As I understand it, the primary issue with big-backbone-ISPs getting
big chunks and startup ISPs not getting seperate chunks is that the
small ISPs often grow big and want to do things like multihome.
I know that's what has happened with scruz.net, where Matthew's
speaking from, and several other ISPs I know of.
As I understand it, people are generally reluctant to break up
a large CIDR block to let some of it leave because it introduces
at least 2 new routes globally... you end up with 3 routes,
the block above the missing chunk, the block below the missing
chunk, and the missing chunk from the new backbone provider.
This is considered bad.
It is not, however, completely necessary.
Consider this as an option. Small ISPs signing up with their backbones
agree to a form of semi-portability of addresses. The small ISP agrees
that they will for (n years) maintain a direct IP link into backbone's
network which can be used for getting packets to the block allocated
from backbone's big CIDR blocks. This can be of arbitrarily slow
speed if need be, or contractually obligated to be a 56k or greater,
or T1 or whatever. Implimentation detail.
In exchange, backbone agrees to artificially inflate the reachability
cost in BGP advertisements to the large block and to allow small ISP
to have any other backbone advertise the specific block at an easier
to reach priority level. This allows the small ISP to dual home, or
switch primary home, and sort of take their address with them (in that
the new backbone can advertise the sub-block more attractively, thus
making traffic most likely to flow that way). It also allows the
origional backbone to not break up its block in its own announcements,
and thus minimizes the number of total announcements made globally
to no more than 2... one for the origional block, and a higher priority
one for the sub-block which migrated. This would be no more advertisements
than the case in which small ISP got its own space to start with and got
it routed, and is actually easier on small ISP at the beginning.
ISPs which have gone dual homed, connected to interchange points, etc.
are likely to be able to justify /18 and up allocations directly from
the NIC, and can slowly phase out their legacy blocks at that time.
They can at their option renumber their internal hosts into their
"own" space at that time, or not. They can let customers keep the
origional space they were allocated, as routing for it is guaranteed
one way or another. Or they can give customers the option to renumber
into space the now midsized ISP has better, more direct control over.
To avoid excessive headache, backbones should probably ONLY offer this
sort of deal to bona-fide ISPs (of any size), not end-user customers.
-george william herbert