multiple prefixes

Doug Barton dougb at
Tue Feb 12 06:28:22 CET 2013

On 02/11/2013 08:49 PM, Erik Kline wrote:
>> B) NPT != NAT. Please stop spreading FUD.
> Actually, even this statement is slightly FUDish, depending on one's
> perspective.
>  From a pure technical perspective, it's true.

Good thing this is a technical list. :)

>  From an application writer's perspective it's patently false, as the
> technical difference between the two is irrelevant.

I'm pretty sure you're wrong about that, although the differences are 
pretty subtle. In the unlikely event that the network admin has left the 
incoming ports open to all of the inside hosts, NPT is nothing like NAT 
since the prefix translation creates a 1-to-1 relationship between the 
addresses and the ports. So there is no UPnP/PMP drama to deal with.

However, given that leaving all the incoming ports open is almost 
certainly not going to happen, the same SPIF mechanism that works for 
NATv4 will work for NPT hosts for connections that the hosts originates, 
so in that sense you're right, they operate pretty much the same way.

> What I suspect you mean when you say (elsewhere) that ULA+NPT costs
> nothing is that it costs /you/ (or "the network operator") nothing.
> But there is a cost, and you've just moved it elsewhere.

Can you define what you believe those costs are, and why they are important?


More information about the ipv6-ops mailing list