"Wilfried Woeber, UniVie/ACOnet" wrote:
I think we were talking about increasing the size of a sTLA (when the requirement for that can be documented), rather than allocating another sTLA?!
OK, my last mail was maybe a bit terse. Some background might help. We (KPNQwest, formerly EUnet) are a "supernational" registry. In the IPv4 world this is much like having 6 individual large registries with the corresponding number of open allocations that implies. Now, in the IPv6 world I'm told that we can't get an IPv6 sTLA for our direct backbone customers or for any of our other national networks because KPNQwest Finland (covered by the eu.eunet supernational registry) already has our one sub-TLA. Of course, that one sub-TLA gives us a total amount of address space which is adequate for our current requirements for the whole network but once this is split over each of about 20 separate autonomous systems, each with their own routing policy, this is hardly going to result in optimally aggregatable routing... James
Also, I seem to remember that the NCC reserves some space in the address tree for that, so you might be able to obtain a "2nd" sTLA back-to-back with the original one, which is equivalent to decreasing the prefix length.
I guess you would be free to structure that (combined/extended) address space internally (for distribution to customers by more than one operational unit).
But probably I am missing something essential here.
Wilfried. ______________________________________________________________________
Of course, there would be at least one more sub-TLA allocated if the IPv4 rules for supernational registries were to be applied to IPv6 instead of restricting these to only having a single sub-TLA allocation... :-(
James
-- James Aldridge, Senior Network Engineer (IP Architecture) KPNQwest, Singel 540, 1017 AZ Amsterdam, NL Tel: +31 70 379 37 03; GSM: +31 65 370 87 07
-----------------------------------------------------------------------------
_________________________________:_____________________________________ Wilfried Woeber : e-mail: Woeber@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 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~:~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ First things first, but not necessarily in that order....
James schrieb:
Now, in the IPv6 world I'm told that we can't get an IPv6 sTLA for our
direct
backbone customers or for any of our other national networks because KPNQwest Finland (covered by the eu.eunet supernational registry) already has our one sub-TLA.
Of course, that one sub-TLA gives us a total amount of address space which is adequate for our current requirements for the whole network but once this is split over each of about 20 separate autonomous systems, each with their own routing policy, this is hardly going to result in optimally aggregatable routing...
Actually, you have a /34, haven't you? EU-EUNET-20000403 and DE-XLINK-20000510. Very interesting that KQDE got an assignment as at that time they already where 100% KQ. -- Arnold
I would like to add although we are not a supernational registry and all that implies ;) we have the same issue. We have been allocated our start up space in IPv6 which is fine for now but would it not be better to be more forward thinking when allocating IPv6 space and allocate enough space to aggregate fully throughout the EMEA region and so implement the best possible aggregation. This is not just a cry for more space because we are big so we deserve it, we are seriously looking to a time when IPv6 is used in anger and we have to do real aggregation throughout EMEA. We do not want to assign IPv6 on a per LIR basis, rather sub-allocate IPv6 space to our current LIR structure since we are all in the same network it makes sense. BTW we are currently writing our internal IPv6 deployment policy. Regards, Stephen Burley UUNET EMEA Hostmaster ----- Original Message ----- From: "James Aldridge" <jhma@KPNQwest.net> To: "Wilfried Woeber, UniVie/ACOnet" <woeber@cc.univie.ac.at> Cc: <lir-wg@ripe.net>; <ipv6-wg@ripe.net> Sent: Wednesday, August 08, 2001 2:49 PM Subject: Re: 90 IPv6 sub-TLA allocations made
"Wilfried Woeber, UniVie/ACOnet" wrote:
I think we were talking about increasing the size of a sTLA (when the requirement for that can be documented), rather than allocating another sTLA?!
OK, my last mail was maybe a bit terse. Some background might help.
We (KPNQwest, formerly EUnet) are a "supernational" registry. In the IPv4 world this is much like having 6 individual large registries with the corresponding number of open allocations that implies.
Now, in the IPv6 world I'm told that we can't get an IPv6 sTLA for our direct backbone customers or for any of our other national networks because KPNQwest Finland (covered by the eu.eunet supernational registry) already has our one sub-TLA.
Of course, that one sub-TLA gives us a total amount of address space which is adequate for our current requirements for the whole network but once this is split over each of about 20 separate autonomous systems, each with their own routing policy, this is hardly going to result in optimally aggregatable routing...
James
Also, I seem to remember that the NCC reserves some space in the address tree for that, so you might be able to obtain a "2nd" sTLA back-to-back with the original one, which is equivalent to decreasing the prefix length.
I guess you would be free to structure that (combined/extended) address space internally (for distribution to customers by more than one operational unit).
But probably I am missing something essential here.
Wilfried. ______________________________________________________________________
Of course, there would be at least one more sub-TLA allocated if the IPv4 rules for supernational registries were to be applied to IPv6 instead of restricting these to only having a single sub-TLA allocation... :-(
James
-- James Aldridge, Senior Network Engineer (IP Architecture) KPNQwest, Singel 540, 1017 AZ Amsterdam, NL Tel: +31 70 379 37 03; GSM: +31 65 370 87 07
--------------------------------------------------------------------------
---
_________________________________:_____________________________________ Wilfried Woeber : e-mail: Woeber@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 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~:~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ First things first, but not necessarily in that order....
Hi, On Thu, Aug 09, 2001 at 09:36:15AM +0100, Stephen Burley wrote:
I would like to add although we are not a supernational registry and all that implies ;) we have the same issue. We have been allocated our start up space in IPv6 which is fine for now but would it not be better to be more forward thinking when allocating IPv6 space and allocate enough space to aggregate fully throughout the EMEA region and so implement the best possible aggregation. This is not just a cry for more space because we are big so we deserve it, we are seriously looking to a time when IPv6 is used in anger and we have to do real aggregation throughout EMEA. We do not want to assign IPv6 on a per LIR basis, rather sub-allocate IPv6 space to our current LIR structure since we are all in the same network it makes sense. BTW we are currently writing our internal IPv6 deployment policy.
As far as I remember the IPv6 policy discussions on the last RIPE meetings, one thing that was voiced repeatedly was "if we have to hand out /48's to customers, a /35 for the LIR itself is not enough" (considering hierarchical strutures - either due to multinational networks, or due to hierarchies of resellers having re-selling customers themselves - 13 bits to work in is just not enough). Also, it hasn't really been shown why we need slow-start *in slow-start space*(!). It's not like we want our own TLA, but I think the RIRs are being way too conservative. Old IPv4 habits...? Gert Doering -- NetMaster -- SpaceNet AG Mail: netmaster@Space.Net Joseph-Dollinger-Bogen 14 Tel : +49-89-32356-0 80807 Muenchen Fax : +49-89-32356-299
Hiya Folks, I would like to confirm what Gert wrote. Myself and *many* others stated 13 bits is not enough. There was definitely concensus on this point. There were other comments about "H-ratio's" and why only 13 bits for a Global ISP when a single user gets 80 bits and the RIR's 35. Other comments about Old IPv4 habits were also made. A separate issue of avoiding non-4 bit boundaries was also made. Even ripe is affected here since TWO reverse domain delegations are needed for each current sub-TLA. As a result, various persons (Randy,and I think Mirjam or Nurani) said the TLA issues were going to be re-worked, and that RFC2928 would be made obselete. This is essential in my view. However, either nothing is happening, or it is all happening behind closed doors both of which are wrong in my opinion. Cheers Dave (normally djp@djp.net but posting might be faster from this subscribed address) On Thu, 9 Aug 2001, Gert Doering wrote: ->Hi, -> ->As far as I remember the IPv6 policy discussions on the last RIPE meetings, ->one thing that was voiced repeatedly was -> -> "if we have to hand out /48's to customers, a /35 for the LIR itself -> is not enough" -> ->(considering hierarchical strutures - either due to multinational ->networks, or due to hierarchies of resellers having re-selling customers ->themselves - 13 bits to work in is just not enough). -> ->Also, it hasn't really been shown why we need slow-start *in slow-start ->space*(!). It's not like we want our own TLA, but I think the RIRs are ->being way too conservative. Old IPv4 habits...? -> ->Gert Doering -> -- NetMaster
On Thu, Aug 09, 2001 at 11:18:15AM +0200, Gert Doering wrote:
As far as I remember the IPv6 policy discussions on the last RIPE meetings, one thing that was voiced repeatedly was
"if we have to hand out /48's to customers, a /35 for the LIR itself is not enough"
For your information: We are currently planning a joint session for ipv6 allocation policy issues for the next RIPE meeting. It would be really nice if we can get volunteers from the community who can give a brief presentation on possible solutions. The problem description is pretty clear by now, however, I have not seen any (public) proposals yet on how to solve it. Obviously, there are multiple ways to deal with the issue and it would be nice to discuss advantages and disadvantages of different solutions. David K. ---
During the January meeting in Amsterdam we had presentions from Bernard Tuy (Renater), and from Stuart Prevost (BT), and first consensus on this issue. During the April meeting in Bologna, a very comprehensive document, with the problem description and with clear proposals was presented by Nial Murphy, and again converging views were expressed. See Nial Murphy presentation : http://www.ripe.net/ripe/mail-archives/ipv6-wg/20010101-20010401/msg00035.ht... http://www.enigma.ie/articles/global-ipv6-alteration.html ETNO expressed supports to Nial's proposals and introduced a common ETNO position on this issue (See: http://www.ripe.net/ripe/mail-archives/ipv6-wg/20010401-20010701/msg00016.ht... http://www.etno.belbone.be/site/positions.htm) Do we really need to explore solutions again or do we need a new Draft from the RIRs taking into account those proposals and the consensus expressed around, and able to be approved by the community ? Alain Bidron. David Kessens a écrit :
On Thu, Aug 09, 2001 at 11:18:15AM +0200, Gert Doering wrote:
As far as I remember the IPv6 policy discussions on the last RIPE meetings, one thing that was voiced repeatedly was
"if we have to hand out /48's to customers, a /35 for the LIR itself is not enough"
For your information:
We are currently planning a joint session for ipv6 allocation policy issues for the next RIPE meeting.
It would be really nice if we can get volunteers from the community who can give a brief presentation on possible solutions. The problem description is pretty clear by now, however, I have not seen any (public) proposals yet on how to solve it. Obviously, there are multiple ways to deal with the issue and it would be nice to discuss advantages and disadvantages of different solutions.
David K. ---
Hiya all, There were also many concrete proposals in my mail of 8.2.2001 to these lists and available for browsing at: http://www.djp.net/ipv6/proposal.html for those who do not wish to access the mail archives. It's time this and the other proposals, which are all approximately in agreement, were put into a revised allocation policy. We already have 166 routes in the IPv6 routing table. The sooner we adopt the revised allocation policy the less likely it is the wrongly sized routes will need to hang around for ever - Cisco memory is very expensive - IPv6 addresses are plentiful. Cheers Dave On Thu, 9 Aug 2001, BIDRON Alain BRX/DAP wrote: ->During the January meeting in Amsterdam we had presentions from Bernard ->Tuy (Renater), and from Stuart Prevost (BT), and first consensus on this ->issue. -> ->During the April meeting in Bologna, a very comprehensive document, ->with the problem description and with clear proposals was presented by ->Nial Murphy, and ->again converging views were expressed. -> ->See Nial Murphy presentation : ->http://www.ripe.net/ripe/mail-archives/ipv6-wg/20010101-20010401/msg00035.ht... ->http://www.enigma.ie/articles/global-ipv6-alteration.html -> ->ETNO expressed supports to Nial's proposals and introduced a common ->ETNO position ->on this issue ->(See: ->http://www.ripe.net/ripe/mail-archives/ipv6-wg/20010401-20010701/msg00016.ht... ->http://www.etno.belbone.be/site/positions.htm) -> ->Do we really need to explore solutions again or do we need a new Draft ->from the RIRs taking into account those proposals and the consensus ->expressed around, and able to be approved by the community ? -> ->Alain Bidron. -> ->David Kessens a �crit : ->> ->> On Thu, Aug 09, 2001 at 11:18:15AM +0200, Gert Doering wrote: ->> > ->> > As far as I remember the IPv6 policy discussions on the last RIPE meetings, ->> > one thing that was voiced repeatedly was ->> > ->> > "if we have to hand out /48's to customers, a /35 for the LIR itself ->> > is not enough" ->> ->> For your information: ->> ->> We are currently planning a joint session for ipv6 allocation policy ->> issues for the next RIPE meeting. ->> ->> It would be really nice if we can get volunteers from the community ->> who can give a brief presentation on possible solutions. The problem ->> description is pretty clear by now, however, I have not seen any ->> (public) proposals yet on how to solve it. Obviously, there are ->> multiple ways to deal with the issue and it would be nice to discuss ->> advantages and disadvantages of different solutions. ->> ->> David K. ->> ---
In your previous mail you wrote: Also, it hasn't really been shown why we need slow-start *in slow-start space*(!). It's not like we want our own TLA, but I think the RIRs are being way too conservative. Old IPv4 habits...? => protection of a business... (if this is not the case someone can believe this). Francis.Dupont@enst-bretagne.fr
Hi, On Thu, Aug 09, 2001 at 03:52:00PM +0200, Francis Dupont wrote:
Also, it hasn't really been shown why we need slow-start *in slow-start space*(!). It's not like we want our own TLA, but I think the RIRs are being way too conservative. Old IPv4 habits...?
=> protection of a business... (if this is not the case someone can believe this).
There are rumors to that extent, yep. But it's unlikely that IPv4 requests will stop so suddenly that RIPE hostmasters will all lose their jobs... Gert Doering -- NetMaster -- SpaceNet AG Mail: netmaster@Space.Net Joseph-Dollinger-Bogen 14 Tel : +49-89-32356-0 80807 Muenchen Fax : +49-89-32356-299
On Thu, 9 Aug 2001, Gert Doering wrote:
As far as I remember the IPv6 policy discussions on the last RIPE meetings, one thing that was voiced repeatedly was
"if we have to hand out /48's to customers, a /35 for the LIR itself is not enough"
(considering hierarchical strutures - either due to multinational networks, or due to hierarchies of resellers having re-selling customers themselves - 13 bits to work in is just not enough).
Also, it hasn't really been shown why we need slow-start *in slow-start space*(!). It's not like we want our own TLA, but I think the RIRs are being way too conservative. Old IPv4 habits...?
This was certainly the concensus also of the GEANT IPv6 Working Group, the opinions coming from representatives of the National Research Networks in Europe. At a minimum we would expect the removal of slow start /35 to offer /29's. Moving the bit boundaries is a secondary concern, but one worth considering sooner rather than later. It is not clear that a /48 is sufficient for a university with satellite locations and that also feeds local colleges/etc, and that offers remote access to staff and students. Thus a national research network would struggle within a /35. Recommendation #2 of http://www.enigma.ie/articles/global-ipv6-alteration.html is a minimum requirement. As a group, we have not discussed more ambitious suggestions such as those at http://www.djp.net/ipv6/proposal.html where a /16 is suggested for the "supernational" organisations. Tim
Hiya all, In my view it is not so ambitious to be recommending /16 allocations to "supernational" organisations as this is the policy originally recommended in RFC2374 and elsewhere. On the other hand: "Will these supernational organisations be advertising parts of this /16 into the global routing table". If the answer is yes, then I think they should be making multiple regional or national requests, and receiving multiple /20 or /24 allocations according to their likely longterm requirements in each region. Everytime an LIR requests and gets additional addresses because of an insufficiently small original allocation (whether through the 80% rule, or 90% according to RFC2450 !! or my suggested 10% rule), the RIR's have effectively made a mistake as this means one unnecessary route in the routing table. I'm not suggesting the RIR's give a /16 (or /20,/24,/28,/32, for that matter) to anybody who asks. The requester must justify that such an allocation is appropriate (with the RIR's taking a much more generous stance in contrast to what they need to do with IPv4). Cheers Dave BT Ignite GmbH, On Sun, 12 Aug 2001, Tim Chown wrote: ->As a group, we have not discussed more ambitious suggestions such as those ->at http://www.djp.net/ipv6/proposal.html where a /16 is suggested for the ->"supernational" organisations. ->Tim
Colleagues, I've just done some calculations that shows the maximum theoritical utilisation that can be achieved is 75% whilst maintaining the minimum size of routing table. That is if you take a large number of subnets, each subnet containing a random number of hosts, and assign to each subnet the nearest power of 2 larger than the number of hosts, the utilisation you get is 75%. This is a 75% utilisation per level of network hierarchy. So if we assume 3 levels of network hierarchy and each level doing perfect routing aggregation and perfect address allocation we will get an overall utilisation of 0.75^3 = 0.422 == 42% overall utilisation for the TLA. I'd like to bet that if we have a network with enough hosts to justify 64 bits of address space it'll also be large enough to require more than 3 levels of network hierarchy. Any requirements to get high address space utilisation out of IPv6 can simply be demonstrated to lack scaling qualities. Regards, ------------------------------------------------------------------------------- Peter Willis | E-mail: peter.j.willis@bt.com IP Technology Strategist | Phone: 01473 645178 Fax: 01473 644506 BTexact Technologies CTO | -------------------------------------------------------------------------------
This is not including the needed aggregation for multi-national registries, its fine for a single network but, still would tie your hands when sub-allocating to multiple LIR's. Regards Stephen Burley UUNET EMEA Hostmaster
Colleagues,
I've just done some calculations that shows the maximum theoritical utilisation that can be achieved is 75% whilst maintaining the minimum size of routing table. That is if you take a large number of subnets, each subnet containing a random number of hosts, and assign to each subnet the nearest power of 2 larger than the number of hosts, the utilisation you get is 75%.
This is a 75% utilisation per level of network hierarchy.
So if we assume 3 levels of network hierarchy and each level doing perfect routing aggregation and perfect address allocation we will get an overall utilisation of 0.75^3 = 0.422 == 42% overall utilisation for the TLA.
I'd like to bet that if we have a network with enough hosts to justify 64 bits of address space it'll also be large enough to require more than 3 levels of network hierarchy. Any requirements to get high address space utilisation out of IPv6 can simply be demonstrated to lack scaling qualities.
Regards,
--------------------------------------------------------------------------
Peter Willis | E-mail: peter.j.willis@bt.com IP Technology Strategist | Phone: 01473 645178 Fax: 01473 644506 BTexact Technologies CTO | --------------------------------------------------------------------------
This is not including the needed aggregation for multi-national registries, its fine for a single network but, still would tie your hands when sub-allocating to multiple LIR's.
This is simply accomodated by mutliplying the maximum utilisation figure by 0.75 for every layer of hierarchy or aggregation in the network. So for a 3 layer network plus another layer for international aggregation the maximum utilisation falls to 0.75^4=0.316 == 32%. This is the theoritical maximum address utilisation that can be achieved without breaking aggregates. I think we should penalize anyone who gets a better address utilisation than this because they are obviously announcing more routes than they need to ;-) Peter.
Hi, I think the RIR staff working on the new policy draft understand the issue. I believe the new draft will reflect this by moving away from a fixed percentage to using the huitema/durand ratio which is meant to give a consistent view of space utilization when using variable levels of hierarchy. Joao Damas RIPE NCC At 17:27 +0100 14/8/01, Peter Willis wrote:
Colleagues,
I've just done some calculations that shows the maximum theoritical utilisation that can be achieved is 75% whilst maintaining the minimum size of routing table. That is if you take a large number of subnets, each subnet containing a random number of hosts, and assign to each subnet the nearest power of 2 larger than the number of hosts, the utilisation you get is 75%.
This is a 75% utilisation per level of network hierarchy.
So if we assume 3 levels of network hierarchy and each level doing perfect routing aggregation and perfect address allocation we will get an overall utilisation of 0.75^3 = 0.422 == 42% overall utilisation for the TLA.
I'd like to bet that if we have a network with enough hosts to justify 64 bits of address space it'll also be large enough to require more than 3 levels of network hierarchy. Any requirements to get high address space utilisation out of IPv6 can simply be demonstrated to lack scaling qualities.
Regards,
------------------------------------------------------------------------------- Peter Willis | E-mail: peter.j.willis@bt.com IP Technology Strategist | Phone: 01473 645178 Fax: 01473 644506 BTexact Technologies CTO | -------------------------------------------------------------------------------
--
Will the new draft include the priciple of MIR's which i detailed on the list (which stangly got no negative response) and will it also understand the concept of sub-allocation? Regards, Stephen Burley UUNET EMEA Hostmaster ----- Original Message ----- From: "Joao Luis Silva Damas" <joao@ripe.net> To: "Peter Willis" <pjw@ip-engineering.bt.com>; "Dave Pratt" <djp-ripe-lists@djp.net> Cc: "lir-wg" <lir-wg@ripe.net>; "ipv6-wg" <ipv6-wg@ripe.net> Sent: Wednesday, August 15, 2001 3:03 PM Subject: Re: 90 IPv6 sub-TLA allocations made
Hi,
I think the RIR staff working on the new policy draft understand the issue. I believe the new draft will reflect this by moving away from a fixed percentage to using the huitema/durand ratio which is meant to give a consistent view of space utilization when using variable levels of hierarchy.
Joao Damas RIPE NCC
At 17:27 +0100 14/8/01, Peter Willis wrote:
Colleagues,
I've just done some calculations that shows the maximum theoritical utilisation that can be achieved is 75% whilst maintaining the minimum size of routing table. That is if you take a large number of subnets, each subnet containing a random number of hosts, and assign to each subnet the nearest power of 2 larger than the number of hosts, the utilisation you get is 75%.
This is a 75% utilisation per level of network hierarchy.
So if we assume 3 levels of network hierarchy and each level doing perfect routing aggregation and perfect address allocation we will get an overall utilisation of 0.75^3 = 0.422 == 42% overall utilisation for the TLA.
I'd like to bet that if we have a network with enough hosts to justify 64 bits of address space it'll also be large enough to require more than 3 levels of network hierarchy. Any requirements to get high address space utilisation out of IPv6 can simply be demonstrated to lack scaling qualities.
Regards,
---------------------------------------------------------------------------
Peter Willis | E-mail: peter.j.willis@bt.com IP Technology Strategist | Phone: 01473 645178 Fax: 01473 644506 BTexact Technologies CTO |
---------------------------------------------------------------------------
--
At 15:39 +0100 15/8/01, Stephen Burley wrote:
Will the new draft include the priciple of MIR's which i detailed on the list (which stangly got no negative response) and will it also understand the concept of sub-allocation?
I can't answer this directly as I haven't seen the latest version, but Mirjam's keyboard is on fire from all the typing going on, so I think the new draft is coming real soon now. Joao
Regards, Stephen Burley UUNET EMEA Hostmaster
--
Hi, On Wed, Aug 15, 2001 at 03:39:56PM +0100, Stephen Burley wrote:
Will the new draft include the priciple of MIR's which i detailed on the list (which stangly got no negative response) and will it also understand the concept of sub-allocation?
Without knowing anything about that draft :-) I can only speculate. On the concept of MIRs - I don't think they are necessary. This is something that a LIR should do internally - and this means the concept of sub-allocation should be legalized (and formalized) now. LIRs already do sub-allocations, because it's necessary. The hostmasters know this, but frown on it, because it's not in the official policies. So let's change 'em (at least for IPv6, where aggregation is MUCH more important than conservation). Gert Doering -- NetMaster -- SpaceNet AG Mail: netmaster@Space.Net Joseph-Dollinger-Bogen 14 Tel : +49-89-32356-0 80807 Muenchen Fax : +49-89-32356-299
participants (3)
-
BIDRON Alain BRX/DAP -
Dave Pratt -
David Kessens -
Francis Dupont -
Gert Doering -
James Aldridge -
Joao Luis Silva Damas -
Nipper, Arnold -
Peter Willis -
Stephen Burley -
Tim Chown