Cell Relay Archive

[dtcc310@ibm.net]

Oren Minzer wrote:
> 
> Hal Murray wrote:
> 
> (-snip-)
> 
> > > How to handle cell lose at ATM layer?(AAL layer error recovery?)
> >
> > I don't know of any good work in this area.  Some people would
> > claim that it's a bad idea.  For example, a voice sample might
> > be too late if you retransmit so the retransmission just wastes
> > bandwidth.  Or TCP uses lost packets to indicate congestion.
> 
> In the example of profeesional CBR video transmission over an ATM WAN,
> we have used AAL-1 with standard forward error correction, so
> as to achieve perfect transmission in imperfect conditions, without
> retransmission (which is too late anyway).
 
> > I think most people would say you should avoid cell loss instead.
> 
> That's certainly true.
But as far as I know, cell lose is unavoidable anyway, because
complexity of the
network and user application requirements makes call admission control
impossible to conduct
perfect admission control. Also, network node(ATM switch) failure can
cause sever cell lose. So I think upper layer participation for cell
lose at ATM is
necessary. For data communication(with reliable, in sequence data
transmission as protocol design requirements), rather than put the
functionality of reliable data transmission at transport layer(TCP, for
example), why not put it at lower layer(for example, 
a new AAL layer) to achieve less overhead? 

However, if forward error correction at TCP layer works perfect even for
the most possible sever cell lose condition, the differnce between TCP
and AAL error control would be minor. But I think FEC suffers a lot from
sever cell lose(?). 

What I said is about data transfer, not voice or video transmission,
which can still work 
with some cell lose.

I am a newcomer with little industrial experience in ATM networking.
Please critisize my 
comments.

Regards 
Yi Wang
University of New Brunswick
E-mail: yi.wang@unb.ca