RE: multi6-functional-dec and re-homing

[<john.loughney@nokia.com>]

Marcelo,

> i am not sure that the wording is appropriate here...
> I mean failure detection and rehoming involves at least the following 
> items:
> - Detecting the failure, which involves different heuristics that 
> depnds on the apps, and other external triggers like ICMP error 
> messages, and then probably some attempt to perform an explict 
> reachability test, which would fail
> - then exploring alternative paths and identifying one that is working
> - and finally moving the communication to the new path, which 
> probably 
> requires a reachability test to the new locator pair (in order to 
> prevent flooding) and perhaps a signaling message to inform about the 
> locator change (something like a BU in mip)
> 
> Now, even though all this is involved in the failure and rehoming 
> event, some parts of this fall within the protocol, (in particular, the 
> messages to verify reachability, and the messages to change the locator 
> being used), while otoh some other items belong to the state management 
> part, like when to determine that a failuer is occurring and launch the 
> rehoming process
> 
> I guess that we have on one hand the protocol that is needed to do this 
> (reachanility test, alternative path exploration messages and rehoming 
> signaling) and on the other hand we have some heuristics that will be 
> used to determine failures and rehoming events.

My proposal was for something a lot simple.  I think we need:

a) General architecture for shim6;
b) General protocol mechanism for shim6 (stressing general);
c) document on multihoming triggers - basically figuring out when to
   invoke the shim6 protocol mechanism.

For c) there are some no brainers like starting up an interface or removing
an interface; slightly more tricky might be some mobility related events
or layer 2 indicators; more tricky still are failure detection.  Failure
detection is tricky because what constitutes failure - 4 TCP retransmissions?
And so on.  This includes different heuristics, etc.

I think that a general protocol could take the multihoming trigger document
and then do the things you list above - reachability tests, path exploration,
new locator pair selection ...

Am I making sense here?

John