[e2e] TCP over IP over ATM

[Detlef Bosau]

First of all, I'ld like to thank for the many papers I've been pointed 
to. I'm still to read them in detail.

My objective is not to study TCP/ATM in great detail, however it's the 
obvious analogy to wireless networks which is interesting.

Up to now, people tend to think all networks to be wired networks - with 
invisible wire.
The extremely strong assumption being so made is no more and no less 
that a link's service time and a link's workload capacity remain more or 
less constant as long as the packet size remains constant.

Either assumption, concerning the service time and concerning the 
capacity, is wrong.

This is hardly admitted by anybody from the wireless community - in ATM 
it is "don't care" because, at least to my knowledge, ATM fades out.

However, before I fail to convince the community of adverse effects of 
link variations on TCP in mobile wireless networks again and again I 
thought it useful to have a glance on a wired "analogy", where the 
problems are basically pretty much the same.

Not because ATM were error prone. But because ATM, particularly with 
ABR, may expose a behaviour similar to that of a mobile network with 
error correction to upper layers.

Would you agree? Or would you contradict here?

Detlef