Re: (ngtrans) Update to Provider Independent addressing format drafts

[Brian E Carpenter <brian@hursley.ibm.com>]

Remember that v6 renumbering can include a period of overlap between the old
and new prefixes, when both will work and neither of them is deprecated. For
a large site with the kind of issues you describe, I would expect this overlap
to be quite long - at least a month, why not longer? This will greatly simplify 
the coordination problem. 

In fact, a single homed site changing to a new provider would become
multihomed during the overlap period; and an n-homed site would become
(n+1)-homed during the overlap period. A good multi6 solution *is* a
renumbering mechanism.

  Brian

Michel Py wrote:
> 
> I could not agree more with Tim. Add to the picture that Tim painted
> connections with dozens of customers, providers or partners, all of them
> having BGP peers, static routes and access-lists on their routers
> referring to the block to be renumbered, and you make renumbering almost
> impossible because you have to coordinate it with a score of third party
> companies that might have other priorities than adapting their own setup
> because one needs to renumber.
> 
> Michel.
> 
> -----Original Message-----
> From: Tim Chown [mailto:tjc@ecs.soton.ac.uk]
> Sent: Thursday, October 25, 2001 3:49 PM
> To: Keith Moore
> Cc: multi6@ops.ietf.org
> Subject: Re: (ngtrans) Update to Provider Independent addressing format
> drafts
> 
> On Thu, 25 Oct 2001, Keith Moore wrote:
> 
> > > Renumbering a 100-router, 300-server, 24x7 $1M/day revenue
> generating
> > > room is quite another one.
> >
> > I've been involved with renumbering of ~300 IPv4 machines of widely
> varying
> > platforms, in less than an hour, on multiple occasions.   It took
> prior
> > planning, a bit of coding, and a mechanism for quickly finding the
> machines
> > on which the renumbering failed.  But it wasn't rocket science.
> > If we had standard tools to do this it would be even easier.
> 
> Hi Keith,
> 
> Is that 300 servers, or mostly clients?  If you have a few hundred, or
> thousands of domains with (virtual or otherwise) mail, web and other
> (perhaps black box) services hosted then while I agree the renumbering
> isn't rocket science, it's not much fun logistically.   There's domain
> registrations to update, firewalls to change, a fair few Windows
> applications which insist on IP numbers being keyed in manually, smooth
> mail transitions, updates on other people's references to your IPs,
> and plenty of places where IPs are hard coded on Unix systems.  I agree
> prior planning is key, but to suggest most sites can press a button
> and have it done in an hour seems optimistic :-)
> 
> At present operators such as Worldcom have smaller enterprises over a
> barrel for bandwidth charges.  As bandwidth costs fall, and you see
> rival
> operators prices looking far more attractive, perhaps up to 50% less,
> it's not a no brainer to move operator because you have to renumber,
> and your current operator knows that, and can keep their fees higher for
> you regardless.   If IPv6 were able to offer less painful renumbering
> (which will need help from the likes of Checkpoint and Microsoft), or
> we had some non-NAT form of provider-independent addressing, great.  But
> as it is, smaller enterprises who don't have leverage to take address
> blocks with them suffer when seeking competitive bandwidth rates.  With
> NAT perceived as the "easy" form of provider independent addressing, I'm
> not convinced we won't see IPv6 NAT.
> 
> tim