Resemblance to CLNP ends with that my one uses node address;
- Node addresses are local with me while it is global with CNLP.
- I don't carry subnet locators.
In my model, packets are not routed on the subnet locator (which is missing
in my model). Rather,
- Each subnet router keeps track of the membership of nodes inside its
subnet.
- Whenever there's differential in the membership, by either a new
incoming node or a leaving node, this change of membership would be
broadcast to other subnet routers in the same domain(AS).
- Assuming the an LS routing is used inside a domain, such differential
can also be distributed as an addendum to normal LS update packets.
- So, each router will know where any node should reside in a subnet
managed by which router.
- Since the routing is still LS, each router keeps knowledge of the
whole network topology so that shortest paths can be determined, which will
then indicate what will be the next hop router for a destination node from
the current router.
- All this can be done without padding locators (or addr of subnet
router) to each packets. Packets carry only node address and AS numbers
instead, the latter not being used within an AS but only to be used for IDR.
- The whole intra-domain routing operation might look similar to bridge
operations. Not totally, however. Just similar.
- The penalty is that we deal with flat node addresses.
- The gain is that mobility is provided directly by routers as part of
inherent operations of routing. No extra infrastructure like HA / FA or
ID/Loc mapping servers. This way, faster mobility can be provided since it
is done directly by routers.
- The whole structure is much friendlier to mobility. Mobility is an
inherent feature, not as added/patched-on function.
The whole point is:
- In contrast to common belief that you need ID and Locator separately
to provide seamless mobility,
- the same can be done only with an identifier of a node (now I call
node address), without extra bothering about locators.
This description is more from the mobility perspective, but also renders
consequences of routing in general:
- Address is local, name is global.
- IDR operates only on AS numbers or Domain addresses. (Node) addresses
are used only in inter-domain routing.
- In a sense, IDR is quite symmetrical to intra-domain routing, with
migrating(multi-homing) domains seen as network nodes.
- So, any routing can be used for IDR. People might still prefer policy
routing and hence BGP4-modified.
- Size of the routing table is limited by the AS# space. Too big? Then,
split in multi-tiers.
- No Internet governance since address are local. (Although that for AS
will still in place. But much less political.)
On Sun, May 23, 2010 at 1:54 PM, Joel M. Halpern <[email protected]>wrote:
> Note that although CLNP use Node Identifiers, these were coupled to subnet
> locators. That is, the NSAP specifically identified the subnet to reach the
> host on. Further, out expectation was that a preamble to the NSAP would be
> the service provider providing connectivity. SO that the NSAP was a
> topologically sensitive address, with a NODE identifier as the last portion.
> (Yes, there are parallels to some of the other ideas this research group
> has discussed. We did not take that work anywhere near there at the time.)
>
> Yours,
> Joel
>
>
> Dae Young KIM wrote:
>
>> On Sun, May 23, 2010 at 2:06 AM, Noel Chiappa <[email protected]>
>> wrote:
>>
>>> > From: Dae Young KIM <[email protected]>
>>>
>>> > the role of PoA is already served by MAC address, and not has to be
>>> > duplicated by extra 'Locator'.
>>>
>>> i) Not all networks have a MAC (although most do, now).
>>>
>>> ii) A MAC serves to globally _identify_ an interface, but it is not
>>> enough to
>>> _locate_ it (i.e. to be useful to the path selection). It will be
>>> necessary
>>> to add some extra information, such as a structured name of the network
>>> the
>>> interface is connected to, to make it an interface name which the path
>>> selection (routing) can use. After all, if you do not have that extra
>>> information, you basically have a world-wide bridged network.
>>>
>>
>> Not necessarily. I think I can build a sound network topology only
>> with node addresses. CLNP, for example, didn't need interface address
>> to do routing.
>>
>> Remember, my question was in the context of a node which has _two_
>>> interfaces, to widely separated (in network connectivity terms) networks,
>>> such as i) a particular wireless LAN, and ii) a 3G cellular network.
>>>
>>
>> Why should it be a problem in my picture?:
>>
>> - At inter-net level, my node will have two links each pointing to a
>> remote router at the other surface of either LAN or 3G.
>>
>> - LAN will identify one of the interface with its MAC address while
>> 3G will do the other with its L2 address(whatever it is).
>>
>> > mobility is inherently supported by routers without resorting to
>>> extra
>>> > mapping (ID>Loc) or agent(HA/FA) infrastructure.
>>>
>>> Only within an AS. What happens if the node leaves the AS (perhaps to a
>>> different wireless LAN)?
>>>
>>
>> Moving across LANs within an AS won't affect its fast mobility.
>>
>> There'll surely be a problem in fast mobility when a node leaves an AS
>> for another AS:
>>
>> o In its most primitive fashion, the TCP connection would break.
>> You have to reconnect.
>>
>> o A fast inter-AS context-switching mechanism might be devised.
>>
>> o DNS should be updated with a new mapping, name > (addr, AS-new).
>> This, of course, is a challenge; DNS synchronization is slow. How to
>> work around this?
>>
>> I have yet to work on these inter-AS mobility issues. Hope I won't totally
>> fail.
>>
>> Noel
>>>
>>>
>>
>>
>>
--
DY
_______________________________________________
rrg mailing list
[email protected]
http://www.irtf.org/mailman/listinfo/rrg
