[afnog] New IPv6 Address Block Allocated to the RIPE NCC

[Brian Candler]

On Tue, Jan 03, 2006 at 11:26:20AM +1100, Geoff Huston wrote:
> "Or assuming an 80% utilisation rate, it means they are planning to connect 
> 429 million customers with a /48 each over the next two years."
> 
> Of course IPv6 does not use a flat 80% address efficiency metric.
> 
> Under the current 0.8 HD ratio metric and under the current /48 end site 
> allocation point, the ISP needs to have some evidence of a proposed rollout 
> across some 5.6M - 9/6M end sites.

Thanks for the URL poitners. Well for one thing I have misunderstood "80%
utilisation" versus "0.8 HD-Ratio". However it's still not clear to me why a
/19 should be needed for "5.6-9.6M end sites"

Of course, the utilisation *within* a customer /48 is vanishingly small, so
that has to be ignored. But even then, a /19 for 10M /48 end sites implies a
utilisation of 1.86%

Thinking from the point of view of network administrator, I can see a few
approaches to allocation which might be considered:

a. You stick a /48 route in your IGP for each end customer. In this case
utilisation can approach 100%.

b. You allocate a suitable sized block for each POP (scaled for a
utilisation of between 25% and 50% after two years), and announce them into
your IGP [*]

c. Half way house: you pick a block suitable for your smallest POPs, say /36,
and announce (not necessarily aggregated) multiples of that for the larger
POPs

Now I'd be the first to admit that 10M routes in your IGP is not going to be
a happy situation, so I'll discount (a).

In case (b) the utilisation of each block is never worse than 25%, but you
have the problem of dividing your address space into multiple block sizes
for the different sized POPs to maximise aggregation, and that process also
gives wastage. But in the absence of incompetence in address allocation it's
hard to see utilisation going under 10%, which requires a /21.

I quite like option (c). That gives you 4096 customers per block. For a POP
with less than 4096 customers, there will be proportional wastage; for a POP
with more than that, the wastage will never be more than 4096 (i.e. all
blocks apart from the last will be at 100% utilisation)

Utilisation for big POPs is therefore better than for small POPs. Taking the
worst case of everything being a small POP, and assuming 2000 customers per
POP, your 10M customers will connect to 5,000 POPs and use no more than
5,000 routes in your IGP, and you will need a /23 (giving 8192 /36's, so
plenty of growing room there too)

So I'd still be interested to know the details of an addressing plan which
requires a /19 for this.

Of course, this demonstrates that IPv6 is just as much as a PITA for address
allocation as IPv4, despite the supposedly huge amount of address space
available :-(

Regards,

Brian.

[*] I don't see any reason to preallocate more space than this to a POP.
It's easy enough to add a second or third block to a POP later when needed.