At 03:53 27/11/2004, Valdis.Kletnieks@xxxxxx wrote:
On Sat, 27 Nov 2004 02:33:54 +0100, "JFC (Jefsey) Morfin" said:
> But why to spend time and money and to take risks to change something which
> is not broken. IPv6 has no problem in keeping the same host numbers if the
> used addressing plan uses a numbering scheme designed with that purpose in
> mind, like the telephone numbering scheme. You change of telephone
> providers - or use several at the same time - without changing number.
That's because the phone number is more akin to a DNS name than an IP address.
I'm pretty sure that if you investigate the insides of how the telco system
makes that transparent provider-change work, you'd not be as interested in
using it as an example. (Or did you *want* to go back to the days when routing
tables were shipped around and installed on the fuzzballs twice a week? I got
tired of *that* back in 1984. And back then, the routing tables were only a
few hundred lines long, not the 150K routes we have now...)
Dear Valdis, OK, I will document it more.
Back in 1978 when we introduced the structured naming on the public international packet switch network, we based it upon a routing technology separating addresses from routing. This is why *since its inception* I am tired by dumb table routing. As Harald puts it: IETF is to maintain IPv4, develop IPv6 and imagine mobile IP. I submit these are totally different problems than to analyze and build from an existing solution, and to analyze a need and design a new solution. So, please let use the past as a bank of experience, but let not talk of "going back" when referring to analyzes also used (and tested) in the past. We simply know them better.
Call establishment needs four informations: - (a) where the caller calls from - (b) how the caller identifies the callee in a directory of called addresses - (c) how the callee designates himself in a directory of called addresses - (d) routing information from caller to callee
From Internet past usage (b) is named "name" and uses structured "0-Z" "numbers", (c) is named "IP address" and uses "0-255" "numbers" also expressed in 0-9 and (d) uses various information sources and systems (DNS, header of the address, routing tables). This results from an optimization and network principles based upon Louis Pouzin's catenet principles and Vint's generalization (EIN78) backing in 1974/1978. This is a cute compromise between the distributed nature of the needs and the centralization required by the computer capacity of the time, in a decentralized academic governance : it works well for 30 years. Other communications systems had, have and will have different compromises. For example, telephone has its own compromises you refer to.
This leads to a hierarchical vision described as "network of the networks". I submit that this time is over and that the current usage paradigm (hence the usage specification trend) is far more subtle. I would describe it as a "continuity of involvements".
As you say, these compromises were over a few lines. They supported well the growth because of the growth of the computer power and the decrease of the prices. This have hidden that the constraints which lead to these compromise exist no more, and that new - far more user oriented - constraints appeared which should lead to new compromises. This is all what I try to make accepted and introduced in the Internet standard process in being in here.
From the current regulatory thinking evolution, I see that
- considering a telephone number as a sequentially distributed ID nationwide,
- and to attach in it service information (a service prefix) rather than routing information (area code)
are for the few years to come concepts they _want_ to publicly investigate.
IMHO, that approach addresses most of the addressing reasons NATs are used for, since host/subhost addresses are stable and quite unlimited.
But it is true that this could lead gigantic tables size and to an incredible processing if we wanted to have the QoS we initially had on the international public network before OSI. Or to keep and better the current telephone call establishment QoS.
I submit that this problem is far easy to solve than when it was investigated 30 years ago. Because of the experience of the last 30 years, and because of the network capacity/computer power of today (150 K is a big spam mail size. Every morning my AVG loads a 1.5 Meg update and rebuilds my tables). So, I am not concerned by the size to load and by the refresh. But I am concerned by the table size with the growth in the number of what could be "wild" IP addresses and the resulting CPU load on routers. This is why I submit that what we name an IP address, should be understood as an IPv6 header container for call management matrix as much as it is an "IPv4 longer addresses" to transparently migrate from IPv4 to NGN.
This management matrix could for example contain:
- the numbering plan block (for example 010 or 011)
- the type of usage : unicast, anycast, multicast, admin (2 bits)
- the service : IP, Phone, S/MMS, TV, Radio, OSI, Postal, others (hopefully 3 bits)
- the NetworkID (6 Bytes)
- the HostID (6 Bytes)
- the InterfaceID (3 bytes)
total : 16 bytes or /128
I submit that in having a National Digital ID provided to every citizen and incorporated entity and then upon additional request, and being usable the same as Network ID (with a fee to pay for routing management) or as a Host ID, most the problems belonging to the needs of the users side are addressed. As Harald says: the standardizers job is to deliver the protocol and the numbering space availability. A plan as I documented can be suggested/discussed by IAB/IETF, but is the job of the users (that is to say of the ITU or of a "Uninum" consortium) after consulting the IAB/IETF. It will then be afterward to the IAB/IETF to discuss the additional services which can be devised to better support it.
I submit that if this approach permits to make addressing and routing orthogonal as NAT only partly do (two ISPs call for 2 NATs and two host addresses) it also delivers more routing information than the current addresses, because of the national Host ID header. This might permit to reduce the ratio tablesize/number of addresse, to develop the grid of local exchanges city networks will ask for, and to delay the design of a distributed routing supervisory solution.
Let us understand that if their functions are similar (it could be imagined as supporting addresses as numeric names), a supervisor which would match Network or Host addresses with other addresses would not be a DNS, because they would not be at the same layer and used the same way.
I will end with this: the interbox experimental robust architecture we start working on is pure NGN (a "corebox" to interface the services over IP or any other datacoms solution and support OPES). It will be seen as a NAT or pure IPv6 depending on such an addressing plan availability. Time has come when we will sell for $ 100 what I sold for $ 100.000 in 1984 but on a 20 billion addresses system instead of 20.000. I do not think anyone can delay that tide anymore.
jfc
_______________________________________________ Ietf@xxxxxxxx https://www1.ietf.org/mailman/listinfo/ietf