-----Original Message----- From: Bound, Jim [mailto:Jim.Bound@hp.com] Sent: Saturday, April 08, 2006 5:52 AM To: Tony Hain; PPML; address-policy-wg@ripe.net Cc: Richard Jimmerson; Latif Ladid ("The New Internet based on IPv6"); Davis, Terry L; ollivier.robert@eurocontrol.fr; narten@us.ibm.com; Brig, Michael P CIV DISA GES-E; Pouffary, Yanick; Green, David B RDECOM CERDEC STCD SRI; Bound, Jim Subject: RE: Question
Tony,
Excellent response and educational for sure. It is my belief that the corporate business model today for operating networks may be broken and I think you supported that below? If not my apologies for bad
Jim/All I am going to respond in two parts here on PI issues; one in terms of aviation and one in terms of corporate. This one is on aviation. The next two paragraphs are from an original response to Thomas Narten, that I didn't see make the list. ---- I view systems that run "critical infrastructure" entirely different from those used to run anything else; especially systems that can directly impact the safety of the people using or relying on them. Safety engineering is just like security engineering; both depend on our ability to build in layers of defense and reliability trying to never rely entirely on a single system. By forcing an industry like aviation to accept the potential of address changing in a global fleet, an element of extreme risk is added as the system's overall reliability is decreased. ---- We know that in the next decade that there will be development initiated for a new air traffic control system. It will likely be built upon IP and if so, likely IP-v6. And ICAO currently has a working group studying this and the committee is leaning towards IP-v6 although there is a strong component that is pushing for IP-v4 and a continuation the NAT type usage currently required in the aviation industry by Arinc 664. And I do definitely agree with Jim here, the use IP-v4 and NAT would create huge risks; if in nothing else, the potential for mis-addressing through one of the hundreds of NAT gateways that would be required. I'll respectfully disagree with Jim in that I believe address change in a complex global system like air traffic control can create a hazard. Keep in mind, that the air traffic control system spans virtually every nation on globe and most everything manmade that flies. Likewise the technical and operational capabilities vary from extraordinary to very minimal; like the 30 or so aviation operators that the EU just banned from flying into EU countries because of their poor safety and maintenance performance record. Coordinating an address change across this type of infrastructure with aircraft and ground infrastructure in almost every nation on the globe, is simply beyond my ability comprehend. Assuming the technology would work flawlessly (discussed below), the politics of when and how to implement the change would likely end up on the floor of the UN for debate. Likewise, if a decision was made to implement a change, we would be dealing with such different levels of expertise around the world that no amount of pre-planning could ensure that implementation failures would not occur. Now just a bit about where ATC systems are likely going and why their criticality will likely grow over the next couple decades. Unless we suddenly develop anti-gravity capabilities to allow slow vertical takeoffs, we are stuck with the airports we have and only minimal abilities to expand them (cost, environmental, noise, etc). The only real way we can expand their capacity is with bigger airplanes and more flights. The "more flights" part is where this gets complicated and critical. To handle more flights, we have to decrease landing and takeoff separations and speed up aircraft ground movements so an airport can handle more aircraft per hour. We are about to human capacity with the current systems which means that these improvements will need to move more and more to relying on precise control systems; a minutes interruption here will be a really big deal. Also we as an industry are just beginning to migrate from bus data communications on the aircraft to networks. The commercial aircraft flying today are already largely computer controlled and as I mentioned above we try very hard not design the aircraft to be critically reliant on any one system. In almost all cases, it requires a cascading series of failures to present an aircraft with a catastrophic hazard. Now as I said, we are starting to put networks on the aircraft and as Arinc 664 shows; we are not the world's greatest network engineers (at least not yet..). In a decade or so, we will have hundreds of networked systems on an aircraft. I think the risk here in re-addressing is clear; how well will they all react. And yes we can probably take most of the risk down in certification testing but keep in mind variation in technical competence of the operators around the world and that we are continually accepting upgraded systems from our vendors as replacement parts and this could also inject potential failures in re-addressing. If we were to use 3178 without a single global address space, I still don't think this would scale as we then would be using probably in the neighborhood of 50 or more ISP's (you don't always get to pick your ISP's and while a country might accept addressing from an industry block, they'd probably insist on using theirs otherwise) around the world for the service. And the way I read it, I would still have lots of unnecessary backhauling to the other side of the planet and some very complicated policy routing to set up. Besides and then with mix of address spaces, I would probably be perpetually leaking with the global Internet in what should be a closed network. Finally at the moment with our existing certification processes, I'm not sure that we would even be permitted to change the aircraft addresses without re-issuing all the affected software with new part numbers. (I'll bet you assumed we used DHCP to address the current aircraft; nope we hard code address everything, remember "bus engineering" 101 ;-) With today's current rules, we haven't put any "critical systems" on anything but a closed onboard network. We are just discussing the ability upload new IP_tables/firewall-rules and authentication certs/passwords to the non-critical networks and I believe that this will be solved in the next couple years. And now also keep in mind that every aviation rule-making body around the world would also have to approve of the address change for an ATC network and define how they were going to certify the change. ====================================================================== Finally now having said all this Jim, I think it is possible for aviation to remain conforming. We have probably only two primary needs for stable IP addressed networks; one for Air Traffic Control and one for Airline Operations. These are industry traffic type designations that have safety related functions that are carried out over them. As we have discussed before, I expect both of them to be run as "closed networks" and should never (IMHO) be seen in the global routing tables; a closed network will provide them with a layer of security, better routing performance, the multi-homing that an aircraft needs, and more options for mobility solutions. Further, two organizations already exist that could legitimately hold the addresses; ICAO for the ATC network as they already govern it and the AEEC for "airline operations" whose members already essentially own "Arinc" which is an ISP already. If it were possible to convince these orgs, to apply for space and the registries to grant them, that would seem to be a solution. Take care Terry PS: Apologies for the length.. PSS: Back to "critical infrastructure" networks a moment, I'd say that any network that wanted to declare itself "critical infrastructure" could obtain PI space, BUT to me this type of network should always be run as a "closed network" with exchanges to the Internet only through "mediation gateways" operating at the application level, not at the routing level. Just food for thought but perhaps there is a class of IP-v6 networks for "critical infrastructure" that have their own PI space, but are prohibited from the participating in "Internet routing". Such a concept might solve lots of problems. parsing?
Their models were fine for an IPv4 world where NAT was required and
some
even confuse NAT with securing ones network (and some programs in the U.S. Government) and that is simply bad policy and view.
In the interim can this be resolved by RIRs creating some kind of additional wording that address reclaim will be done in manner that is negotiable, and do no harm to corporate or government business operations? This would buy us time to work on the issue and stop the FUD around this topic?
Also I am willing to sponsor a world wide IPv6 Forum BOF on PI and addressing you can lead as ajunct to one of our regular meetings you can lead for an entire day and we get the right players in the room. So think about that as another option too.
But do enjoy the beach this thread does not have to be resolved this week :--)
Really want to hear from all of you and discussion Terry D., Latif, Yanick, Dave G. Mike B. etc.
Thanks /jim
-----Original Message----- From: Tony Hain [mailto:alh-ietf@tndh.net] Sent: Friday, April 07, 2006 7:57 PM To: 'PPML'; address-policy-wg@ripe.net Cc: 'Richard Jimmerson'; Bound, Jim; 'Latif Ladid ("The New Internet based on IPv6")'; 'Davis, Terry L'; ollivier.robert@eurocontrol.fr; narten@us.ibm.com; 'Brig, Michael P CIV DISA GES-E'; Pouffary, Yanick; 'Green, David B RDECOM CERDEC STCD SRI' Subject: RE: Question
A public answer to a private question as I have been sitting on a beach for awhile without the laptop and missed some related conversations ... :)
Is the outcome really open for discussion on the PI issue? It doesn't sound like it is.
In the minds of some the route scaling issue outweighs any argument for PI. When taken to its extreme, there is a valid point that a broken routing system serves no one. At the same time the dogmatic stance by the ISPs enforcing lock-in is just as broken both for large organizations with financial or legal requirements for operational stability, and the individual consumer/small business with limited budgets looking for true competition. The hard part is finding the middle ground in a way that limits the exposure to a potential routing collapse.
I personally refuse to declare some needs legitimate and others not, as the only point of such differentiation is to establish a power broker. When all uses are legitimate, the problem boils down to the technical approach that can be scaled as necessary to contain growth in the routing system. This is the logic that leads me to the bit-interleaved geo that can be aggregated in varying size pockets as necessary using existing BGP deployments. We can start flat and implement aggregation over time when a region becomes too large to handle. One nice side effect of this geo approach is that it mitigates the continuing political demands for sovereign rights to IPv6 space.
Any aggregation approach will force the business models to change from current practice. That is not as bad a thing as the alarmists will make it out to be, because their accountants are claiming the current model is a broken money looser as it is (which if so means they will eventually change anyway). The primary difference is that there will need to be aggregation intermediaries between the last-mile and transit providers. The current model eliminates these middle-men by trading off their routing mitigation service against a larger routing table (actually they already exist in the right places but are currently limited to layer2 media aggregators). The anti-PI bunch is trying to use social engineering to directly counter the bottom line business reality that the customer will always win in the end. Rather than accept this situation and constructively work on the necessary business model and technology developments, they effectively stall progress by staunchly claiming there is no acceptable technical approach that works within the current business structure.
Making the RIRs be the police deciding who qualifies for PI and who does not just adds to their workload and raises costs. The beneficiaries of this gatekeeper approach are the ISPs that claim they need full routing knowledge everywhere, while the cost burden for supporting the waste-of-time qualification/evaluation work is borne by the applicant. Given that the most vocal and organized membership in the RIR community are the ISPs it is easy to understand why it would seem like the PI issue is already decided as closed. I tend to believe it will just drag out until enough of the corporate world becomes aware of the IPv4 exhaustion in light of their growth needs that they collectively appear at their RIR and demand an immediate solution. Unfortunately this 'wait till the last minute' tactic will likely result in a reactionary quickie with its own set of long term side effects.
A while back I tried to hold a BOF on geo PI in the IETF, but was told that shim6 was the anointed solution. Now that at least nanog has told the IAB where to put shim6 it might be possible to get the current IESG to reconsider. In any case the result would be a technical approach that would still require RIRs to establish policies around. As long as they are dominated by the ISPs it will be difficult to get real PI.
Tony