On Sun, Jun 13, 2010 at 09:11:31AM +0900, Dae Young KIM wrote:
> On Sat, Jun 12, 2010 at 4:55 PM, Tony Li <[email protected]> wrote:
> >
> > Hi,
> >
> >> ? ?1. Why, in the first place, did people allow sites to inject PI
> >> addresses in DFZ? Why not simply reject PI in DFZ, limiting their use
> >> strictly inside a site?
> >
> >
> > Because that doesn't solve the multi-homed site problem.
>
> I'd believe/hope that there could have been solutions for multi-homing
> even with keeping this scenario, if people would have enough research
> before ever starting to distribute PI addresses.
It's probably worth mentioning that in the "early days" of the Internet
(before CIDR came along in the early/mid-1990s), there really was no concept
of "PA" and "PI" prefix assignments; every new site connecting to the
Internet obtained a unique block of non-aggregatable address space and a
route for that new block was explicitly propagated into the global routing
system. The distinction between "PI" vs. "PA" and consequent idea of
"rejecting PI" is a relatively new development.
One might argue (and some did argue) that solving the multihoming problem
through locator/identifier separation should have been a fundamental goal
of any "next-generation IP" proposal effort during the 90s. Unfortunately,
such arguments fell on deaf ears so that's why we're faced with retrofitting
a solution now. Not surprisingly, this implies backward-compatibility and
transition issues which tend to make an architecture or design a lot less
elegant than might be desired.
--Vince
_______________________________________________
rrg mailing list
[email protected]
http://www.irtf.org/mailman/listinfo/rrg
