RE: [EME] Re: transport recovery at the APP layer ?

>The transport layer no longer becomes an E2E service, and 
>all the capabilities it provides - reliable transport in particular
>must be reinvented at the app layer.

This seems to be the case for all P2P applications running on some overlay network such as the openDHT. It certainly applies to P2P SIP.

Please see http://p2psip.org 

Transport is indeed not e2e due to NAT and as we have seen, Skype for example manages to provide e2e over discontinuities at the IP layer.

In the case of P2P SIP, the ICE protocol provides the NAT traversal at the application layer.

Thanks, Henry

-----Original Message-----
From: Joe Touch [mailto:touch@ISI.EDU] 
Sent: Wednesday, November 15, 2006 11:05 AM
To: Rémi Després
Cc: eme@irtf.org
Subject: [EME] Re: transport recovery at the APP layer ?

Rémi Després wrote:
> Joe Touch wrote :
>> It'd be useful to decide whether recovery includes the assumption that
>> the app layer will reconnect and restore E2E context. If that's
>> _assumed_ here, it'd be a first IMO.
>>
>> Joe
> This is IMO an important question.
> 
> We agree, I believe, that forbidding middleboxes that do some transport
> relaying is unrealistic.
> (In other words, for the real world Internet,.the EME model must be
> accepted in addition to the E2E one.)

I'm not sure we agree.

I agree that we can call a proxy a middlebox, and so long as the
endpoint KNOWS it's not talking to the ultimate end, that's fine.

I don't agree that this is OK when the endpoint doesn't know about it.
I.e., if you prefer: E2E trumps EME.

> Then, there seems to be no simpler place, to recover from transport
> relay failures, than just above the transport layer..

The problem with this model is that all it does is move E2E to the
application layer. The transport layer no longer becomes an E2E service,
and all the capabilities it provides - reliable transport in particular
- must be reinvented at the app layer.

That's been a step backwards, IMO.

> It is thus not a surprise that  real applications, that had to adapt to
> the real transport world with no extra layer between transport and
> themselves, already have that recovery mechanism.
> 
>  It would indeed be a first _as an recognized model_, but AFAIK not in
> implemented code.

Right. I don't agree we should recognize it where it breaks E2E. (it
doesn't break E2E when the middle is explicitly known).

JOe

_______________________________________________
EME mailing list
EME@irtf.org
https://www1.ietf.org/mailman/listinfo/eme