[arin-discuss] IPv6 End User Assignments

[Aaron Hughes]

Matthew,

You are highly likely correct.  Regional aggregation will be implemented by any reasonable ISP. Also, if ISPs did assign /48s to each customer your math is correct, each ISP of this massive size would need a /21ish.  Also, I agree a /48 is excessive for each customer.  

For a moment we should separate policy from operations.

As architects in the planning phase of a v6 roll-out we get to plan for company needs, customer needs, aggregation, reachability, scalability, etc etc etc. One of the aspects we should all evaluate is wasting space.  

When I rolled out v6 to my customers, I made the company policy decision to assign /64s to customers by default and /48s to those who requested more than one subnet.  This made it rather easy to have 2 pools of aggregatable regional space for customer assignments.  The company policy / network architecture fits within certain, what I will call, ARIN guidelines in that it does not violate ARIN policy. Policy has never dictated operational practice.

That being said, policy should _allow_ for an implementation that requires /48 assignments to the LIRs respective customers. There are implementations where this makes sense.  It is highly unlikely that very large ISPs will be assigning 48s to each customer as it would be a waste of space.

This list, just like Planing / Design / Architecture / Operations groups are full of opinions about how things should be implemented. Time will reveal best practices and policies will get updated to reflect them as long as the operational community is involved in the policy process.

I am involved in many v6 implementations and none of them assign 48s by default.

Cheers,
Aaron


On Wed, May 06, 2009 at 11:28:47AM -0600, Matthew Wilder wrote:
> This is where it gets interesting.  I doubt the worst case is a /23.  Remember, IPv6 has so many bits for  the very purpose of clean summarization and easy subnetting.  
> 
> Comcast might want to regionalize their subnetting.  And then within each region, they might want to have a nice big block for each edge router so they don't have to constantly add address resources to the router.  All of a sudden, instead of assuming a 90% utilization of that block (which is heinously unreasonable and inconsistent with IPv6 intentions) you are looking at maybe 20 - 30% utilization at the /48 assignments.  Now they need probably a /21 for those customers.
> 
> This gets this sort of ISP into the hairy edge of what the HD ratio allows in the best case.  Assuming a /48 assignment to an end user counts as 100% utilization of the entire /48 subnet, then they will probably squeak through on the Threshold (https://www.arin.net/policy/nrpm.html#six7).
> 
> And this discussion here is exactly why I originally through out the question.  Why would an ISP assign a /48 so that a consumer can have two large layers of subnetting (16 bits of subnet address to be exact) at the expense of their own routing and summarization?
> 
> MW
> 
> 
> Aaron Hughes wrote:
> > US population is roughly 300 million.
> > A /19 would cover 536,870,912 /48s
> > A /27 would cover 536,870,912 /56s
> > 
> > 7 billion in the world.
> > A /15 would cover 8,589,934,592 /48s
> > A /23 would cover 8,589,934,592 /56s.
> > 
> > Number of total Internet users in the world roughly 1.5 billion or 20% of 
> > the population.
> > Number of total Internet users in the US roughly 220 million.
> > 
> > Let's say you are Comcast.. ~ 25 million customers. Worst cast you are 
> > looking at a /23 to give each one a /48, or roughly best case a /39 for 2x/64s 
> > per customer.
> > 
> > This is not a repeat of v4.
> > 
> > IPv4 ISPs gave a single host to the outside interface of the CPE AND some 
> > flavor of space in (RFC1918) 10.0.0.0/8, 172.16.0.0/12, 192.168.0.0/16 for 
> > their inside interface.  If we implement NAT in v6, we will stop progress 
> > with end-to-end application development and make the same silly mistakes we 
> > made with v4.  The mistake was not wasting space but rather not making the 
> > leap to IPv6 when we identified the potential for growth so many years ago.  
> > Instead we focused on CIDR/VLSM and NATing everything we could to extend the 
> > life of a dying protocol.
> > 
> > It is perfectly reasonable to have standard assignment sizes to create an 
> > appropriate customer expectation. Your customers do not need to know what 
> > a subnet is.  If the standard was, for example, to assign a /64 to the WAN 
> > and /64 to the LAN with SLAAC enabled, the customer behaves the same way 
> > they do today.  Those who request more space know what they are doing 
> > (generally speaking).
> > 
> > Cheers,
> > Aaron
-- 

Aaron Hughes 
aaronh at bind.com
(703) 244-0427
Key fingerprint = AD 67 37 60 7D 73 C5 B7 33 18 3F 36 C3 1C C6 B8
http://www.bind.com/