Re: [GLOBAL-V6] New draft available: IPv6 Address Allocation and Assignment Global Policy
Hi, On Wed, Feb 06, 2002 at 02:25:46PM -0800, CJ Wittbrodt wrote: > It is a figure that is "large enough that nearly every active LIR today > can get an IPv6 allocation NOW" (having somewhat over 3000 LIRs in RIPE > land, of which some are not active any more, others have merged, and so > on), while at the other hand being small enough so that *if* this > turns out to be a mistake, it means "6000 'IPv6 swamp' prefixes in the > global routing table", and this is something the routers can handle. > > I just want to be clear. My understanding of what happened in the > meeting wasn't that each existing LIR could get one, but anyone who > became an LIR could get one. This means that existing LIRs can > indeed get one, but so can anyone who is willing to become an LIR. > That was my understanding. Yes, this is what was agreed upon. Every LIR, no matter whether old or new, could get a /32 by asking for it. I do not think that this will lead to a "land rush" on /32s (to answer that concern en passant). Why should it? What's the benefit for an end site? It gives them a provider block that won't be filtered. If they're multihomed this is a benefit. Most of the companies that want to be "independent" will find one way or the other to achieve this - either by announcing /48s all over the region and possibly the world, or by opening a LIR, or by claiming they want to be soooo multihomed (and maybe setting up a peering with some other "independent" company to prove it). We have to be able to solve *this* - teach 'em that BGP multihoming with "PI" space is just one of many solutions, and develop more attractive solutions - instead of hindering IPv6 progress any further. Why would they justify or claim anything? They can just become an LIR with no justification and get a /32. ---CJ
Hi, On Wed, Feb 06, 2002 at 02:54:28PM -0800, CJ Wittbrodt wrote:
Most of the companies that want to be "independent" will find one way or the other to achieve this - either by announcing /48s all over the region and possibly the world, or by opening a LIR, or by claiming they want to be soooo multihomed (and maybe setting up a peering with some other "independent" company to prove it). We have to be able to solve *this* - teach 'em that BGP multihoming with "PI" space is just one of many solutions, and develop more attractive solutions - instead of hindering IPv6 progress any further.
Why would they justify or claim anything? They can just become an LIR with no justification and get a /32.
Yes. But then they need an upstream provider that is willing to announce a single-homed customer /32 - which might just be much more expensive than giving the end site a /48 of their space. Which I could imagine as a way to let market regulate whether someone "needs" their own /32. I am truly convinced that we can NOT solve the "I want to be independent" problem at registry level (or at least "not without doing real harm to some of the people we want to be 'in'"). Adding a rule like "you get a /32 only if you're multihomed" will make things worse, because "they" will then get a /32 *and* an AS number. Gert Doering -- NetMaster -- Total number of prefixes smaller than registry allocations: 71770 (72395) SpaceNet AG Mail: netmaster@Space.Net Joseph-Dollinger-Bogen 14 Tel : +49-89-32356-0 80807 Muenchen Fax : +49-89-32356-299
Gert Doering <gert@space.net> writes:
Yes. But then they need an upstream provider that is willing to announce a single-homed customer /32 - which might just be much more expensive than giving the end site a /48 of their space. Which I could imagine as a way to let market regulate whether someone "needs" their own /32.
Why? If it there was a significat effect of costs on prefix announcement, we wouldn't see the huge number of (pseudo-) PI prefixes in today's table. I don't see a technical difference between announcement of an IPv4 PI /24 and an IPv6 /32 which would imply a significant difference in costs. Thus you will get an upstream for your v6 /32 as easily as now do for your v4 /24. The proposal poses the interesting question of other conditions to becoming LIR apart from costs. Currently RIPE requires a minimum of imediately assigned address space. What if anyone wants to become LIR for IPv6 only? Was this discussed?
Adding a rule like "you get a /32 only if you're multihomed" will make things worse, because "they" will then get a /32 *and* an AS number.
Most of those who went to the trouble of becoming a LIR, be it only for the reason of getting "PI" space for IPv6, will surely get their own AS number too. Thus for size consideration you should assume one AS per prefix. BTW, don't get me wrong - I like this proposal much better than the current draft which IMHO is biased far too much towards conservation of address space and poses too high requirements to receive an allocation. Sorry to ask since I wasn't at the meeting: What was said about larger allocation sizes than /32, e.g. for multinational ISPs for which a /32 does not provide sufficient space for reasonable aggregation? Robert
On 7 Feb 2002, Robert Kiessling wrote:
Sorry to ask since I wasn't at the meeting: What was said about larger allocation sizes than /32, e.g. for multinational ISPs for which a /32 does not provide sufficient space for reasonable aggregation?
Or for a national provider who may want to give a static /48 to a few hundred thousand customers. How much of the address assignment maths includes static vs dynamic provisioning? tim
At 1:01 AM +0000 2/7/02, Tim Chown wrote:
How much of the address assignment maths includes static vs dynamic provisioning?
Tim, The HD Ratio analysis is based on the historical precedent of other addressing plans that have not used dynamic provisioning, at least in the sense I think you mean. Steve
Hi, On Thu, Feb 07, 2002 at 02:00:25AM +0000, Robert Kiessling wrote:
Gert Doering <gert@space.net> writes:
Yes. But then they need an upstream provider that is willing to announce a single-homed customer /32 - which might just be much more expensive than giving the end site a /48 of their space. Which I could imagine as a way to let market regulate whether someone "needs" their own /32.
Why?
It's a possibility to solve *that* issue. Yes, it might not happen, but different from IPv4, "renumbering is easy" (supposedly) and assignment of a /48 is going to be a lot less paperwork and administrative work than RIPE-141's. So providers might *want* to charge more for "if you bring your own /32, we have more work, so you pay more". Or upstream providers start charging per /32 that a downstream ISP announces, so there is an incentive for that ISP to bill it to their customers. It's just a possibility, but not an unlikely one.
If it there was a significat effect of costs on prefix announcement, we wouldn't see the huge number of (pseudo-) PI prefixes in today's table.
Today there isn't, which is part of the problem. Today there is no cost for deaggregating a /20 into /24s either, and this HAS to change. [..]
The proposal poses the interesting question of other conditions to becoming LIR apart from costs. Currently RIPE requires a minimum of imediately assigned address space. What if anyone wants to become LIR for IPv6 only? Was this discussed?
Actually RIPE requires the address space not for becoming a LIR, but for getting your initial allocation. If you don't want that, you can become a LIR just by signing that you have understood all the rules and are willing to pay your fees. [..]
Sorry to ask since I wasn't at the meeting: What was said about larger allocation sizes than /32, e.g. for multinational ISPs for which a /32 does not provide sufficient space for reasonable aggregation?
The draft states that those can get "whatever needed", but you have to argue it by listing customers, displaying network plans, and so on. As far as I remember, there was no opposition against this. Gert Doering -- NetMaster -- Total number of prefixes smaller than registry allocations: 71770 (72395) SpaceNet AG Mail: netmaster@Space.Net Joseph-Dollinger-Bogen 14 Tel : +49-89-32356-0 80807 Muenchen Fax : +49-89-32356-299
At 9:28 AM +0100 2/7/02, Gert Doering wrote:
different from IPv4, "renumbering [in IPv6] is easy" (supposedly)...
No, no, no, no, no. IPv6 has some features to make renumbering easiER, but it still may not be easy for some or even many sites, due to the many ways and places in which IP addresses are used (for most of which we do *not* have automated means of renumbering). Please do not adopt policies that assumes renumbering will be entirely painless for end customers, let alone for ISPs. Among the conflicting demands of conservation, aggregation, availability, etc. should be included "stability", where a "stable" address assignment is not a permanent- for-life assignment, but rather one that ought to be good for at least a couple of years in most cases, and one that ought not to be changed for less-than-compelling reasons. And on the general topic of discussion, I agree with Thomas. Unless and until we figure out how to do globally flat routing, it is vital to ensure aggregatability to a reasonable number of prefixes. I understand and appreciate the difficulty of defining "who's an ISP", and especially "who's a top-level ISP"; the RIR's "slow start" approach seems like the least bad scheme that anyone has yet been able to come up with, despite its flaws, so I'm surprised to see it discarded -- for IPv6 (if that's what's happening). I do wish we could converge on one of the possible multihoming solutions that would reduce the pressure for multihomed sites to seek PI prefixes... As for the research nets serving 50 universities, I don't think they should get /32s. If they were instead given longer prefixes and if, at some point in the future we end up with so many such longer prefixes that ISPs are forced to filter some of them out, the (presumably) relatively small number of especially "worthy" long-prefixes could be granted exceptions in the filter rules without too much operational effort (the real effort being the political one of identifying the worthies). The HD Ratio calculations that have been used to justify the current address field partitioning assume that the high-order /48 is being carefully managed, not squandered, so non-extravagance is still required at that level (i.e., something more generous than "conservation" but less generous than just giving huge blocks of /48s to anyone who asks). Steve
At 12:03 AM 2/7/2002 +0100, Gert Doering wrote: <snips>
Adding a rule like "you get a /32 only if you're multihomed" will make things worse, because "they" will then get a /32 *and* an AS number.
I know this is a bit out of the original subject but correct me if I'm wrong but there is only ~65000 in total where some are private ASN. We have plenty of IP addresses now for a while (20-30 years or so... ) What about AS numbers? How long will they last?
---
Roger Jorgensen (rjorgensen@chello.com) System Engineer @ engineering chello Broadband handles: ROJO1-6BONE ROJO9-RIPE RJC10-NORID
participants (1)
-
CJ Wittbrodt -
Gert Doering -
Robert Kiessling -
Roger Jorgensen -
Steve Deering -
Tim Chown