[Chapter-delegates] What to do about IPv4 address scarcity

[John Schnizlein]

Your micro-question gets at the heart of the problem.

Some partial answers:

For some time, the normal practice has been that you get address space  
from your Internet access provider so that details of routing to your  
site is contained within your ISP's network rather than requiring an  
entry in every routers global route table.  You would have to renumber  
if you change providers.

The addresses allocated prior to the development of RIRs (regional  
internet registries) and especially the Provider-based addresses are  
called "legacy".  The provider-independent (PI) legacy address  
prefixes are the subject of much debate, at least partly because they  
were allocated under different rules and expectations for the Internet  
(before the crisis of success).

The growth of the global route tables was (temporarily) slowed by the  
adoption of the practice of a site getting its address from its  
Internet access provider, which got addresses from the RIR.  The  
aggregation of routes to a (shorter) prefix that was enabled by  
matching the hierarchy of address allocation to the topology of the  
Internet helped keep everybody's tables from growing too quickly.

Many ISPs will provide routing for the legacy address prefixes.  Their  
view is usually that it would be silly to turn down business, and the  
impact of another entry in the global route table is shared by all  
their competitors as well as them.

The real question is how quickly and how much would the global route  
table grow if new networks were to connect to Internet access  
providers with addresses from somewhere else.  When your ISP doesn't  
have an address block large enough to satisfy you, if the accept an  
address prefix for you, that adds to the global route table in a way  
that aggregation does not cover.  How big a problem is this likely to  
really be?  Is it comparable to the growth of the global route table  
from traffic engineering and multi-homing that has already occurred?

All the proposals I have seen limit transfers to within the same RIR.   
How much worse is the impact on the global route table if address  
transfers between RIRs were allowed?

On 2008Sep8, at 1:50 PM, Sivasubramanian Muthusamy wrote:
>
> One reason that a market for IPv4 addresses might need to be managed
> is to preserve the constraint on growth of (default free) route tables
> produced by provider-allocated addresses, which has been essential to
> protecting the global routing infrastructure.  If address prefixes
> from one region are advertised on another continent, what is the
> impact on Internet route tables?  Please send pointers to any
> research, especially numerical estimates, on this.
>
> I have a more micro-question here. At present, if I have an internet  
> connection from ISP ABC, I need to get my address allocated by this  
> ISP. What if get my address space from ISP XYZ but choose to buy the  
> connection from ABC? Or what if I buy my address space direct from  
> ICANN and go to ISP ABC for an internet connection for hosting  
> bandwidth? Is this kind of "portability" possible and if possible is  
> portability conceded to users at present?

We should be careful about the use of "buy address space".  As things  
stand now, nobody buys addresses.  Rather they are allocated based on  
need to RIRs and then to ISPs and then to customers.  Each layer in  
the allocation has a vested interest in keeping the complexity of the  
global route table from growing too quickly.

John