On 01/11/2009, at 8:20 PM, jamie baddeley wrote:

Yeah, I guess that equation changes if you don't consider your customers
should be able to serve 'content'.

"end-to-end principles" are all lovely in a world with available IPv4 address.  But at some point having a real world IPv4 address all to yourself is going to be an extra cost item.  So, either you'll pay more or use IPv6 for serving content.

My attitude is that customers should be able to directly connect to each
other - with nothing in the middle getting in between that.

For that we have IPv6.   

Maybe that's too idealistic in these sunset years of IPv4 allocation?

Your idealism will need to get dual stacked Jamie.  



On Sun, 2009-11-01 at 19:13 +1030, Matthew Moyle-Croft wrote:
SP-NAT doesn't have to scale to 300 customers per IP.  Just more than

One IP address may only support 4-20 customers with SP-NAT.   But
that's 4-6x what I currently get per IP.   ie.  take one of the /15s I
use now for dynamic addresses - we've gone from 132k (1-1) to as many
as a few million.

SP-NAT is inevitable - at some point we'll have no more IPs to have
and more customers than we have IPs requiring IPv4 connectivity.

IPv6 is a way of diverting traffic away from SP-NAT and having to only
ever buy SP-NAT boxes once.   


On 01/11/2009, at 6:49 PM, jamie baddeley wrote:

On Sun, 2009-11-01 at 15:15 +1030, Matthew Moyle-Croft wrote:
On 31/10/2009, at 5:58 PM, TreeNet Admin wrote:

The huge problem is all the home customers with ancient

I don't think they're a huge problem as they're the least likely
notice the implementation of SP-NAT in front of their connection.
know at least one large ISP in our region considering this as
phase #1  
of an SP-NAT implementation.

I saw a great presentation at the IPv6 Hui that were held in
Christchurch, Auckland and Wellington recently.

It was presented by Dr Hiroshi Esaki from the WIDE project in Japan.

He made it pretty clear that SP-NAT does not scale. See here:

Start at Page 9 :-)

The point he makes is this. TCP has a limited numberspace. Stuff on
internet in terms of number of connections per session can be large
unknown frankly).

iTunes has in excess of 200 connections per session. Divide 65K by
connections and you're left with an equation that says you can only
serve iTunes to about 300 odd users from one NAT box.

But you know that :-)


The big issue right now is:  the lack of IPv6 native support in
CPE at  
all.   If we had the larger CPE vendors starting to implement (*)
with a 2 year replacement time for most CPE we'd be fine by the
time a  
lack of v4 addresses starts to pinch.

In NZ at the moment with VDSL2 starting to be rolled out it'd be
PERFECT time to start squirrelling (or whatever marsuipal/mammal
guys have) it into people's houses as people replace CPE.   But
guessing that it's not happening that way alas.

We need to start getting IPv6 out there to end-customers to start
ball rolling to start shaking the problems down.  (And believe me

there a whole LOT of problems with IPv6 in actual real production
DSL customers ...)

(*) Let's face it - given that almost all the CPE is Linux based
just laziness on the CPE vendor's part as it's already bloody
done for them.


Matthew Moyle-Croft
Peering Manager and Team Lead - Commercial and DSLAMs
Internode /Agile
Level 5, 162 Grenfell Street, Adelaide, SA 5000 Australia
Email: mmc@internode.com.au    Web: http://www.on.net
Direct: +61-8-8228-2909      Mobile: +61-419-900-366
Reception: +61-8-8228-2999        Fax: +61-8-8235-6909

Matthew Moyle-Croft
Peering Manager and Team Lead - Commercial and DSLAMs
Internode /Agile
Level 5, 162 Grenfell Street, Adelaide, SA 5000 Australia
Email: mmc@internode.com.au    Web: http://www.on.net
Direct: +61-8-8228-2909      Mobile: +61-419-900-366
Reception: +61-8-8228-2999        Fax: +61-8-8235-6909