The IP over ATM Mailing List Archive by date

[rajeev@trillium.com]

> From: shah@lds.loral.com (Nimish Shah)
> Subject: Re: one tcp per vc ???
> 
> > To maintain the current tcp/ip software in most of the existing
> > os, several Q93b or signalling implementations only do the ip to
> > vpi/vci mapping (e.g., SPANS, VINCE) instead of tcp/ip to vpi/vci
> > mapping to avoid the issue for the specifc call tear-down.
> > 
> > --
> Are there any signaling packages available or announced in the market 
> that do tcp/ip to vpi/vci mapping ?

I think there is some misunderstanding here about the capabilities of
the signalling protocol. The job of signalling is to set up SVCs. Protocols
to this effect are defined in ATM Forum's UNI 3.0/3.1 and ITU's Q.2931.

The job of mapping higher level applications, such as TCP/IP, to ATM
services is done by a separate protocol layer such as RFC 1577 or
LAN Emulation. It is this layer that translates higher level protocol
addresses, such as IP or MAC addresses, to ATM addresses. Once this
translation is done, the signalling protocol is invoked to establish
an SVC with the desired ATM endpoint.

Having said that, let me comment on the original question about mapping
TCP directly to ATM services. The work of this group was focussed on
mapping IP to ATM services, because ATM is seen as yet another subnetworking
technology in the Classical IP over ATM model. This leads to the 
situation of TCP (connection oriented) sitting over IP (connectionless)
sitting over ATM (connection oriented), and TCP being unable to pass
its connection specific parameters to ATM. To obtain this optimization,
TCP must work directly over ATM. In this context, ATM is seen as an
internetworking technology, not merely a subnetworking technology.
Cole's framework document exposes the issues when ATM is seen in
different models as a different technology.

I think the fact that TCP to ATM mapping is not yet defined (or being
worked on) is because we are making our way up the stack, to prove the
viability of ATM. ATM Forum's LAN Emulation group is working with
mapping MAC layer services to ATM. This group is working on mapping IP
level services over ATM.  The strength (or weakness) of these efforts is
that they hide the underlying ATM services from higher level
applications.  At some point in the future, when ATM services (and APIs)
become more stable, there would definitely be attempts to map higher
level applications to ATM. That, after all, is the promise of ATM!

-------------------------------------------------------------------
Rajeev Gupta                       Email     : rajeev@trillium.com
Trillium Digital Systems, Inc.     Voice (w) : (310) 479-0500
Los Angeles, CA 90025.             Fax   (w) : (310) 575-0172
* I speak only for myself, but make no claim to original thought *
-------------------------------------------------------------------