Jun-ichiro itojun Hagino wrote:
Let me see if I understand this. Without PI, the enterprises say
no, and with
PI, the ISP's say no. Got it.
I believe that a more constructive assessment is that enterprises are
unwilling to pay non-trivial costs to renumber, and ISPs are
unwilling to pay non-trivial costs to support a non-scalable routing
subsystem.
my persistent question to the enterprise operator is this:
how frequently do you plan to switch your isp, or how many times
did you do that in the past?
Each of my Sandelman/Xelerance locations switch ISPs about every 3-4 years.
There are presently 5 such locations, so there is a renumber about every
year. I am about to renumber my SOHO, which is an IPv4/25 in in PA space.
Ironically, my IPv6, which comes through a tunnel, won't need to be
renumbered. (One of the reasons I liked Tony Hain's geographical assignment.
Tunnels don't bother me)
In IPv6, we would be quite happy to always have two IPs on every host
(even though Linux does source address selection poorly, Itojun's code in BSD
does very well).
One PI-ish address which never has to change, which is kinda-site-local,
but might not get ultra-high-bandwidth, and one PA-ish address which is much
faster.
PI addresses don't have to imply centralized NAT to get out. We can do it
in the hosts instead...
That's nicely what shim6 is doing, hmm.
Renumbering is not fun. IPv6 doesn't make it easier, since the DNS dynamic
update stuff never really got fully implemented. (specified, yes, I think)
Wearing my XDS hat, the company survives entirely using rfc1918 internally,
with a dozen IPv4 NATs + VPNs. We use all of 192.168.0.0/16 for enterprise
production. Testing occurs in 10.x. This leaves 172.16/20 for my customer
facing networks. Not enough, so I will have to re-use addresses. Thus my
interest in IPv6.
Using a single prefix massively simplifies the VPNs. That's why
enterprises want a single PI-ish prefix.
The shim6 multiple address possibilities is one reason why I'm vague about
whether or not I need to multi-home my address space.
I suspect that I *won't* need to in the end -- I'll be able to use shim6.
_______________________________________________
Ietf mailing list
Ietf(_at_)ietf(_dot_)org
https://www1.ietf.org/mailman/listinfo/ietf