On May 3, 2010, at 10:25 AM, Chris Engel wrote:

> On the network level, I basicaly want something that entirely abstracts my 
> internal architecture from my external advertisement of services...and 
> essentialy functions as a proxy/intermediary between my internal devices and 
> thier external presence at the boundary between internal/external. NAT very 
> handly does that currently in IPv4. From the discussions that I've had with 
> alot of people involved with IPv6...and many of the people who have strongly 
> argued against any sort of NAT in IPv6... they basicaly seem to be 
> disagreeing not just with the particular method I want to use....but with my 
> end goal itself.

I would agree that many in the IPv6 community disagree with your goal. Their 
fundamental objective is end to end transparency, and your fundamental 
objective is end to end opaqueness. You disagree with each other. I understood 
from your previous emails in this thread that you felt they were "wrong" or 
that you felt that they felt you were "wrong"; I don't think either is "wrong" 
per se, but you certainly disagree.

Personally, I think the term "NAT" in the moniker for this draft was 
particularly unfortunate, and said that to Margaret when I agreed to co-author. 
Since we are talking about something fundamentally different than IPv4/IPv4 
NAT, we have the same issue that has been discussed with the term "realm". We 
wind up with this discussion, which is not a stupid discussion, but is off 
topic with respect to the proposal in question. When the IETF had a BOF on the 
topic, that was incredibly clear. We had a proposal on the table which was most 
decidedly *not* re-implementing IPv4/IPv4 NAT in an IPv6/IPv6 world, but the 
entire discussion from the floor was "we don't want to re-implement IPv4/IPv4 
NAT in an IPv6/IPv6 world", largely non-responsive to the topic at hand. It 
winds up consuming a lot of cycles, is very emotional, and drowns out 
reasonable discussion on anything else. Take a look at this thread as an 
example. We have collectively exchanged 39 messages (this is the 40th), of whic
 h perhaps half a dozen have been on topic and the remainder have been related 
to re-implementing IPv4/IPv4 NAT in an IPv6/IPv6 world, something which is 
specifically not the topic of the list.

Regarding NAT, though, in http://www.ietf.org/rfc/rfc4864.txt, some strong 
proponents of the transparent model addressed what they thought were the 
primary reasons for the use of NAT. They didn't cover this one, so to speak. 
But it might be worth your while to acquaint yourself with their viewpoint. In 
short, NATs not only achieve a certain level of obfuscation of the network 
layer, they make it harder to build applications of a certain class - 
applications in which the address of one's peer must be known at the time one 
wants to use it. Real time applications, such as VoIP and Video/IP, fall in 
this category. More generally, the entire realm of peer-to-peer applications 
does. The imposition of NAT has not altogether stopped such development, 
though; what it has done is make it a lot harder, resulting in companies like 
Napster, BitTorrent, and Skype building byzantine overlay networks to overcome 
the issues it presents. In the realm of voice and video, it has engendered RSIP 
 and the use of SIP Proxies as gateways between domains. NAT hasn't actually 
protected any networks, I will argue; it has merely made scaling the walls a 
little harder. 

The goal of those that would like to not have any translation at all - here, 
Keith, I'm putting words into your mouth, so please feel free to correct them - 
is essentially the counterpoint of that. The world is poorer for the 
applications that have not been developed due to firewall and NAT boundaries, 
and they would like to be able to build interesting applications that don't 
need the byzantine structures to bypass what they consider to be ineffective 
and deluded security architectures. If using NAT to flatten the addresses 
expressed at the DMZ has been ineffective in preventing attacks (the vast 
majority of which come from behind the firewall anyhow) and have made 
application development more difficult, what's the point? They would rather go 
with the end to end principle as stated by Saltzer 
(http://web.mit.edu/Saltzer/www/publications/endtoend/endtoend.pdf) and enable 
users to make the Internet and the applications that use it more valuable.

For my part, and the part of this proposal (and the ILNP proposal that RRG is 
putting forward), I am all for the free development of useful applications. I 
don't see the value of NAT, given its history of not stopping attacks and of 
preventing applications. I see a problem that needs addressing though, related 
to management of the backbone route table. In short, we now have O(10^5) 
prefixes in the IPv4 route table, and perhaps two years from now will easily 
have O(10^6) as people buy and sell prefixes. As an equipment vendor, I have no 
problem with that - if you'll pay for the memory, I'll happily sell it to you. 
But I don't want to hear about either capex or opex, I don't want to hear about 
"green" aspirations, and I don't want to hear about the price of the equipment. 
You make the bed, you lie in it. As we collectively move to IPv6, we have a 
choice. We can replicate the IPv4 swamp, complete with its route table, or we 
can change it. I vote that we change it.

In the area of addressing, networks at the edge are pushing back very hard on 
the provider-allocated addressing model, and are driving for 
provider-independent addressing. Reason: they don't want to be captives of 
their providers. I'm sympathetic with them. But consider the implications: if 
they all have PI addresses, we are enumerating the objects at the edge, and 
even today we have O(10^7) or more objects at the edge. PI addresses take us in 
the direction of a swamp at least as bad as that in the IPv4 network. Service 
providers find the PA model very attractive, both because it provides a market 
lock on their customers and because it helps them manage their route tables.

>From my perspective, I would like to achieve some of the goals of all of the 
>players. I obviously can't both lock and unlock customers, so whatever I do is 
>wrong there from someone's perspective. But by Network Prefix Translation, the 
>subject of this list, I can give edge networks the appearance of PI (they are 
>independent of their providers for addressing) and the service providers the 
>appearance of PA (they enumerate the ISPs at the edge, O(10^3 to 10^4) instead 
>of objects at the edge, O(10^5, 10^6, or 10^7)). The boundaries are 
>address-translucent, not either transparent or opaque, so while I don't give 
>you what you want I give you something a step in its direction. And I give 
>Keith what he wants - if he can find it in his heart to have applications use 
>names instead of addresses, he gets all of the access between applications 
>that he needs. And for those that operate networks, it reduces their capex and 
>opex. It requires everyone to think a little differently, but if they
  are willing, they get most of what they are looking for.

I hope this is useful.
_______________________________________________
nat66 mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/nat66

Reply via email to