Re: [Dime] [RFC3588bis-34] - Host-IP-Address AVP

[jouni korhonen <[e-mail]>]

On Sep 25, 2012, at 12:19 PM, Ralph Loader wrote:

> 
>>>> Yes for the first part and no for the second part.. you are not supposed to send
>>>> addresses that you cannot be connected at.
>>> 
>>> But in general, a Diameter implementation can't reasonably be expected to know which of its IP addresses will be usable by a particular peer.
>> 
>> A Diameter node is supposed to know which addresses it can bind and
>> did bind to. The addresses it might put into host-ip-address do
>> not come out of blue.
> 
> That is irrelevant.  The problem is that just because the Diameter node has bound to an address, does not imply that it's possible for a specific peer to try and connect to.  
> 
> E.g., node A, node B are both behind different firewalls, node A binds to 192.168.1.1, on node B's network, 192.168.1.1 is assigned to a completely different machine.

Right, private addresses + NAT -> you know you will hit your head to wall.
It is not worth putting any energy to solve NATted environment related
problems.

> Do you have a specific suggestion for the algorithm a node should use in selecting which of the nodes addresses are put in the Host-IP-Address field?

Not any specific algorithm than the normals:
 o At startup the node scans for the interfaces and addresses in those
   it can bind to.
 o the selection of interfaces can be controlled by many means like
   config file, and the selection of the "main" IP address can also
   come from a config file etc.

Now there is an obvious question what happens if some addresses on
some interfaces are just not reachable e.g. from outside. I would
argue that every permutation of a deployment can be _made_ to break
with some effort regardless of the protocols you got.

>>> Best just to treat the field as informational.
>> 
>> Don't fully agree. In case of TCP I could. In case of SCTP I don't.
> 
> For SCTP Host-IP-Address is utterly redundant.  SCTP already supports multi-homing and telling peers of additional addresses (although in practice, I've seen that go hilariously wrong).

- Jouni

> 
> Ralph.
> 
> 
>> 
>> - Jouni
>> 
>> 
>>> 
>>> Ralph.
>>> 
>>> 
>>>> 
>>>> 
>>>> 
>>>> - Jouni
>>>> 
>>>> 
>>>>> 
>>>>> Ankur Garg
>>>>> 
>>>> 
>>>> _______________________________________________
>>>> DiME mailing list
>>>> [e-mail]
>>>> https://www.ietf.org/mailman/listinfo/dime
>>