[address-policy-wg] IPv6 prefixes / PI

Hi,


On Tue, 6 Dec 2005, Marcus Gerdon wrote:

> Hi @All,
>
> (in advance I'm sorry for the typo's and mis-chosen wordings as I'm not
> native English and just out of my favourite pub ...)
>
> I'm following this discussion for quite some time now, but to me it looks
> like many of you are going for a 'holy war'.

I also would say that is a strong possibility. :-)


> Although I'm surely misunderstood, I've the impression (as for most
> discussions & votes) only the larger members here will get heard and/or will
> discuss this as the larger part of the members can't effort someone getting
> payed exclusively for community work.
>
> Let me start introducing a simple statement:
>
> Drop IPv6!

Wow! Quite a radical statement. However, if we are going to talk about 
IPv6 deployment, i can agree that the population using IPv6 is rather 
small than the one not using it... so, convincing everyone to remove IPv6 
from their networks is a shorter path than getting worldwide IPv6 
deployment... 
(but not as interesting!)
:-)


> No one (ok, nearly no one) is going to use IPv6.

Besides the fact people are already using it, some people are also
using it without any knowledge about it (i.e. apps using Teredo, 
transparent proxying, newsfeeds, ...).

If we look at the current active ASNs, we should see that networks (ASNs)
using IPv6 are still a small set, yes! - but we don't know how will this 
evolve in the (near) future. Any crystal ball around?

I usually keep my antennas tunned in with what Geoff Huston says about 
IPv4 address exaustion, and try to keep an eye on IANA's list... i hope we 
don't get a new "Y2K" issue, but i've been more optimistic about it than 
nowadays.


> I'm speaking out of experience for de.mainzkom (a few months ago merger to
> de.citykom) being a regional ISP in Germany. Out of approx. 50 business
> customers I've spoken to TWO were interested in  IPv6 trials. Only ONE of
> our transit customers explicitly asked for v6 (a university).

I read this as:
Do you want to satisfy your customer demand? even if that means only 2% 
of your customer base?

Then, some interesting questions arise:
- is the customer able to pay for what he is demanding? (IPv6 is not free, 
you know... IPv6 also means packets getting in and out of your network...)
- is this an opportunity to deploy a service, that 98% of the 
current customer base is not interested, but that can help bring new 
customers?
- are you sure, that 98% figure won't go down in a year's time? and in 2 
year's time?


> How do you think a 'very large' member is going into the discussion for a
> lot more than a /32 ? Same way I stated need for our /32: let's count ANY
> end consumer (may it be a dynamic DSL account) a /64 and each business
> customer a /48.
>
> I had worries myself satisfying the 200-customer-rule - I solved it by
> assuming each of our DSL customers getting assigned a /48 ... regardless
> whether he needs/uses it. In the meanwhile applying this method the /32 is
> filled and to small ... could I argue for another one or maybe /30 or even
> more ?

Usually people don't bother. a /32 is perfectly ok for a small-medium ISP. 
And about the 200-rule... well, that will fade out with time, and when 
sensible policy proposals are accepted! *hint*, *hint*...


> There's a point to the increasing number of IPv4 PI space allocated:
>
> Enterprise customers WANT to be independent f their ISP. Quite a number of
> them being governmental organizations or such HAVE TO open a Request for
> Proposal every few years (2-3) and change their ISP equivalently. Last time
> I had to explain that there's IPv4 PI but no IPv6 PI available. What do you
> think ? They dropped v6 completely!

<probably political incorrectness>
As I see it, most of LIRs are ISPs, hence ISPs are making policies.
ISPs like to keep their customers. Saying to a customer they'll have to go 
through the pain of a renumbering process is a good argument to keep 
anyone aboard their network. :-)
1+1=2

Then there is "the routing table size problem" -- if everyone gets its own 
PI, some routers will not be able to keep up... but that's also a danger 
on the IPv4 world... if someone wants a PI, it's only a matter of money 
and imagination. :-)
</probably political incorrectness>


> Surely, you'd have to recitate RIPE rules and deny the request for PI - but
> what would happen to your relationship towards your potential customer ? Can
> you afford dropping them - multiple a year ?

Surely not. :-)
If ISP1 doesnt find a way to adapt the customer's situation to the rules 
in order to please the customer, ISP2 or ISP3 gladly will...


> Making v6 PI available may overload memory of current routers. And I'm the
> last ignoring this matter.

What is a "current router" ?
My ASN uses a Juniper M10, a Cisco 7206 and a Cisco 12410 to deal with 
BGP4+ (about 175k v4 routes + about 500 v6 routes). Every one of them 
deals with IPv4 and IPv6 routes, if i may add.
Some years ago we used two Cisco 75xx. I dont really know which equipment 
will form my "border router team" 2-3 years from now...



> There IS a limit any ISP can effort expenses on equipment.

Surely. And if you provide a better network, you should mirror the costs 
of a better network to your customers...


> At least in
> Germany by the current market structure I'm glad of anybody NOT signing an
> end-consumer DSL contract. The 'the-cheaper-the-better' mentality is sooner
> or later killing all of us. By a 20 month ROI for a single DSL where do I
> get the money for upgrading routers from ? By not offering IPv6 so I can
> keep with my current routers ?

Last 4 year experience says you can deploy IPv6 with minimal costs. Just 
use simple engineering... :-)
If you don't want/cant to mess out with the IPv4 routers, just use a cheap 
alternative (quagga/zebra/...)


> Enterprise customers are asking (seldom
> enough, but they do) for Multicast and increasingly for v6 in addition to
> MPLS services - but these are the minority.

Luckly, we would expect most of the multicast thinggy to be intra-domain 
:-)


> Where's the point the average ISP is eaerning money ?

The point where it satisfies customer demand, without charging less for 
the usage of their infrastructure?


> Many of you (I've had no sense for looking up each one's approx. size
> specifically) sounds like one of these german (wanna-be)-monopolists: we're
> that large, customer have to come to us. You deny any small ISP their right
> for existence.
>
> Maybe there's need for an organized structure of the internet. Let's get the
> routing and address design be dictated by the known tier 1's & maybe 2's.

That's even a dangerous path. ISPs also want tier-1/tier-2 independence! 
:-)))



> You want that ? I think the mayority of attendents of these discussions do
> (as none of the smaller is able to effort time for RIPE-WG's, DECIX-WG's,
> AMS-IX-WG's etc...). When I get time for these I follow and sometimes
> response to WG's, but that's neither part of my job nor paid.

Strongly agree. :-)


> You'd prefer a 200 pseudo-customer enterprise above a 100-customer ISP.
>
> To my personal opinion there're already too many non-ISP AS'es out there and
> much more - sadly mostly ISP - announcing nonsense (Gerd, I appreciated the
> discussion about aggregation some time ago) but HOW do you want to
> distinguish ? Maybe RIPE - and/or becoming a LIR in general - should be
> closed for non-ISP ... But where to draw the line ?
>
> Therefore I think we should go for IPv6 PI - with some constraints:
>
> - PI space (no matter whether v4 or v6) should be charged an annual fee
> - AS numbers should be charged an annual fee
> - a minimum size for PI assignments should be established

That is a possible set of ideas, yes. You can also raise the question of 
charging or not to your customers when announcing their PI prefixes. 
Thruthfully, when an ISP announces an extra PI prefix, its peers will 
have to accept it or not. 
Usually they do, but how about if they asked: "Hey, what do i gain by 
accepting it? I only agreed to receive your PA prefixes in the first 
place..."


> I once got assigned a /24, /26, /27 of v4-PI. This satisfied the request
> made but when asked for routing we had to deny this as longer than /24 is
> simple pollution of the routing table.

Is there an authority that can regulate what is pollution and what is not? 
:-)


> Why not go for v4 PI always being a multiple of /24 (if at least /25 is
> explained - any longer is nonsense by itself) and v6 being at most /40 (out
> of a 'dedicated' /32 or shorter) with more stringent criteria ?
>
> I always have a hard thought about equipment needs and cost - but do you
> remember the primary reason there's such a variety of ISP ? This reason are
> the variety of customer needs and the CUSTOMER's will.

Sure. And often *one* customer will get a better service from an ISP with 
a 50 customer base, than *one* customer on an ISP network with a 5000 
customer base... unless its monthly numbers enter the TOP-whatever in the 
ISP's accounting dept. :-)


> regards,
>
> Marcus
>
> PS: I'm sorry for the mix of thoughts above ... just had a '42' day .... :-|

Found your e-mail to be very useful, and very, very market-wise.
Thanks!


Regards,

./Carlos
--------------

          Wide Area Network (WAN) Workgroup, CMF8-RIPE, CF596-ARIN
FCCN - Fundacao para a Computacao Cientifica Nacional  http://www.fccn.pt

  "Internet is just routes (175261/555), naming (millions) and... people!"