Cell Relay Archive

[manfredi@arl.bna.boeing.com]

In article <5r6m3n$q3o$1@dismay.ucs.indiana.edu>,
  "S.Ramkumar" <ram@wipinfo.soft.net> wrote:

> I am thoroughly confused why you said that the term
> "connection oriented" applied to ATM is a misnomer.  To send a cell,
> you need to establish a VCC between the end points.  Doesn't this
> characterize ATM to be a connection-oriented protocol?
>
> If not, what are the qualities required to call some protocol
> "connection oriented" or not?
>
> Also I thought ATM was packet (cell rather) switched.

Well, this seems to be one of those semantics arguments which start
religious debates. Still, I think that the misuse of the term causes
endless confusion.

A circuit switched network technology is one that simply establishes a
data path before the data packet(s) are allowed to flow. A packet
switched net is one in which each data packet independently finds its way
to the destination.

Above this network technology, you can layer connection oriented
protocols (TCP, TP4, DECnet, CO XTP, etc.) or connectionless protocols
(UDP, CLNP, CL XTP, etc.). The connection oriented protocols operate
between the end users, and do not necessarily imply anything about the
network technology below. These CO protocols also happen to be reliable,
so that each correctly received data frame is acknowledged by the
destination device.

The fact that ATM switches cells along its VCs is, in my opinion, a lower
level detail. The more fundamental point is that the VC must be
established first, just like a SONET pipe must be established before data
bytes flow. The only difference between setting up an ATM VC and a SONET
pipe is one of implementation minutiae and the fact that we sort of have
an idea that it takes a much longer time to mess with SONET pipes than it
does with ATM VCs. But in principle, you're doing the same thing.

Although this might seem like a lot of religious hooey to many, I do
think that a lot of confusion could be avoided if the concepts of end
user protocols and network switching schemes were kept separate.

It's interesting to note that if we keep reducing the life of a circuit
in a circuit-switched network, we eventually get to something that is
very similar to a packet switched network. For example, suppose that an
ATM-variant were designed to keep an SVC up only as long as it takes to
transmit one IP frame's worth of cells, then it gets torn down. Next IP
frame comes along, and ATM signaling once again establishes the path,
sends the cells, and tears down the VC. This would be functionally very
similar to a packet switched net, in which each IP frame could
potentially be routed differently, during a single TCP session. The only
problem with such an approach is that we assume, deep down inside, that
all this signaling would be too much. But I think this example is good to
differentiate these different concepts and keep them in proper context.

Bert
manfredi@arl.bna.boeing.com

-------------------==== Posted via Deja News ====-----------------------
      http://www.dejanews.com/     Search, Read, Post to Usenet