RE: [GLOBAL-V6] Comments on AP Consensus

["Craig A. Huegen" <chuegen@cisco.com>]

> -----Original Message-----
> From: Michel Py
> Sent: Monday, April 15, 2002 10:52 PM
> To: Carlos Friacas; global-v6@lists.apnic.net
> Subject: RE: [GLOBAL-V6] Comments on AP Consensus
> 
> > Carlos Friacas wrote:
> > These two "rules" imply that Provider Independent networks
> > shall not exist ???
> > IMHO, this is a good thing!
> 
> IMHO, this is a mistake. What do large (non-ISP) companies want?
> 1. Their own address space (because they don't want to be held hostage
> by an operator).
> 2. Multihoming.

I'll second this and add comments similar to those I sent to multi6.

Why shouldn't very large, multi-homed end sites be able to get PI
address space?

We're struggling with the addressing structure that should be used
internally for Cisco IT.  The currently proposed scheme of handing one
/48 per top-tier root of an end user's provider tree is simply not
reasonable for the large, multi-homed end users.  Operation of such a
network will be difficult at best, and very expensive:

1. The current scheme says that operators in end-user sites should
assign a /64 from each of the /48's to links.  Cisco has nearly 15-20
top-tier roots in its Internet connectivity structure globally.  In
addition, the corporate HQ infrastructure is large enough that a /47 is
necessary to maintain a scalable addressing model in the "site" (what we
call "region") without completely overhauling the network (assumes
applying a /64 where there's an IPv4 segment today).  Maintaining all of
these prefixes across all segments is going to be very, very difficult.

2. Without significant operational overhead, the originating *host* in a
multi-homed-to-multi-homed scenario picks both providers
(inbound/outbound) based on common prefix bits, because it picks the
source address and destination address to use, which are tied to
providers.  Enterprise IT organizations want and need control over the
policy of the network while multi-homed.  The host's selection of the
routes to use (through its selection of addresses) tends to pull control
away from the network operator and leaves it to the host, without
significant operational configuration/overhead.

3. I've seen some comments that imply now that we have address
deprecation, that renumbering is somehow a zero-cost process now, so
that any amount of renumbering an end-user must do is a non-issue.
That's absolutely not true -- it costs significant amounts of money (in
terms of labor) to renumber a very large site.  We still have the same
problem as IPv4, except that we've at least said that organizations get
the same prefix size.

I think that there is a solution gap for large enterprises -- we've
covered the ISP model, we've looked at the smaller sites, but there
hasn't been much thought about the very large networks.

/cah

---
Craig A. Huegen, Lead Network Architect       C i s c o  S y s t e m s
IT Transport, Network Design & Technology           ||        ||
Cisco Systems, Inc., 400 East Tasman Drive          ||        ||
San Jose, CA  95134, (408) 526-8104                ||||      ||||
email: chuegen@cisco.com       CCIE #2100      ..:||||||:..:||||||:..

-
- This list (global-v6) is handled by majordomo@lists.apnic.net