James Kempf wrote:
Pekka,
I agree with Joe and Tony. Tunnels are a tool for virtualizing the address
space. If you are going to propose that they are a flawed tool, then I think
you need to propose an alternative that has "better" (for some sense of the
word) properties. The only alternative I can think of (swapping IP
addresses in the header, i.e. NAT) is worse, but maybe there are other
alternatives.
There is at least one, i.e. the approach taken by shim6, which explicitly
virtualizes the address ("identifier") seen by upper layers, and
uses a variety of addresses ("locators") at a slightly less virtual
level known as Layer 3, which by the way was invented to virtualize
Layer 2 addresses about 30 years ago.
If we are thinking forward, I would like to repeat a remark I've made
a number of times, in all seriousness: the flaw in the OSI model is
that it has a finite number of layers instead of being recursive.
It seems that we need to make "Layer 3" recursive to have a clean
virtualization approach. Tunnels slot naturally enough into a recursive
view of the nextwork layer.
Brian
_______________________________________________
Int-area mailing list
[email protected]
https://www1.ietf.org/mailman/listinfo/int-area
