Re: Wedge between transport and ULP

[Iljitsch van Beijnum <iljitsch@muada.com>]

On 3-mrt-04, at 7:12, Dave Crocker wrote:

> Geoff mention an architectural choice that is not currently
> represented by any proposals.  It places a multiaddressing wedge layer
> between the transport layer and the applications layer.

We've discussed this a while ago (where you on the list yet back 
then?). I'm not sure if everyone agreed in the end, but my position was 
and is that this can't work in a sane way.

> The wedge fits rather nicely into that realm the OSI folks like to
> call 'session layer'.  Things put at that layer tend to need to
> replicate control (eg, reliability) mechanisms already done in
> transport.  This will be particularly true if the layer is expected to
> work across multiple transport associations.

The problem is that you can't know how much data made it to the other 
end at any particular moment in time so retiring one TCP session and 
continuing over another means you have to resync. For some protocols 
(most HTTP, FTP) this can be done with relative ease, but solving the 
general case means effectively tunneling TCP over TCP, which makes no 
sense.