]

Funny to see how IPv6 address space can run out  ;D

But okay. Bit disappointing HE is running out of /48s. I have plenty of useful purposes for tunnel /48s, not just for experimental usage. For example: routing between two locations, and subnetting/dividing the upstream subnet I get from HE. We all know the RFCs describe a /64 as the absolute minimum at the customer premises (but often you can use smaller IP6 subnets-but RAs won't work with such subnets), and if a tunnel user needs to split it up for routing purposes it will get tricky if he only has a /64.

With lots of ISPs around that aren't yet IPv6-capable and a dissapointing amount of tunnel brokers (that crappy Freenet, HE and SixXS with their stupid admins) we need some good tunnelbrokers which can provide usable and complete IPv6 services to fill the gaps while ISPs aren't ready for IPv6 yet. And that includes /48s (or even better: /56s, a lot less waste of address space: a user really doesn't need 65535 /64s isn't it?).

IMHO, HE should take a look at implementing /56s instead of those large /48s. For example: make it possible to request a /56. And furthermore, you could add a form where users can apply for a /48 but only if they have a very good reason for it. @broquea: what about such an idea?

[SimRett]

Agreed, /56 would be fine; heck, even /60 should suffice for me. In my specific case I have two subnets to serve with ~10 users and would like to use RAs, since some OSes don't support DHCPv6 yet.

[marcusw]

If you don't mind me asking, why would one even want more than one /64 subnet on a single LAN? Don't subnets become irrelevant now that there is global routing for everything? Wouldn't having subnets increase the number of hops and therefore latency between nodes?

Pardon my ignorance.

[jimb]

No.  Subnetting is still required.

You need the /48 when you need more than one LAN, otherwise a /64 suffices.

There's nothing magic about IPv6 which makes sizing LANs much different than under IPv4.  Just easier since there are pretty much fixed LAN sizes now and you'll never have to worry about running out of host numbers etc or networks.  However, you still don't want more than so many nodes on a single LAN for the same reasons you don't under IPv4.

I have started to see another trend towards "flattening" networks again ... similar to one I saw back in the 90s.  But as I said, having too many nodes on a single LAN has similar pitfalls under v6 as it did under v4.  Too large of a "multicast domain" is just as bad as too large of a broadcast domain.

Also, as for /48 vs. /56 etc, etc, this ground has been covered many times.  Basically, /48s are what the IANA recommends to assign to end users with more than one LAN, or businesses, per site.  It might seem wasteful, but the IPv6 space is so huge that it's not a concern.  Having large assignments like that also reduces the size of routing tables and makes routing faster/less expensive (less expensive TCAM needed, etc).

[hisken]

Also, as for /48 vs. /56 etc, etc, this ground has been covered many times.  Basically, /48s are what the IANA recommends to assign to end users with more than one LAN, or businesses, per site.  It might seem wasteful, but the IPv6 space is so huge that it's not a concern.  Having large assignments like that also reduces the size of routing tables and makes routing faster/less expensive (less expensive TCAM needed, etc).

Interesting matter. But is there really any performance hit when a ISP for example has like 30.000 /48s versus 30.000 /56s in their BGP? Obviously the number of subnets allocated from the /32 would be the same in both situations. Correct me if I'm wrong. Even when very big ISPs would be using /56s they would only need one /32 in most situations. Of course, the routing table of that /32 would be a lot heavier compared to multiple /32s with /48s in it. Is that the kind of performance hit you are referring to?

[marcusw]

No.  Subnetting is still required.

You need the /48 when you need more than one LAN, otherwise a /64 suffices.

There's nothing magic about IPv6 which makes sizing LANs much different than under IPv4.  Just easier since there are pretty much fixed LAN sizes now and you'll never have to worry about running out of host numbers etc or networks.  However, you still don't want more than so many nodes on a single LAN for the same reasons you don't under IPv4.

So I would want a /48 if I had, for example, more than about 500 hosts on my LAN? Or are there other reasons? It can't be organisation, since things are the same connectivity wise whether there's a router in the middle or not. Thanks for explaining this to me.

[hisken]

So I would want a /48 if I had, for example, more than about 500 hosts on my LAN? Or are there other reasons? It can't be organisation, since things are the same connectivity wise whether there's a router in the middle or not. Thanks for explaining this to me.

Yeah, that could be one reason though those 500 hosts can obviously perfectly fit in a /64.

You'd want a /48 or /56 for subnetting things up. For example like when you are setting up a VPN server at home, you could use a second /64 for it and make it routable. If you didn't have a /48 or /56 in that case at your home, your only options would be splitting a /64 (not recommended since minimum "official" size is a /64) or a Layer-2 VPN bridge (much more overhead, also not recommended).

[jimb]

Heh yeh I thought about the routing table size thing to after I posted that, and it probably wouldn't make a big difference in routing table size in the end.

[xezlec]

Also, as for /48 vs. /56 etc, etc, this ground has been covered many times.  Basically, /48s are what the IANA recommends to assign to end users with more than one LAN, or businesses, per site.  It might seem wasteful, but the IPv6 space is so huge that it's not a concern.

It has been covered many times, and the obviously wrong answer reached and stubbornly defended every time.  A /56 is ludicrous overkill.  A /48 is just throwing bits away for no reason at all.  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 middle-ground projection of future world population is 15 billion.  Dividing by the current population of internet-enabled users, we see that growth by a factor of 32 (5 bits) is reasonable to expect.  Allocating larger prefixes higher up in the network hierarchy genuinely might help with routing (though I don't see how allocating extra bits at the level of a customer's home network does that), so we can expect to lose, very conservatively, another factor of 64 (6 bits) for that.  Internet growth is exponential even in regions where most of the population already has internet access (due to the proliferation of new ways and places that an individual accesses the internet throughout his day, i.e. all of the new mobile devices now coming on the market).  We can expect at least a factor of 16 (4 bits) from that.  There's a factor of 8 taken out by the IANA because of the reservation of most of the /3's.  Putting it all together:

+ 32 bits currently occupied by the world
+ 5 bits population growth and internet uptake
+ 6 bits elbow room for routing
+ 4 bits device proliferation
+ 3 bits reserved by IANA
----
50 bits needed

And yet we're giving each residential home user a /48?  And you don't think it's a concern?  And this only takes into account likely, known scenarios.  What about the unknown unknowns?  Shouldn't we plan for that?  What if we want to assign IP address space to other astronomical bodies or something (sure, there's lag, but that's not a huge problem for streaming multimedia).

Having large assignments like that also reduces the size of routing tables and makes routing faster/less expensive (less expensive TCAM needed, etc).

Why does allocating a bigger prefix make routing more efficient even when the extra bits will obviously not be used by the customer?  Name a case where an ordinary user will need thousands of subnets.  Even a futuristic sci-fi case.  I can't think of one.

So I would want a /48 if I had, for example, more than about 500 hosts on my LAN?

Nope.  One subnet is always a /64.  You would need a /48 if you had about 500 subnets, each with multiple computers and a router of its own.  If you only have a hundred or so subnets, not thousands, a /56 is fine.  In reality, no one has more than one subnet at home, so a /64 is fine.  If some weirdo really wants multiple subnets at home, they should request special treatment from their provider and expect to pay more.  /56 by default is silly and /48 is indefensible.

In the year 3000, there will not be one single IP address in the world with any nonzero bits between the 48th and 56th bit (I invite you to try and argue otherwise), but the address space above that will be crammed and crowded as all heck.  There's something wrong with that.