At 3:04 PM +0700 2/16/01, Robert Elz wrote:
> From: "Paul Francis" <[EMAIL PROTECTED]>
>
> | On a separate note, I know this is probably just digging up dead
> | discussions, but have people considered creating a new IPv6 space
> | which is "globally unique but not globally routable" for use in
> | this sort of internal site communications?
>
>The chief objection to this, as I understand it, has always been
>how such things would get assigned, who would do that, on what
>basis, etc.
Exactly. The last thing we want to do is establish yet another numbering
space that needs to be globally administered, which then becomes another
ICANN political football (like our existing global name and numbering
spaces), and for which we have to figure out some sort of allocation
mechanism/policy to prevent anyone from coming along and grabbing
such site IDs by the millions ("Hey, I'm going to be the next McDonalds,
so I'm going to need a million site IDs! Trust me."), and for which
we'll get dragged into the issue of charging money for site IDs
(initially just to cover the cost of running the allocation service,
of course, but then as a disincentive to grabbing a million addresses,
eventually degrading into a fight to get a piece of the site ID
allocator's revenue stream). And the other thing that would likely
happen is that sites would obtain these non-globally-routable site IDs
and then a few months later some of them would shop around to find an
ISP who would agree to inject their (unaggregated) site IDs into the
global routing system. Now maybe these fears are overblown, but looking
at the current state of IP address allocation, DNS name allocation, and
the growth of unaggregated prefixes in the BGP tables sure makes me wary
about going in that direction.
One could possibly finesse the problems of establishing a new
numbering bureaucracy by exploiting an existing one, e.g., by
saying that site can choose, say, one of its own Ethernet addresses
(that is, a globally unique number from a space managed by the
existing IEEE bureaucracy) to stick into the empty field of its
site-locals -- unfortunately, there aren't enough free bits
there to hold an Ethernet address. How about using IPv4 addresses
for that purpose? Well, some sites today can't get even one IPv4
address, and that's likely to get worse. Besides, we already have
half a dozen IPv6 address formats that have IPv4 addresses embedded
in them somewhere, and I shudder to have to explain yet another one.
Use a random number generator? Well, someone would have to
figure out the Birthday Probability of duplicates and understand
how bad the consequences would be if two sites with the same site ID
got accidentally or intentionally merged, even if it were a very
unlikely event.
And then there may be other counter arguments I've forgotten. It's
certainly true that we've been over this question a number of times
in the past, so if we're really going to dredge it up again, we
should at least do some digging in the archives to make sure we
don't spend a lot of time rediscovering (or worse, failing to
rediscover) previously identified issues.
Steve the Wet Blanket
--------------------------------------------------------------------
IETF IPng Working Group Mailing List
IPng Home Page: http://playground.sun.com/ipng
FTP archive: ftp://playground.sun.com/pub/ipng
Direct all administrative requests to [EMAIL PROTECTED]
--------------------------------------------------------------------
