[address-policy-wg] Policy Proposal 2006-02 (IPv6 Address Allocation and Assignment Policy)

[Bill, see somewhere half way through this rumble, a Q about why
3ffe::/24 is still alive and announced ;) ]


JORDI PALET MARTINEZ wrote:

>>>> "The assigned prefixes must be longer than the one allocated by RIPE
>>>> NCC." is useless as the LIR can't pass a shorter prefix down as they
>>>> don't have been assigned that portion.
>>>> <----- white space is useful
>>> The goal of this part of the text is to avoid that a prefix is received and
>>> assigned totally to the same customer.
>>> <----- white space is useful
>> So you want to avoid a /48 to be allocated to an organization who can
>> then use it at their own site? What is the use of changing that then?
> 
> No, I want to avoid if an LIR get a /32 (minimum allocation as per existing
> policy) that it allocates the complete /32 to a single customer.

Why would you want to avoid that? Maybe they can justify the need for
it? Most sites won't be able to do that though, it is a bit too much.
But as you like looking at other regions, in those other regions several
large corporations (that is without primary objective being ISP) have
received an allocation. Do you really think that they are going to play
ISP for other companies? As you want to equal the policies out,
according to your proposal, did you take this in consideration with the
above line? Also they still have to justify this to the RIR, so why is
this line needed again? It is just an artificial barrier.

>>> So yes, can't pass a shorter prefix,
>>> but could pass the same one and this will not be good as it can be a kind of
>>> "bypass" of the rule in order to provide an alternative way for doing PI to
>>> customer, which is not the intend of this proposal.
>>> <----- white space is useful
>> You want to avoid "PI" but you also state in the proposal to make the
>> policies the same as other RIRs. Bit strange as they don't have these
>> changes, they have a separate PI policy.
> 
> Out of context. What I'm saying is that other RIRs (LACNIC and AfriNIC)
> don't have already the requirement for 200 /48s, and I think is ridiculous
> for us to be more strict, especially when this is an artificial limit.

If you don't have even a PLAN to ever use 200 of the /48's out of the
65536 that you will be assigned, then what is the use of getting a /32
of address space?


If people only require a /48 then give them a /48, if they require a
/40, give them a /40. Don't waste address space; yes there is enough,
but throwing it away is futile too.


>>>> The "must be advertised" part causes any non-internet usage to be 'illegal'.
>>> Yes, this has been already raised before.
>>> <----- white space is useful
>> You stated that all the previous discussion was nullified and that all
>> arguments should be raised again.
> 
> No. What is not useful here is the "yes I'm in favor" or "I'm against" for
> the previous version of the proposal. Discussion is always useful, and in
> fact, this version tried to accommodate all the discussions in the previous
> version.

Saying "yes" or "no" without argumentation is useless.
And I've made enough arguments to show why.

[..]
> I clearly stated that I'm not saying all are valid, but one among these 3

Why name something when you know it is not valid or even relevant?

[..]
> No need to renumber, they just need to use one more prefix (a global one),
> and keep using the existing ULA addressing if they still need it for
> internal routing, keeping separate overlay networks, or whatever.

Ever ran a network larger than ~25 computers?

Multiple prefixes on the same link leads to a lot of issues that a lot
of people want to avoid. It is a great idea, it is great that the
possibility exists, but it is not so great in practice. If having those
multiple addresses, in multiple prefixes was easy to do then there would
not be a huge multihoming discussion, people would be using it don't you
think?

Ever tried running multiple prefixes of the same host and making sure
your source address selection is correct on all the platforms that are
in a large business!?

> I think that somebody who plans adequately, will perfectly know if they need
> a global prefix in one year term, and they can announce it even if they
> decide not to allow incoming/outgoing traffic to the rest of the network.

So you do think that planning is a good idea? Then planning for 200
/48's in a LIR should not be a problem would it?

> The proposal is not asking to explicitly make all the internal network to be
> visible, just to announce the prefix.

Announce it where? If it is 'announced' internally it might just be a
static route. Really the RIR's have nothing do with with routing,
Network Operators take care of that. As I mentioned before, they decide
what gets blocked, announced, accepted etc, it is their network, not the
RIR's. If they get payed enough or deem a site important enough they
will route the prefix, how many times a RIR will say they can't or not.

Same goes for protocols, proto-people don't like bittorrent, but
everybody uses it, that is the freedom of the Internet.


> Also we are not asking RIPE to verify to every RIR every detail that they
> need complain with every policy at every minute. Some flexibility is
> reasonable, and that provides in most of the cases, extra time to comply
> with the policy, without any impact in the community.

You actually are saying what (as you mention courts in your proposal)
every court will dislike: that a RIR can, because of this wording,
randomly pick out a LIR and pick on them whatever they want.
Now that is illegal business practices.

>> Most networks will end up being connected to the Internet or to another
>> network one day or another, being able to globally route it is thus a
>> requirement for quite some organizations. ULA's thus don't cut it.
>> Unless you want to see those popping up in a BGP near you.
>
> Filter them, in fact this should be a strong rule already in place in every
> router, same as per link-locals, etc.

That is why the 6bone (3ffe::/24 that is) is still reachable at most
sites. And also why there are people having loads of problems with
69.0.0.0/8 (See http://69box.atlantic.net/).

The C&W folks, who got the first /20 can also tell you how long it took
them to get their perfectly valid prefix unfiltered. And the IPv6 BGP is
only comprising of a relatively small number of ISP's at the moment.

Filter is NOT good. It will hurt, not today, but it will tomorrow.

[..]
> So what is the problem then ?

As I mentioned, thus again: the sentence is useless, take it out.

> If we actually have a problem on this, let's take it step by step, with
> another policy proposal. I learn the lesson that trying to make very complex
> changes in policy proposals is a "no way".

And you know very well how long it takes to make a proposal and to get
people to more or less agree on it. If the old one was 'good enough' and
'the same' then why did you change the wording?

>>> As explained in previous emails, if we want to change that, it will be
>>> better to do in a different policy proposal and have a different discussion
>>> about that, otherwise it may become difficult to achieve consensus in the
>>> main change here (removing the 200 /48).
>> <----- white space is useful
>> This proposal changes the wording, you can then just as well also
>> completely take out that part instead of fumbling with words.
> 
> I will be happy with that, but the consensus that I perceived is that the
> community don't like that.

Your perception of something is not the final answer.

> They want to have at least a plan, and I think is fair.
Funny that you say that 'having is a plan is fair', while the whole
point of this proposal is to remove having a plan, as, according to the
proposal, that would be bad in the courts. Which court btw? Do remember
that there is no such thing as an Internet court.

The plan is anyway implicit, if you don't have a need for address space
then you should not get it in the first place as you don't show demand
to actually use the address space.


The 200 number is there solely to delimit between organizations getting
a /32 or ones who are only endsites and should get a /40 or a /48.
It might be out of thin air but it is a reasonable measure. Maybe a
better number for getting a /32 is actually 20.000 sites, as then one is
actually filling up that /32 of address space. It is only a PLAN. Nobody
will enforce it, nobody can enforce it (except for other ISP's, if they
don't like you they won't accept/route/transit your prefix, if a RIR
assigned it or not)


> I think we must rely on RIPE NCC staff to accept the plan or ask for
> further justification if its fuzzy.

Which is what they are already doing and have been doing for years.


>>>> A way to stop it would be requiring S-BGP but that will not be done for
>>>> the coming years. Also most likely S-BGP will still allow the owner of
>>>> the certificate to announce more specifics.
>>>> <----- white space is useful
>>> Yes, this is a possibility, but I think there is a more realistic one. Let
>>> market decide, if you break aggregation and as a consequence this cost money
>>> to the rest of the ISPs, some of them may decide to filter strictly on
>>> allocations and not accept more specifics. This already happens today.
>> <----- white space is useful
>> If "let the market decide" is the answer then there is a MUCH easier
>> one: just take 8000::/3 and start picking random blocks from there. No
>> need for RIRs to be involved, saves a lot of hassle, and as long as you
>> are large enough (content wise, user wise, money wise) you can get that
>> prefix to be accepted anyway. Same goes for DNS and all other internet
>> resources, they are just numbers. The biggest one wins.
> 
> I think you are being too extreme. What I'm saying is something already
> happening and not being a chaos. What you propose, will be a chaos :-(

I don't propose that (did I fill in some form somewhere?) it is what one
could do if one really wanted to.

If, say, Google would take 198.18.0.0/15 and start serving YouTube and
Gmail from it, stopping access over IPv4, I am very confident that a lot
of ISP's will start routing it for them. Same goes for IPv6. The RIR's
have no power over the routing system, and there is also nobody who can
sue anybody over it. (Fastweb using 23.0.0.0/8 and a number of others
for instance)


>>>> The whole more-specifics business depends on one factor: Money
>>>> If you pay people enough or it is important enough one can announce it
>>>> anyway and no RIR is going to stop that.
>>>>
>>>> Including that portion is just a lousy way of trying to inhibit routing
>>>> table growth which will not work; as can be seen by the multitude of
>>>> longer prefixes being announced into the global IPv6 routing tables
>>>> already. (*)
>>> <----- white space is useful
>>> Having the text in the policy helps to take measures if needed.
>> What can a RIR do against it? Say that the LIR can't use the prefix
>> anymore? How will a RIR do that exactly? Tell them they are bad and
> 
> SBGP is not enough ? Some (just a few) LIRs filtering the prefix will make
> it unusable.

No, because a LIR is typically a leaf. Now if a *transit* would filter
it, then it would wreak havoc. But fortunately transits get payed by
their customers, who are the leafs, thus if they complain hard enough,
or simply not pay, then it will be restored again.

Again: there is no single power on the Internet. For that matter, there
is no single "Internet". The "Internet" is what people perceive as
useful and try to use.

Next to that, prefixes can also be used internally of course, which is
why I mentioned that the 'announce' clause is useless too. Announce it
where? on the local IX, which is that WRT54GS in the cupboard connecting
me and my neighbors? It is an IX, as they have IPv6 /48's and we
interconnect and we are also multiple parties.

>> evil? Come on, 3ffe::/24 is also still announced and that is now IANA's
>> block again and in effect 'illegal' to be in use.
> 
> I don't see those prefixes, are filtered out, and many people do the same, I
> guess.

Not enough people, see GRH.

> So not an issue, not useful for who is using those prefixes.

If it is not useful, why does somebody then announce them? Bill can you
give an answer? :)u

> I think
> right now only 1 entity in the world in just 7.5 months since released, not
> bad, actually it took less than 3 months as I recall, to get only 1 entity
> announcing it.

Because I and some other people have been actively bashing people to get
it out of their routing systems. A lot of places where not even AWARE
that 6bone was going away, even though it was very clearly announced.
Oh, before you ask, even people holding and announcing pTLA's didn't
know about it; a large amount of the folks who shutdown after 6-6-6
where in that camp. Why? -> they didn't care less or simply forgot all
about it.


>> [Bill, did you see the black helicopters around your house already? :) ]
>>
>>> Not having
>>> that text will not help at all, while keeping the text provide a possible
>>> way for the community to take actions, if required, and meanwhile doesn't
>>> hurt.
>>> <----- white space is useful
>> ISP's are already doing it and nothing is being done against it.
>> They will and are creating their own PI without the 'community' being
>> able to do anything about it: except filtering, which is not happening (yet)
> 
> I can tell you that some are getting filtered. Even people receiving /48 PI
> from ARIN has issues, as they told me two weeks ago. May be they will be
> lucky and will get sorted out, but filtering works.

So here you say yourself that filtering is hurting people, but you still
want people to filter as you mention above!?
Quite a bit of a contradiction don't you think?


>>>> Lastly "# have a plan for making assignments within two years.",
>>>> everybody has a PLAN to do something, actually doing it something else.
>>>> As the number of assignments is removed, there is no way to check up on
>>>> this, next to there not being any requirement of actually registering
>>>> these assignments in the RIPE db, thus there is no way to check.
>>>> <------- this one was missing which made you confused
>>> The point is not asking LIRs to lie about the "size" of the plan. Some LIRs
>>> could have a plan for just a few customer, and actually if they say the
>>> true, they will not get an allocation because they don't have a plan for 200
>>> customers. So they could tell to RIPE that they have a plan for 200
>>> customers and get the allocation. This is completely artificial.
>> If it is artificial then why is it included?
> 
> I'm not sure what text are you reading. No longer 200 included. That was
> artificial and we need to get rid off it.

Mind the quoting, all the 200's where your text, I only wrote a single
line: "If it is artificial then why is it included?" I suggest that you
don't fumble the quoting by removing white space between the replies,
that saves on this muttering. Note the line with "<---" for clarity.


>> If they only have a plan for a few customers then they should say "we
>> only need a /40" to RIPE and should be able to get that /40.
> 
> Not under existing policy. And if we agree on that, it will need to be in a
> different policy proposal offering different allocation lengths depending on
> the number of possible customers, or whatever.

Then can this proposal and write one which actually makes sense.

> I'm not saying yes or not,
> just take each thing step by step if we want to move after so many years and
> so many failures and work of so many people to remove this artificial
> barrier.

Doesn't work if there are so many wrong things with this proposal, and
the things I mentioned you still have not addressed.

>> Then again, address-space wise there is not much wasted in giving a /32
>> to everybody who wants it. (16^2 is a lot :)
>>
>>> I don't believe anyone NOT willing to provide services will ask for an
>>> allocation.
>>> <----- white space is useful
>> Why not? Maybe some company just wants address space to be able to say
>> they have it. Just check the current allocations which are allocated but
>> have not been announced in the last couple of years.
> 
> People is getting ready, it takes time, and priorities change, but they will
> use it. If not, following policy, could be reclaimed, if the community
> decides to do so.

People _are_ getting ready, but more because they have to to win their
government contracts than that they see an actual need.

Reclamation does not happen. The cost is too high and the winnings are
too futile. Especially for IPv6 this is the case, at least for the years
too come till we run out of 2000::/3 ;)

>> Maybe they just want to secure their address space for later, how can
>> you know, it is their company not yours.
> 
> I don't think this is a general rule, and actually I think it is easier for
> small ISPs to start providing the service than for bigger ones. Smaller ones
> tend to me much more "agile".

What you are actually saying with the above is that a small company
doesn't need to have 5 layers of management approve things.

A large company can also get address space now, set up a tunnel broker
internally, or use ISATAP or a similar technology and start providing
IPv6 to their clients without much of a problem.
Which is actually how a number of large ISP's solved their IPv6 issue.


>>> RIPE can check that at any time by asking customer contracts.
>> And then they show 2x /33, one for their games department and one for
>> their work department. Doesn't really help does it?
> 
> No, existing policy ask for justification if you assign more than /48 to a
> single customer, so it will not work, even they will be in troubles before
> doing the assignment because RIPE NCC will not accept a justification for a
> /33 for their games department unless they demonstrate that a /48 is not
> enough.

Then why remove the plan if they can't use the assignment anyway?
This also comes back to the "do not assign larger than own prefix" line,
if the RIR can decide then why include it?

>> As mentioned even if RIPE then finds out that it does not qualify, then
>> what are they going to do?
>>
> 
> Reclaim the space, check with the community, whatever.

"whatever", indeed. Have you ever seen address space being reclaimed?
Ever tried to calculate how much effort it would be?

> We should be doing the same with IPv4, and may be sooner or later we will end up doing so. But
> if we don't have a way to "secure" it in the policy, if time arrives could
> never do that.

Did you ever read slides from Geoff Huston or various of the other
people talking about IPv4 address depletion and that reclamation is more
trouble than it is worth?

>>> I thought all the /48s assignments need to be registered (I may be wrong).
>>> So this is not changed by this policy proposal.
>> <--- white space is useful
>> Clearly not as there are enough allocations that don't have a single
>> assignment but they are announcing the whole block.
> 
> Ok, but then take it as a different issue, and as a member of the community,
> ask why LIRs don't do that, and if there is no need for it remove that part
> of the policy, etc. 

Simple: they don't want to put in a public database who their customers are.

> As said, let's work step by step.

Then the first good step is to can this proposal and write a proposal
which makes sense. Argumentation in the various emails I, and others,
have already sent about this.

>> Please check BGP before you try to make up a policy.
> 
> :-)

Smileys don't help, have you already checked it by now?
http://www.sixxs.net/tools/grh/ has a lot of information for you.
And of course don't forget to check http://ris.ripe.net which is also a
great service to get at least a little bit of background in these matters.

Greets,
 Jeroen


-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 311 bytes
Desc: OpenPGP digital signature
URL: </ripe/mail/archives/address-policy-wg/attachments/20070319/8ab7bf9c/attachment.sig>