Re: Identifier/locator recap

[Masataka Ohta <mohta@necom830.hpcl.titech.ac.jp>]

Ran;

> 	I concur with Erik Nordmark's analysis of things.
> 
> 
> On Wednesday, Mar 19, 2003, at 02:45 America/Montreal, Iljitsch van 
> Beijnum wrote:
> > Doing it in middleboxes is more complex,...
> 
> 	Middleboxes break the end-to-end principle and are generally not
> desirable architecturally.

Right. However, to be honest, DNS servers are middleboxes.

But, they are not intelligent but dumb middleboxes.

Load concentration, caused by the violation of the e2e principle,
is taken care of by light weight protocol. Never say X.500.

Lack of reliablity, caused by the violation of the e2e principle,
is taken care of by multiple servers.

DNS servers are necessary evil that the point is not to
introduce yet another evil

> 	Since we have at least one architectural approach (i.e. 
> identifier/locator
> separation) that does not need/use any middleboxes, please let us all 
> agree
> that middlebox approaches are not desirable here -- and not discuss them
> at any further length in this WG.

But, we need DNS.

With IPv6 with 16 bytes of an address, we are a lot more motivated to
favour domain names over raw address than with IPv4.

With multiple addresses (or locators), we are motiveted even more.

In general, if we want to introduce structure unrelated to the
network topology, we have similar problem as DNS.

							Masataka Ohta