[ppml] Arguments against Policy Proposal: IPv4 Soft Landing

[Iljitsch van Beijnum]

On 22-mei-2007, at 23:25, David Conrad wrote:

>> 2) This creates an increase risk of a 'run to the bank' before
>>    the current set of policies change

> The IPv4 free pool is nearing exhaustion. Given human nature, a "run
> on the bank" will almost certainly occur

So what would that look like?

The way I see it, the only people who can successfully request really  
big blocks of address space, are people who already hold really big  
blocks of address space, which would be the world's largest ISPs.  
Anything below a /16 is not worth our time, that's only 15% of the  
yearly address use. It is, howerver, 91% of the allocations.  
Reclaiming one or two /8s a year is enough to fulfill the < /16  
requests so for people needing less than a /16, IPv4 will effectively  
never run out. It will for the large ISPs, but does it really matter  
much in the grand scheme of things whether those play nice or start  
padding their requests? At some point, they'll have more customers  
than addresses, and that's the moment of truth. The only people for  
which all of this really makes a difference is new big address users:  
those don't have the track record to prove to the RIRs that they  
really need the addresses, and they don't have large existing address  
blocks to spread a little thinner over the increasing numbers of  
customers.

> The point of increased requirements such as the 3rd party audit is to
> reduce the likelihood of people simply making up numbers to get
> allocations.

Did you have any request size in mind for this? An audit for a /9  
request makes some sense. For a /24 request it's complete overkill.

> However, folks at router vendors who can reasonably be assumed to
> know what they are talking about have indicated that router
> scalability isn't a concern for 5 to 10 years.

Not exactly. There are already significant power, cooling and cost  
issues. Will there be a moment when vendors are going to tell us that  
there is no way they can build a bigger router? Not likely. But the  
number of people that can afford the routers they do manage to build  
will decrease as the routing table grows, as has happened over the  
past decade. See the results from the IAB routing and addressing  
workshop last year.

>> 4) Efficiency above 80%.

>>    For an organization that does allocation from fairly large blocks
>>    all the way up to end devices , 80% efficiency is already a very
>>    high tartget to achieve and it is not feasable to go much above.
>>    We may shoot for 81 or 82% with great difficulties, but 90% and
>>    above is irrealistic.

> I would be interested in understanding this point more, perhaps
> privately.  I do not have enough data to fully appreciate your
> concerns here.  However, the IPv4 free pool is nearing exhaustion so
> it is unlikely that the processes and procedures used when IPv4 was
> plentiful will continue to be appropriate.

If you have 50 hosts in one subnet you can have a /26 (64 addresses)  
and have about 80% efficiency. But if you have 500 hosts in 10  
subnets, you'll need a /22 (1024 addresses), which allows for 16 /26  
subnets even though you only need 10. So at the organization->subnet  
level you are 10/16 = 62.5% efficient and then 50/64 = 78.125% at the  
subnet level but your total efficiency is only 500/1024 = 48.8%.

In other words, the maximum reasonably reachable efficiency is  
reduced with each level of hiearchy in your addressing model. For  
really large networks with a good number of hierarchical levels this  
adds up quickly. See RFC 3194 that Alain co-wrote.

>> Given all the above points, I'm not in favor of this policy.

> Understood.  However, given the IPv4 free pool is nearing exhaustion,
> I would be interested in understanding what your view of the
> alternatives are.

I can't speak for Alain, but I am also against this policy proposal.

My ideal scenario is one where nothing changes, except for a possible  
gradual increase in the yearly number of address use, so we don't  
have to keep coming up with new predictions that people are going to  
ignore if they don't like them. Predictability is key.