Re: [sig-policy] prop-111-v001: Request-based expansion of IPv6 default
Hi David,
I should answer these questions.
From: David Conrad <drc at virtualized dot org>
Subject: Re: [sig-policy] prop-111-v001: Request-based expansion of IPv6 default allocation size
Date: Mon, 27 Jan 2014 22:24:22 -0800
| Hi Guangliang,
|
| On Jan 27, 2014, at 9:20 PM, Guangliang Pan <gpan at apnic dot net> wrote:
| > I think that statement refers to early IPv6 allocations from the old /23 blocks. Before APNIC received the /12 allocation from IANA, we use sequence allocation method to make /32 allocations and reserved up to /29 for every allocation. That was the practice for all RIRs in the early stage. I believe this policy proposal is trying to address those reserved space.
|
| That wasn't clear from my reading of the proposal, e.g., the first sentence of the problem statement says: "... while APNIC currently reserves up to /29 for each /32 allocation."
I'm sorry for lacking my explation. I'll revise my proposal text about
this point.
| > Current /32 allocations from the /12 block can grow up to /24 at this stage.
|
| Err. That suggests APNIC is not using sparse (aka bisection) allocation, rather APNIC now just reserves the the /24 instead of the /29...?
I think Guangliang mentioned just about current allocation status with
sparse allocation mechanizm described in:
http://www.apnic.net/publications/media-library/documents/resource-guidelines/ipv6-guidelines#sparse
| In any event, looking at the proposal, I gather there are 3 justifications are provided for a default /29:
|
| 1) traffic engineering, since some folks out there filter on /32 boundaries;
|
| 2) potentially fewer prefixes if the ISP needs to expand
|
| 3) efficiency
|
| Going in reverse order:
|
| I don't understand #3. In this context, is the "efficiency" mentioned related to ease of network design?
Here I would like to say reserved space which will not use into the
future (early IPv6 allocations) can be utilized with this
modification.
| WRT #2, if I understand correctly, all APNIC allocations for ISPs have sufficient space for that allocation to grow to (at least) a /29 within a single prefix. I don't think APNIC staff would be so silly as to allocate from the reserved/extension space non-contiguously.
I did not suppose such non-contiguous case, but there might be a
case that LIRs who get additional continuous space announce two /32s
not one /31 in order to keep existing network stable.
| That leaves #1 which appears to assume the same folks who are currently filtering on longer than /32 won't decide to start filtering on longer than /29. After all, I can easily see the whole point of filtering on /32 by (arguably) overly pedantic network operators as trying to discourage folks from shattering their allocations for traffic engineering purposes to try to limit routing table growth. Once APNIC makes /29 the default allocation size, they could just as easily shorten their prefix filters. Then what, make the default a /24?
|
| Are there any data on how many ISPs are filtering at /32?
I do not have that information, but I observed that I can find some
/35s only in part of looking glasses (and prefixes longer than /32 are
increasing).
Yours Sincerely,
--
Tomohiro Fujisaki