[e2e] What should e2e protocols know about lower layers?

[Dennis Ferguson]

Christian,

> It should be pretty obvious that suppressing congestion control based
> on a naive test such as "same subnet mask" is BAD. OTOH, developers do
> this for a reason: they are generally frustrated that TCP does not take
> full advantage of the capacity of a high speed network, especially in
> the case of short connections. Make no mistake, the pressure is going
> to increase with applications such as ISCSI and gigabit networks. 

There may be pressure to do something to make people's high speed
networks work better, but it doesn't change the fact that this particular
cure isn't just naive and bad, it is utterly wrong.  If you need to
detect grass the fact that you have a green detector doesn't help
much.  You can't assume that everything which is green is grass, even
if some green things are, and declaring all the things which are green
but not grass to be miscolored seems like an attempt to change the problem
to fit a solution rather than dealing with the fact that no solution for
the problem one has currently exists.

A previous note pointed this out succinctly, I think.  Our current notion
of "one hop away" is entirely a construction of routing and addressing,
and is useful only for those things routing is concerned with.  The sole
concern of routing at this level is connectivity, i.e. who it can talk to
and how it can talk to them.  The ability of a subnet to multicast, i.e.
to deliver a single packet to several neighbours, is only that, it is an
aspect of connectivity which routing might make use of when it needs to
deliver a single packet to several neighbours.  Routing has no concern
for the congestion characteristics of the subnets it is using for
connectivity, multicast or not, and has no means or need to carve them
up on this basis.  Routing knows which things are green, but could care
less which are grass.  Grass detection has always been the sole domain
of the transport protocols.

This could be changed, I guess, but what is required is not what exists
now.

Dennis Ferguson