cgrundemann at gmail
Aug 1, 2012, 2:24 PM
Post #9 of 44
On Wed, Aug 1, 2012 at 3:03 PM, Tim Densmore
Re: Current Consensus on IPv6 Customer Allocation Size
[In reply to]
<tdensmore [at] tarpit> wrote:
> Hi Chris,
> Thanks for the link! I had actually read through it recently and found it
> quite useful. I find the back and forth between /64 vs /12 on P2P links
> I've read rfc 3177, and though some of it was above my head, I couldn't
> really find anything in it that really defined *why* a /48 was preferred
> (GSE, 6to4, and Site Local require /48, so /48 is best was what I took
> away). As far as I can tell, the layman arguments in favor go like this:
> - /48 is likely to be equal to /24 (v4) bgp-wise, and anything longer is
> likely to be filtered.
> - /48 for *everyone* allows for uniform customer allocation size.
> - It's "easy" to forsee that someday in the future people will need more
> than 256 subnets in the home, and since nibble boundaries are considered a
> must, then /48 is the only option.
Maybe. What is more likely (imo) is that we will see more and more
complex self-configuring home networks. The /48 gives you 16 bits to
play with which makes algorithmic subnetting without human
intervention much more possible. I don't know that I'd say we will
definitely need >256 subnets, rather that we need the flexibility to
"plug and play" routers in a multi-tier home network. If we break a
home network down by "width" and "depth" it looks something like this:
1 CER [/48] (Customer Edge Router)
15 L1IRs [/52] (Level 1 Internal Routers)
15 L2IRs [/56]
15 L3IRs [/60]
15 L4IRs [/64]
If you use 16 bits for internal subnetting (a /48) you get upto 15
routers (wide) at every level and upto 4 levels of routers (deep)
within the network. You also know that this is the case
deterministically. This makes prefix autoconfiguration easy (if I get
a /56 I'm a L2IR and I should hand out /60s downstream).
You may not go to full width at each level, nor to full depth in each
branch, but having those 16 bits gives you a stable framework to
operate in. Yes, you can get away with less - but why? The future of
innovation is brighter the more flexibility we provide today, and IPv6
is abundant (not scarce like IPv4).
> Is that about right?
> I think I'd have a hard time pitching extra cash outlay for a /28.
> Thanks again!
> On 8/1/2012 1:49 PM, Chris Grundemann wrote:
>> Best Current Operational Practice (BCOP) on IPv6 Subnetting:
>> /48 per site is best. I would highly recommend swallowing the ~$2k/yr
>> and get the allocation you need now, so that your network can grow in
>> a structured, homogenous manner. Rather than fighting fires later to
>> save a buck now (I mean, I have to guess that buying even one router a
>> year blows that cost out of the water anyway - even a line card...).