On 5/3/10 3:46 PM, Dan Wing wrote:
>> Actually, this seems like what IPv6 Privacy Addresses were made for. 
>
> IPv6 privacy addresses have to be managed by the application and/or the OS,
> and to achieve the same result as NAPT would need to be changed _much_ more
> often than every 24 hours (24 hours is the Windows 7 default for changing the
> host's privacy address, per my understanding).  
As an application developer, what I really want for privacy addresses is
to be able to use a different source address for every outgoing
connection, and to not hold on to that address-to-interface binding any
longer than needed to do FIN-ACKs (or equivalent for other transport
protocols) on that connection.    I want the host to keep a pool of
available addresses to use for privacy addresses, allocate more from the
local network when the pool goes below a threshold, and free old
addresses up when they're no longer needed.

I want to be able to specify, on a per-connection basis, whether (a) the
source address should be a privacy address and (b) whether the host
should accept connections at that address.

It doesn't work nearly as well to reuse a privacy address across
multiple connections, because you never know how long an app will need
to use that connection.

But since you said 24 hours isn't short enough (and I pretty much agree)
- do you think that typical NAT44 internal-to-external address binding
times /are/ short enough?  My impression was that most NATs would
maintain such bindings indefinitely as long as there were active
connections using them.

Keith

_______________________________________________
nat66 mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/nat66

Reply via email to