RFC 3576 support

[Alan DeKok]

Arran Cudbard-Bell wrote:
> Ok just the asynchronous nature of CoA requests...  It's not really the
> servers job to process feedback from the various SNMP probes, IDS's , or
> track changes in the authorisation of users or their equipment.

  Yes.  That's what proxying is for.

> I guess I can see very few usage cases for CoA where the server will
> actually make the decision to send a CoA request on it's own, so why not
> just use the client or client libraries ?

  if user uses more than 2G of bandwidth, then kick them off.  This is a
valid decision for a server to make.

  Forking an external program means that it's independent of the server
core, and is more difficult to integrate with SQL, etc.

> How were you thinking of triggering CoA events? Didn't you say there
> were issues with an instance of the server being both a CoA proxy and a
> CoA generator ?

  Yes.  If you're going to proxy CoA requests, there's no need to
*generate* a CoA request for the one you're proxying.

  On the other hand, if you're receiving an accounting request, it may
make sense to generate a CoA request.

> Have to wait for vendor support *grumble*.
> 
> Let me know when you get your trapeze kit so we can compare notes :)

  Will do.

  Alan DeKok.