May 2002 - ipv6-wg - mailman.ripe.net

Re: Getting addresses
by Joao Luis Silva Damas 16 May '02

16 May '02

At 17:27 +0000 13/5/02, Robert Blechinger wrote: >Hi Joao, > >On Mon, 13 May 2002 17:08:17 +0200 >"Joao Luis Silva Damas" <joao(a)ripe.net> wrote: > >> It is does not appear realistic to expect ISPs to "interview" each >> customer and ask them whether they intend to subnet or not. > >so, you think it's not realistic to ask the customer if he intends >to subnet or not ? > >ipv4 view: why not to give all your dialup customers in ipv4 >per default a /28 delegation, and the customer decides to do NAT, >all in one subnet, or more subnets ? > >no, i would give everyone /64 per default, unless the customer >says, "no, i want to do subneting", then i will give him a /48 >without discussion. > >thats my point of view. I don't have a point of view. A lot of people are telling me they will choose the /48 road. Without a detailed definition of site, almost everyone's view will have to be accepted. That's what I am seeing in all these discussions. > > Everyone will probably end up with a /48: it reduces costs for the ISP. > >ipv4 view: some years ago,....: Everyone will probably end up with a /24 >hm...i think the cost point isn't a good argument, sorry. >does it cost so much to do subnetting ? you have one block with /64 >delegations >and one with /48's. >so where is the problem ? take a /64 give it to the customers, if he says no, >thake the other one. The problem is the lack of arguments for an RIR on what is the right thing to do. Joao

4 4

RE: (ipv6-wg) Roadmap to IPv6 multihoming: no PI addresses
by Michel Py 16 May '02

16 May '02

> Francis Dupont wrote: > => MHAP is a two space solution (PAs are locators, PIs are > identities) These locators are provider-independent, but should really _not_ be called PI for two reasons: 1. They do _not_ appear in the global routing table. 2. Most of them are geographically aggregated. > and draft-ietf-ipngwg-esd-analysis-05.txt should be read in order to > understand pros & cons (note: this draft is expired and can be hard > to find). http://arneill-py.sacramento.ca.us/ipv6mh/draft-ietf-ipngwg-esd-analysis -05.txt Note that this draft is not completely relevant to MHAP. MHAP does have strong similarities with GSE, but is a different protocol as well. Michel.

1 0

(ipv6-wg) Roadmap to IPv6 multihoming: no PI addresses
by Michel Py 16 May '02

16 May '02

Ipv6-wg folk, This is my candid view on IPv6 multihoming, comments welcomed. Enjoy, Michel. +-----------------------------------------------+ | Roadmap to IPv6 multihoming: no PI addresses. | +-----------------------------------------------+ Assessment of the current IPv6 multihoming situation: ----------------------------------------------------- - Regardless of the availability of multihoming protocols, independence from the service provider is highly desired by multihomers. - The pressure is increasing on RIRs to allocate PI blocks. - RIRs have delayed PI allocation as it creates long-term known problems. - The current policies effectively prohibit multihoming for everyone but ISPs (even RIRs themselves can not use PI addresses). - It is clear at this time that PA addresses alone do not address today's needs. The current roadmap to IPv6 multihoming is: - Offer a short-term alternative that would prevent the deployment of PI addresses: "geo for now". - Finish developing a long-term, scalable solution: MHAP. - ISP multihoming remains unchanged. "geo for now" ------------- The idea is to create a low-ambition geographic aggregation model that can be adopted using today's technology, requires no changes to physical infrastructure and few changes to current operational practices. Current work includes making geo for now and MHAP compatible to the migration from the shorter-term geo for now to the longer-term MHAP smooth. Geo addresses are: - Allocated depending on the location of the site. - Allocated by RIRs or NIRs - Locally portable. - Globally unique. - Aggregatable. Regardless of the actual aggregation ratio achieved, geo addresses are always preferable to PI; PI addresses will never be aggregated, geo will some day. Therefore, no IPv6 PI addresses must be ever used in any situation and geo addresses must be used instead. In other words, PI addresses offers no hope of aggregation, geo addresses do. The choice between PI and geo is a no-brainer: geo. MHAP ---- This is currently a working document of the ipv6mh mailing list. The draft is relatively mature and an earlier form was submitted to the IETF a year ago (it has evolved since then). MHAP is a full-blown mid to long-term solution. MHAP Features: - Zero impact on the DFZ's routing table. The DFZ's routing table stays strongly aggregated. - Provides multihomed, provider-independent, /48 address space for any site. - Very large organizations that require more can get a /47 or bigger. - Scalable (4 billion multihomed sites, initial allocation). - Addresses are aggregated at geographic areas boundaries. - There is no need for Internet eXchanges at aggregation boundaries. - MHAP is transparent to hosts. No modification of existing stacks is required and end-to-end traffic is unchanged. - More than 90% of routers would not require modification. - Provides global load balancing. - Provides survivability of open sessions. - There is no MTU reduction. - Can be run on hardware available today. - Gradual migration, no "flag day". - IPv6 only protocol. - MHAP provides site multihoming. ISP multihoming is unchanged. MHAP Concepts: - Multihomed address space exists only at the edge. The end-to-end multihomed traffic is carried over aggregated PA address space in the core. - A site gets PA addresses from ISPs and multihomed provider-independent space from a registry. - The process of transforming multihomed traffic to PA and back is called aliasing and relies on the presence of rendezvous points and aggregators. - Rendezvous points and aggregators answer topology requests. They do not carry traffic. - There is a separation between the DFZ's routing table and the multihomed space routing tables. The entire multihomed space is represented by two aggregates in the DFZ's routing table. - The only multihomed traffic on backbones is topology requests and routing updates. - There are two types of multihomed addresses: o Centralized, portable, for large multinational organizations. o Geographical, portable only within a geographic area. - There is no centralized table for geographic addresses. - The routing table for centralized prefixes remains contained to rendezvous points. - No multihomed site receives full multihomed tables. All a multihomed site needs is the small DFZ's routing table.

2 1

RE: Getting addresses
by Michel Py 14 May '02

14 May '02

Joao, > Joao Luis Silva Damas wrote: > My questions so far have not been trying to find a way of giving > any RIR arguments to fight against the deployment of IPv6 or > anything else in the Internet. If the RIRs ever become a stumbling > block for the development of the Internet you might just as well > get rid of them (us). On the other hand, RIRs have a role to play in controlling practices that lead to problems in the long run, such as the v4 swamp. > Currently, the policy refers to the concept of "site" which has > been left intentionally "fuzzy". That is OK as a starting point, > to be refined later, as we (all) gain experience. In practice, > this fuzziness will be translated into acceptance of any definition > of "site" that a requester puts forward. And each site is entitled > to a /48 as per the IAB's recommendation, which seems to be widely > accepted. I think there is a de-facto definition of site, though. - No subnetting = /64 - Some subnetting = using what was used to be called the SLA bits = site = /48. > And we get to the magical word: multihoming. > Organisations which fear not having their addresses routed when > they multihome might generate requests that ensure they get a /32. > If conservation is not a problem, this should also not be a problem. It actually is in the long run, and there is a balance to find here: Even if saving IPv6 addresses is not the goal, one must be careful in not creating monstruosities. A /48 is overkill already for most multihomers. Making it a /32 is a monstruosity. In order to eliminate the fuzziness you mentioned above, the policy needs to either: a) Allocate a /32 to everyone. b) Have clear rules if multihomers get a different size (which I do not recommend at this point in time). > In practice the current policy is equivalent to the one suggested > in January: When you have plans to use IPv6, become an LIR if you > aren't one already. The RIR will register the numbers. Again, this is not a rant about the new policy, I actually support it in the lack of a better one. Nevertheless, in practice the new policy is equivalent to giving a /32 PI to multihomers, as cheating to fulfill a legitimate business need is done. We have seen increases in 6bone pTLA requests, and we will see requests for LIR allocations for the same reason: Get a global prefix that will be seen in the global routing table. We have done that in the past already: it's called the pre-CIDR swamp. Having everyone become a LIR can _NOT_ continue forever. There is a limited time we collectively have before we have to define a new policy. > If the RIRs ever become a stumbling block for the development of > the Internet > And we get to the magical word: multihoming. If a multihoming solution is not deployed during this limited time I just mentioned, I hope that everyone understands what will be the next RIR policy in order not to become a stumbling block for the development of the Internet: "Give a /48 PI to anyone that wants one" The writing is on the wall. I renew my call to people with a brain in working order to join the effort in developing a v6 multihoming solution before it is too late. Michel.

1 0

Re: Getting addresses
by Joao Luis Silva Damas 14 May '02

14 May '02

I agree with Wilfried about the re-iteration, but it is probably just a reflection of the fact that the devil is in the details and you encounter details when you try to actually translate paper into action. My questions so far have not been trying to find a way of giving any RIR arguments to fight against the deployment of IPv6 or anything else in the Internet. If the RIRs ever become a stumbling block for the development of the Internet you might just as well get rid of them (us). One thing the RIRs need, though, is a set of guidelines (policy) that is as clear as possible. Fuzziness leads to interpretations that might lead to a perception of unfairness. Currently, the policy refers to the concept of "site" which has been left intentionally "fuzzy". That is OK as a starting point, to be refined later, as we (all) gain experience. In practice, this fuzziness will be translated into acceptance of any definition of "site" that a requester puts forward. And each site is entitled to a /48 as per the IAB's recommendation, which seems to be widely accepted. Different ISPs will have different perceptions of what services they roll out to customers and how they position themselves in the market. That has always been the case, with ISPs or any other company, and an "IP allocation policy" should not attempt to dictate a business model. The only other criteria that is clearly set down in the current instance of the allocation policy is the number of sites the LIR expects to provide services to. Currently a minimum of 200 "planned" in a couple of years is the threshold. Fine. I understand this is a very flexible way of going about things, and probably a good one until there is a certain critical mass of IPv6 service providers. It also circumvents the need to discuss a "PI" policy now. Since the barrier is so low and refers to medium term plans, in practice almost any organization will be able to get an allocation. This last part, the "almost", is where I see people finding most of the problems. The difference between getting the allocation or not is not whether the organisation gets enough addresses, it is more likely related to the uncertainty of whether those addresses will be useful. In other words, if the routing announcements will make it throughout the whole IPv6 Internet without the need to do anything special. And we get to the magical word: multihoming. Organisations which fear not having their addresses routed when they multihome might generate requests that ensure they get a /32. If conservation is not a problem, this should also not be a problem. In practice the current policy is equivalent to the one suggested in January: When you have plans to use IPv6, become an LIR if you aren't one already. The RIR will register the numbers. Cheers, Joao

1 0

Off-topic university networks. WAS Re:Getting addresses
by Rudolf van der Berg 14 May '02

14 May '02

At 17:09 13-5-2002, Michel Py wrote: >This is a grey area for sure. I don't think that providing access to >currently enrolled students can reasonably be considered bad use of >public money, for the good reason that a simple accounting trick will >assign that to tuition fees students do pay. Besides, access to the >university's computer resources is mandatory for studying, and a >university provided dial pool does have some operational advantages in >terms of security. Oh well, I'll pitch in on the discussion what universities do with their networks, having been the vice-chairman of Twente University's campus network (which included dial-in, ADSL and cable). Twente University saw it as a strategic decision for getting new students, changing the way people study and to support new (business) initiatives of students, to provide all on campus rooms in '94 with 10mbit connections (now 100mbit switched). Later they have extended these connections to every student and employee, so that they can now get a 130.89.x.x connection over cable, adsl, or POTS in the region of the university. For everything off-campus they made use of businesses, who were happy enough to have such a big customer. The university used its power to buy wholesale to get cheaper and better connections for all and realize the strategic objective that employees could work from anywhere. These are legitimate reasons and every business could/can offer to provide these services, so no anti-competitive behavior there. But yes, buying wholesale means that the profit per user is less for the provider, if anyone (Francis?) finds this anti-competitive... go ahead, challenge modern economics. :-) In the Netherlands Surfnet, the NREN, has been very active in hooking up student accommodations with fiber under the Gigaport FttD-project (Fiber to the Dormitory). The motivation for this was, that there would be no commercial provider willing to do this and offer gigabit connections to the dorms. Furthermore these projects are very helpful as pilots for fiber to the home projects. It has always been the goal of academic networks to support new developments. So I don't see a problem there either. A nice added bonus for providers is, that students that have tasted the sweetness of 100mbit links to the internet will generally opt for ADSL of cable services when they leave university. > > In terms of the get-outs, I expect that commercial spin-offs or > > products from the universities (e.g. a university selling > > intelligent widgets) may be hosted from a university with such a > > spin-off. > >Most of these will get grants that include cash from the university, >there is nothing fundamentally wrong in making connectivity part of the >grant. Agree, another way of solving it is what Twente University recently did.. They allowed separate VLAN's over their campus infrastructure, so that commercial parties on campus could hook up to commercial providers connected to the nearby internet exchange. Before this, commercial use was explicitly forbidden, because it violated the rules for use of the NREN. Now the commercial traffic is nicely routed of campus to a commercial party that can charge it. So no anti-competitive behavior there either. Even better... new business opportunities have emerged for some ISP's. Rudolf van der Berg Dutch German Internet Exchange (NDIX) http://www.ndix.net

2 1

Re: Getting addresses
by Xavier Henner 14 May '02

14 May '02

> > It is does not appear realistic to expect ISPs to "interview" each > > customer and ask them whether they intend to subnet or not. > > so, you think it's not realistic to ask the customer if he intends > to subnet or not ? > > ipv4 view: why not to give all your dialup customers in ipv4 > per default a /28 delegation, and the customer decides to do NAT, > all in one subnet, or more subnets ? No : we have not enough space on v4 The RIPE-NCC will not agree Nerim is a small ISP but we give a static IPv4 adress to custommers if they want. If a custommer need a /29 or /28, and can justify it, we send the request to RIPE-NCC All of this costs time. > no, i would give everyone /64 per default, unless the customer > says, "no, i want to do subneting", then i will give him a /48 > without discussion. complicated and unjustified (adress space in v6 is *not* rare) > > Everyone will probably end up with a /48: it reduces costs for the ISP. > > ipv4 view: some years ago,....: Everyone will probably end up with a /24 There is millions of /24 IPv4 There is more than a million of billions of /48 IPv6 not the same thing > hm...i think the cost point isn't a good argument, sorry. > does it cost so much to do subnetting ? you have one block with /64 delegations > and one with /48's. No But paying somebody to change the adress space of the customer cost IPv6 adress space costs *nothing* -- Xavier Henner Responsable de l'exp�rimentation IPv6 Nerim -- Fournisseur d'acc�s � Internet URL: <http://www.nerim.net/>

4 3

RE: Getting addresses
by Michel Py 14 May '02

14 May '02

Pekka and Gert, > Pekka Savola wrote: > I don't think university-like organizations ever need more than > a /48. One and the only really problematic thing is if they > provide access to students/staff/etc., e.g. via DSL, dial-up, > dorms or what. /64 would usually be ok (except very large > universities and the like), /48 would not. Universities are ISPs, and it is perfectly legitimate to assign /48s to students. Invariably, each class I teach IPv6 subnetting, students ask me "ok, now what if want to try this?" and my answer is "go to freenet6 and get a /48". Yes, they're going to use only two or three subnets possibly with only with one host each but they do need more than a /64 and I do not think that allocating anything between a /48 and a /64 is debatable at this point. > How do you think that e.g. Nokia, Cisco or Microsoft will do this? > I think everyone agrees that /48 should be very well enough, Pekka, I am sorry but this is wrong. One of the valuable participants we have in ipv6mh is Craig Huegen, the lead network architect for Cisco, and the _first_ question he asked about the protocols we are developing is how they would work for people that need more than a /48. With an HD of 0.8 Cisco needs between a /46 and a /45. > Gert Doering: > So with some reason instead of a "this can't work!" attitude, I think > this can work well - if one insists on doing non-useful things, it will > break (but yes, this is a problem with the "one site" = /48 rule, because > it's too vague to work unless people are reasonable). Yes, the "one site" = /48 rule leads to terrible HD, but a university with 50K students would indeed fit within a /32. The question is: will people be reasonable. Michel.

12 28

Re: Getting addresses
by Xavier Henner 13 May '02

13 May '02

> > complicated and unjustified > > compilcated ? no! the same as in ipv4 But I didn't say it was simple in IPv4 > > (adress space in v6 is *not* rare) > > along time ago, there where some people who said the same sitting on the ipv4 world. > > > There is millions of /24 IPv4 > > There is more than a million of billions of /48 IPv6 > > > > not the same thing > > it's the same, only the comma is on a different point. There was less than 3 billions of human beings or so when IPv4 was created There is 6 billions of human beings today There will never be more than 30 billions of human beings on this planet (changing planet is not as easy as changing Internet Protocol) With the /48 model, infrastructure need no adress space (Nerim is an ISP, infrastructure is ONE /48, custommers will have 99.999% of the alocation) If you give 2 thousands of /48 per human beeing, you will have spare space. > > No But paying somebody to change the adress space of the customer cost > > the initional request of ip addresses doesn't cost anything in my company. > we only bill when a customers want's an addional network. But your competitors will give a /48 directly without charging. And it will cost them less time. They will not break the IAB recommendations (but you will). > > IPv6 adress space costs *nothing* > > IPv4 space costs *nothing*. did you pay diffrent prices @ ripe for diffrent ISP allocations ? > /16 very cheap, only 500.000 Euro per year ? > if you buy 2 * /16's you get one /20 for free! Fixed costs (part of ISP job); and a big Cisco router costs much more than the RIPE fee. But IPv4 is rare -- Xavier Henner Responsable de l'exp�rimentation IPv6 Nerim -- Fournisseur d'acc�s � Internet URL: <http://www.nerim.net/>

1 0

Re: Getting addresses
by James Aldridge 13 May '02

13 May '02

Pekka Savola wrote: > On Mon, 13 May 2002, Xavier Henner wrote: > > > If you *do* have 200 IPv6 customers *today* (those that pay for the > > > service, not necessarily IPv6 service, but something), I'd surely conside r > > > one a real player here. > > > > > > But an issue is about DSL operators with 1,000,000 IPv4 customers > > > beginning to introduce IPv6 services.. > > > > I don't see any issue with DSL operators. > > Can you explain it ? > > With HD ratio of 80%, the operator would require a /23. Just because a HD ration of 80% is the minimum utilisation required to get a subsequent allocation, this doesn't mean that higher utilisation can't be achieved in some (and shouldn't be an aim in all) cases. I can imagine DSL operators offering /64 (single subnet) and /128 (single host) assignments to some customers (in the same way that single IPv4 addresses are often assigned now), in which case these 1,000,000 customers could be handled with an allocation of between a /39 and a /103 while maintaining an HD ration of 0.8 within that allocation - i.e. one /32 would be more than enough! James

4 5