Re: Layer 3.5 is actually layer 4

[Brian E Carpenter <brc@zurich.ibm.com>]

Masataka,

You have made your point repeatedly. Some people disagree with you,
and at this point, more repeating will not change that. So could
we please close this discussion?

Of course, the architectural analysis to be developed in and
following Seoul will have to take this discussion into account.

Thanks

    Brian

Masataka Ohta wrote:
> 
> Christian;
> 
> > Thinking of UDP as a transport protocol is misleading. UDP is not really
> > a transport protocol, it is a pass-through. It only performs one of the
> > classic TP functions: multiplexing.
> 
> We can be fine to say UDP is a transport layer protocol to
> pass-through.
> 
> OTOH, if you implement TCP in a way that kernel does multiplexing
> only and all the other TCP functionality (connection management,
> congention avoidance, retransmission and so on) is performed in
> user space, it is no different from UDP doing similar things in
> user space.
> 
> So, we can also be fine to say TCP is a combination of a transport
> layer protocol to pass-through and an application layer protocol
> to do other things.
> 
> Note that there are OSes where kernel and user spaces are not
> separated.
> 
> That is, debate on what funcitonality is of transport and what
> else is of application is not meaningful, because the difference
> is internal implementation one not visible from the network.
> 
> There can be a difference when the network support port number
> based QoS assurance, as is the case with networks with RSVP,
> where we can definitely say that the multiplexing functionality
> does belong to the transport layer.
> 
> Anyway, that
> 
> > In a multi-homed
> > environment, the application will also handle multi-homing.
> 
> is enough to deny 3.5 fallacy and UDP is not the only special
> case.
> 
>                                                 Masataka Ohta