[ppml] Address Space versus Routing Slots

Michel Py michel at arneill-py.sacramento.ca.us
Thu May 4 10:25:35 EDT 2006


>>> Vince Fuller wrote:
>>> (the best thing that could happen to a router vendor would be a
>>> return to the replace-the-core-every-2-or-3-years heyday of the 90s)
>> Michel Py wrote:
>> Vince, you can't deny that preserving the status-quo on BGP is the
>> best way to insure that it will happen again.

> Carrying the "PI address" scheme of multihoming (i.e. assign a
> globally-visible prefix to anybody who wants to multihome) from
> IPv4 to ipv6 is the best way to ensure that it will happen again.
> And there are a lot more potential ipv6 prefixes than IPv4 prefixes.

This is very true but it's too late to do something about that, which is
why I support 2005-1. It's not good, but it's the lesser of two evils.
Try to be realistic and look what the alternatives are today for
enterprises that need multihoming and need to deploy IPv6. These
alternatives are:

1. Don't deploy IPv6.
2. Become a LIR.
3. Use RFC4193 and try to get it routed.
4. NATv6 RFC4193 addresses if 3. fails.

If 2005-1 is adopted, the list becomes:

1. Get a PI block from 2005-1.

Note that I'm not saying that 2005-1 will trigger massive IPv6 adoption
as there are numerous other show stoppers, but that's another

> The IETF's position on this matter is that shim6 is the answer. My
> view, that an ID/LOC split is a fundamental requirement for scalable
> multi-homing, is considered heretical among the IETF leadership.

Keep in mind that trying to convince me on this is akin to preaching to
a choir. Ever read draft-py-mhap-intro-00.txt? That's not the point; the
point is what we can do about it.

> Please don't confuse anything that I write with an IETF position
> because the two couldn't be more different.

Vince, one could wonder. I'm reading your slides from the GROW

> Is it worth doing an IAB-sponsored experiment, workshop, or
> other IETF-sanctioned activity along these lines to re-examine
> GSE or explore other solutions?

You have been around the IETF long enough to know that the answer to
this is no, and will continue to be no for the foreseeable future. Why
put in on the table in the first place? Are there any developments that
I am not aware of that would make this fly without being another suicide

Reality check: PI sucks. Do you have a solution with reasonable chances
of adoption that sucks less? If no, please support 2005-1. It sucks, but
all the other alternatives are worse.


More information about the ARIN-PPML mailing list