Re: Réf. Re: Alcatel Space interest about IP/DVB

["Kearney" <clausen@cosy.sbg.ac.at>]

From: "Keith Hogie" <Keith.Hogie@gsfc.nasa.gov>
To: <ip-dvb@erg.abdn.ac.uk>
Sent: Wednesday, April 17, 2002 10:24 AM
Subject: Re: Réf. Re: Alcatel Space interest about IP/DVB

In my opinion the big difference between connectionless = datagram and
connection-oriented on the lower layers is the use of the addresses. In
connectionless systems you use the full address in every packet as
exemplified in Ethernet or Toke ring LANs, whereas in connection-oriented
systems you use thte full address only once for the set-up and then continue
using a "label", VC number etc. With adresses - or device_IDs getting bigger
and bigger (IEEE MAC -64, IPv6 ->128 bits) thjere is increased interest in
using shortel "labels - see the MPLS developments.

All I wanted to point out is the way in which we might treat the addressing
in the encapsulation. In my opinion the PID is just a "virtual broadcast
channel" and NOT a virtual circuit or path. It can be used as a
pont-to-point link but I doubt that this very sensibel; if there is a need
for this type of addressing it should probably be delegated to the
link-level i.e. the encapsulation should include an "address/label" field.

--Horst Clausen ("kearney")
>
>
> Kearney wrote:
> >
> > Hello Stéphane>
> > > Thanks a lot for your reviewing and comments !
> > >
> > let's keep the discussion cooking!
> > > By "Ethernet-like" we only mean "connectionless layer 2 based on
broadcast
> > > medium".
> > Question - why does it have to be connectionless?
> > Most of the applications are anyhow sessio-oriented - TCP, HTTP, and in
> > particulatr multicast applications are based on the concept of a
session. If
> > you "tune" your receiver to a PID you are basically opening a session -
so
> > why wouild you like to have the next higher layer (link/encapsulation)
> > connection-less?
> >
>
>   There may be various definitions of "connectionless" and "session"
> going on here.  In one sense things like an ATM or Frame Relay
> Permanent Virtual Circuit (PVC) are a "connection" but that is a
> often a different definition than a TCP connection.
>
>   I see the OSI terms "connection-oriented" and "connectionless" refer
> more to upper layer protocols and not to things like a PVC.  Their
> "connection" term relates to protocols that do some sort of handshaking
> like X.25, PPP, or TCP.  The protocol actually initiates a "connection",
> which may also be called a "session", by exchanging packets between
> the two ends of the connection and creates the connected state.
>
>   I think the comment on "Ethernet-like" "connectionless" was aimed at
> making sure you don't start running a connection-oriented lower layer
> underneath a connection-oriented upper layer like TCP.  This relates
> back to the days when people ran TCP over X.25 lower layers.  TCP is
> fully prepared to wait for retransmissions and having X.25 underneath
> it also doing retransmissions is redundant and causes other problems.
>
>   When you ran multiple TCP connections over a single X.25 connection,
> lost data packets could cause X.25 to slow down and wait for
> retransmissions. This would delay all of the TCP connections running
> over it even though the lost packets only affected one TCP connection.
> This is where Frame Relay came from.  It removed the X.25 flow control
> and retransmission and just accepted frames and relayed them on.
>
>   You can call the lower layer data path a "connection" if you want
> but the real point about "connectionless" is that the lower layers should
> not be introducing any sort of flow control.
>
> ----------------------------------------------------------------------
>   Keith Hogie                   e-mail: Keith.Hogie@gsfc.nasa.gov
>   Computer Sciences Corp.       office: 301-794-2999  fax: 301-794-9480
>   7700 Hubble Dr.
>   Lanham-Seabrook, MD 20706  USA        301-286-3203 @ NASA/Goddard
> ----------------------------------------------------------------------
>