- ipv6-wg - mailman.ripe.net

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 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

RE: Getting addresses
by Michel Py 13 May '02

13 May '02

Tim, > Tim Chown wrote: > but yes a university may feel it wants to offer connectivity to > past students 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. However, providing access to past students could definitely be seen as using public money to compete with businesses. > staff. I do not think staff is a problem. They get a paycheck that contains public money, why can't they use a dial-up that is paid by public money? > 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. > the really fuzzy area is universities who use their academic > connectivity to give students access from halls of residence, which > if done with open ADSL-like rules may well let students run > commercial services. This is not that bad. Students have 10 meg connectivity in dorms. And yes, we know they'll use it not only for studying but also to surf porn and download pirated .mp3s. From time to time, an idiot will run a warez ftp server and will get busted. Frankly, I'd look the other way if I see a student that runs a website for his on-campus pizza delivery service that helps pay his/her tuition. I think it all comes down to this: fast and reliable Internet access is something that students need. Money and side jobs are also something lotsof them need. They are students for 4 years, after that they get a real job and pay for their own. Michel.

1 0

Re: Getting addresses
by James Aldridge 13 May '02

13 May '02

Pekka Savola wrote: > > 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! > > IAB/IESG recommendation state /48 should be used in scenarios like this. Personally, I don't agree with the "/48 for all" policy but it's there so I suppose we have to live with it... Note, however, that rfc3177 is just the IAB/IESG *recommendations* in the form of a "best current practices" RFC. James

1 0

Re: Getting addresses
by Wilfried Woeber, UniVie/ACOnet 13 May '02

13 May '02

Sorry, Francis, >=> one detail which is never called to mind here is an university which >gives internet access to its students *outside* the university (i.e. >in a context where students can need a /48) should be sued by all >commercial ISPs for illegal competition using public money... but this statement is simply ridiculous.... and completely off-topic as well. Wilfried. _________________________________:_____________________________________ Wilfried Woeber : e-mail: Woeber(a)CC.UniVie.ac.at UniVie Computer Center - ACOnet : Tel: +43 1 4277 - 140 33 Universitaetsstrasse 7 : Fax: +43 1 4277 - 9 140 A-1010 Vienna, Austria, Europe : RIPE-DB: WW144, PGP keyID 0xF0ACB369 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~:~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

2 1

RE: Getting addresses
by Michel Py 13 May '02

13 May '02

David, > David Kessens wrote > You can present whatever you want, but I don't have a 'default > route' that points to any of my ipv6 peers. This is by choice and not by necessity, because you could have one if you wanted to. Here are a few definitions I glanced in what we are currently working on: 1. At the top level, the `default-free zone', it's not possible to define a default route. BGP speakers in the default-free zone really do need to have a route for every possible Internet destination. 2. The core of the Internet is also called the default-free zone (DFZ) because backbone service providers must maintain a complete map of all globally visible network prefixes reachable throughout the Internet: no default routes are allowed. 3. The Default-Free Zone (DFZ) of the Public Internet is where routers MUST maintain routing entries for all reachable destinations. In short: A router in the DFZ can NOT have a default route. These are tier-1 transit providers. Opposed to tier-2, where you CAN configure a default route to one or more of your transit providers; it actually is common practice for tier-2 to have a floating static default route as backup in case of BGP snafus. As of today, I have not seen a single IPv6 router where you could not configure a default route, as filtering (or the lack thereof) makes almost every peer a transit provider. The ipv6mh mailing list had this discussion a while ago and here is the definition we currently use: DFZ: the subset of routers that do not have a default route and do not receive a full routing table from a single peer. Michel.

1 0

RE: Getting addresses
by Michel Py 12 May '02

12 May '02

> This is very I advocate separate, "earmarked" blocks for > dialup/DSL/whatever, if it is done in such a big scale that /48 > for organization or /32 for an ISP would not be enough. > If you just give out an /32 which should be used for everything, > I think the problem is more serious. One can not know beforehand > how big a block of that /32 must be reserved to dialup/DSL/whatever. > And after it is reserved, what if there is a need to expand? These > things do happen within a time span of 2-3 years. This is where HD > ratio is useful. I have not seen many universities growing their number of students that fast, but any there is a word to address this: PLANNING. If one evaluates that they would need more than a /32 in three years and don't have the 7,100 sites now, there is legitimate reason to ask why they think they need more a /32. I don't have a problem with that figure of 7,100 except that I would have placed the bar a little higher. As far as having multiple blocks, I personally prefer having only one. Michel.

1 0

RE: Getting addresses
by Michel Py 12 May '02

12 May '02

> Gert Doering wrote: As a side note: nothing prevents you from > making better usage of a /32 - 7100 users is the minimum value > you have to document to get a bigger block. > Depending on the way the internal network hierarchy works (if > it has few levels, and all entities at one level are similarily > sized) it is perfectly allowed to utilize 20.000 /48s or more :-) > - it all depends on your network structure. Pekka, I don't know how you came with these figures but Gert is absolutely right. The HD of a dial pool or tunnel pool depends on two things: 1. The HD of the hardware itself, as there are some strong operational incentives to allocate a block per device, such as allocating a block of 256 sites to an access server that has only 192 modems. 2. The network structure. For a dial pool that size, there is absolutely no need for more than one level of aggregation (that would be the pop level). The device with the worst HD I have used is Cisco AS5400, it has 648 modems and it is temping to allocate a block of 1,024 sites to it. That device itself has a HD of 0.934. Then you have the pop aggregate. In a single pop, the worst you can have is 33 of these boxes. This is 21384 sites on a /32, which is a HD of 0.899. On a 3-pop setup, the worst you would have 9 boxes per pop, which still is 17,496 sites and a HD of 0.881. This is the absolute worst case scenario. The bottom line is that a typically organized /32 dial pool has up to 40k sites. Michel.

2 1