Re: [mif] MIF at the application level

[marcelo bagnulo braun <[e-mail]>]

Marc Petit-Huguenin escribió:
> marcelo bagnulo braun wrote:
>   
>> but NICE is about how selecting address pairs, rather than just the
>> local address, isn't it?
>>
>> I mean, i would expect that selecting the local address to be a much
>> simpler problem than selecting the address pair
>>     
>
> The well known usage for NICE is to find the best route for NAT
> traversal,

not route, but source destiantion address pair, if i remember correctly
>  but NICE is useful even without NAT.  In a MIF
> environment without any NAT, NICE will find the correct local
> address to use to reach a specific destination, and this without any
> configuration.

agree, but at a high cost. I mean, if i remmeber correctly NICe tries 
the differnet address paris till it finds one that is working
that would certainy work, but i think we can do much better (i.e. less 
latentcy, less traffic) for the problem that you only need to figure out 
which local address to use

>   Before NICE, people tried to find the best route by
> trying to guess what type of NAT was in front of the host, which is
> a problem that cannot be solved.  I feel that choosing the right
> local address is a similar kind of problem that NICE can solve at
> the application level.
>
>   

while the problem maybe similar, my take is that there are much simpler 
solutions that trying every address pair

> The idea behind NICE is that the local address to use is the one
> that have connectivity with the destination to reach.  That seems
> obvious but that was kind of a new concept at the time (and kudos to
> Jonathan for this work).
>   
well, seems to be very similar to ICE :-)

Regards, marcelo


>   
>> regards, marcelo
>>
>> Marc Petit-Huguenin escribió:
>>     
>>> I read the 4 documents listed in the agenda for the MIF BOF, and I
>>> am surprised to see very few discussion about the application
>>> behavior in presence of multiple interfaces.  The documents talk
>>> mostly about operating system configuration but, in the spirit of
>>> the end to end argument, I do think that the final decision on what
>>> network to choose (when there is a choice) should be made by the
>>> application, with fallback to some automatic configuration when the
>>> application is not smart enough to do this.  NICE
>>> (draft-rosenberg-mmusic-ice-nonsip) is a very good example on how an
>>> application can choose between different networks and perhaps this
>>> draft, that does not seem to be maintained currently, could be part
>>> of the deliverables of the future MIF WG.
>>>
>>>   
>>>       
>
>
>