Re: IPv6 addresses for EP
- Date: Fri, 11 May 2001 08:27:56 -0700 (MST)
> I've removed lir-wg@localhost from the distribution !!!
The one message with lir-wg@localhost only went to you, Dave. It didn't go
to the list :> The problems we're seeing appear to be faulty listserv
software at RIPE, imo.
> Looking at a couple of professionally run exchanges in Europe, I disagree.
>
> Exchange support infrastructure needs to be multihomed, and as Randy so
> succinctly pointed out, this cannot be achieved through a transit provider
> (or with their addresses).
Perhaps I missed that argument by Randy. Could you quote the relevant
portions?
As you term "Exchange support infrastructure", I reiterate this has
nothing to do whatsoever with exchange points needing address space to
establish themselves.
There are/should be special assignment policies to networks which play a
sufficiently important role to internet infrastructure. Exchange points
have been widely-accepted as one such sufficiently important role.
The "critical" role they play is to encourage the interconnection of
operational networks to improve routing. As such, as a point of
interconnection for diverse networks, exchange points must have globally
unique address space in their core. This address space is solely for the
interconnection of participants.
Additional activities (support activities) of an exchange point operator
must be considered separately from the activity of interconnecting
participants. Mail servers, Hank's looking glasses, traffic measurements -
all these are perfectly well and good, but are activities which I believe
can be well-supported using upstream space. I do not believe these
activites require RIR-assigned address space. That said, if Randy really
did argue they do, I would like to be reminded of the argument.
> I would see an exchange being in control of a block (at least /48, although a
> normal LIR assignment would make more sense to me to ensure routability). This
> block would be globally routable and provide connectivity to support
> infrastructure.
Operators petitioning the RIPE NCC under an exchange point policy should
be able to define their own needs. Any policy that develops out of this
discussion should not specify address assignment sizes - let's put the
onus on both the petitioner and the RIPE NCC to determine
appropriately-sized assignments.
> We could expand this discussion to cover multihoming in general - why would
> anyone switch to IPv6 from IPv4 if it means they can no longer multihome like
> they did in IPv4.
No really - let's not expand this discussion to cover IPv6
multihoming. Please.
/david
*--------------------------------*
| Global Crossing IP Engineering |
| Manager, Global IP Addressing |
| TEL: (908) 720-6182 |
| FAX: (703) 464-0802 |
*--------------------------------*
