[AFRINIC-rpd] new proposal: "Remove requirement to announce entire v6 block as single aggregate"

[Owen DeLong]

I generally support the idea behind this proposal.

However, I do think there is a serious flaw in this proposal in that it creates the impression that it might be appropriate for an ISP to advertise /48s. In no case should an ISP allocation to a region be so small as to prevent the assignment of a /48 or more to each and every end-site supported by the ISP. (Where end-site means any customer structure, building, or tenant in a multi-tenant structure or building).

I also think we're getting wrapped around the axle on several things that are not so important as they are made out to be. (More details near the bottom).

I would suggest rewording the proposal as follows:

1.	Problem
	The current AfriNIC policy requires the announcement of a single aggregate
	of IPv6 space and, thus, requires any organization needing to make multiple
	announcements for traffic engineering or other localization purposes to actually
	apply separately for each block to be advertised. This is costly and time consuming
	for both the organization in question and for AfriNIC staff.

2.	How the proposal addresses the problem
	This proposal would remove the single-aggregate requirement allowing an
	organization to obtain a larger aggregate IPv6 allocation which could be
	divided and announced in whatever manner best meets the needs of the
	organization.

3.	Proposal
	We propose to delete the following sentence in section 6.1.1 (d)
	from the IPv6 policy “The LIR should also plan to announce the
	allocation as a single aggregated block in the inter-domain routing
	system within twelve months.”

	Add the following to section 6.1.2

	An LIR organization can justify larger space than a single /32 on the basis that
	they will be dividing their allocation up into regions (where the definition of
	a region is left to the organization) and wish to make more specific advertisements
	from each region. In such a case, justification is such that each region can be
	given enough space to provide /48s to each end-site served in the largest
	region with reasonable room for growth.


On May 14, 2013, at 23:39 , Andrew Alston <alston.networks at gmail.com> wrote:

> Hi RPD List,
> 
> While I see the logic behind this proposal, I have to oppose it based on technical grounds.
> 
> If said multi-national applies for a large enough block to cover each country and is allocated space equivalent to a single /32 per country (so in this case, a /28, or 16 /32s), then this proposal would work since the multi-national could announce /32s from each location.  However, 
> allowing for announcements of /48s out of the LIR blocks would not work because of filtering.  There are many that still only accept /32s, other than when the /48 is assigned out of a known PI block.

I must disagree...

1.	Hardly anyone is still filtering out /48s and they have become the de facto accepted routing increment
	for IPv6. We see far too many of these more specific blocks to believe that 

2.	RIR policy should not be driven by router policy, nor should routing policy be specified in RIR policy.
	The responsibility of RIRs is to responsibly distribute number resources. Routing is the responsibility
	of those people who actually operate routers and it is up to each of them to decide which prefixes they
	will or will not accept, advertise, etc. It is only coincidence in most cases that these entities are the same ones
	that interact with the RIR to begin with, so RIR rules on the subject are largely meaningless anyway.

> To allow this would only result in partial reachability for the block and I'm not sure that's a good idea at all.

This is simply factually wrong in addition to being out of scope of RIR policy. First, the organization can always announce the aggregate along side the more specifics resulting in complete reachability if possible. Second, the more specifics are unlikely to get filtered. Finally, the people who operate routers are perfectly capable of addressing routing discrepancies without involving the RIRs in the process and there is no benefit to RIR policy being built to protect network operators from themselves.

> At the same time, huge allocations of /28s or whatever may not be practical though I think that’s another debate entirely.

Why not? In the ARIN region, you can get a /28 practically at the drop of a hat. Large allocations are very practical in IPv6 and we should expect and support them widely. Small allocations are actually potentially harmful.

So far, I have been involved in obtaining a number of /28s and even a couple of /24s for organizations through the ARIN process.

There are 16.7 Million /24s in IPv6 of which, more than 10 million are still completely untouched. Unless you envision a time when the world has more than 10 Million relatively large ISPs, I see no problem with liberal distribution of /28s (of which there are more than 160 million completely untouched).

> I do acknowledge the problem the proposer has, and I do believe the economic problem stated needs to be rectified, I just don't agree with the proposed solution.

I think the proposed solution is perfectly fine. I do not think that the RIRs should be the routing police. As such, the current requirement seems anachronistic at best, IMHO.

As to Sunday's comment... Most RIRs have policy which allows space to be assigned to private or non-connected networks. I see no reason to believe that obtaining space which is not subsequently announced is a bad thing.


Owen