Dual stack IPv6 for IPv4 depletion

Baldur Norddahl baldur.norddahl at gmail.com
Thu Jul 9 06:16:15 UTC 2015


Hi,

With RIPE you can get a /29 with no justification, so if you have any less
it is because you did not bother logging in to ripe.net and hit the get
more button.

ARIN gives you the option to make a network scheme based on nibbles but
RIPE does not, so do not go there. Why try to allocate by the bit at all?
We all use internal routing protocols instead of static routing. The only
concern is to avoid an excess amount of internal routes in your IGP. You do
that by using an allocation size per site, so your local routers can do
route aggregation before redistributing the routes.

We have an allocation size of /42 per site. A site can have and usually
have multiple /42 allocated. This size was chosen because we allocate a /26
of IPv4 per site (=6 bits per allocation or blocks of 64). We are a
residual ISP and it is expected that each customer needs one /32 IPv4 and
one /48 IPv6 prefix. Our allocation of /32 IPv4 and /48 IPv6 is
approximately the same utilization. In fact we made an algorithm that can
convert your IPv4 /32 to your IPv6 /48, so we avoid tracking two
allocations per user.

Our smallest access routers can handle about 30k routes. But long before we
hit that, I expect we would add another layer of aggregation.

The case of using a bit based allocation scheme is so you can avoid that
database (or spreadsheet) keeping track of your allocations. But I found
that you need that either way or you will go crazy. The ability to look at
an allocation and say hmm digit 10 is between 8 and f, so that must be
somewhere in [insert big city here]... well it does just not seem that
useful. Check our reverse DNS if you want to know.

Regards,

Baldur



More information about the NANOG mailing list