RE: FC encapsulation

["Elizabeth Rodriguez" <egrodriguez@lucent.com>]

I would agree with that.  Might even want to add Bob's explanation to an informative annex, so that if this question comes up again, it will be answered in the doc...

 

(Personal opinion here -- w/ chair hat off.)

 

Elizabeth

-----Original Message-----
From: Franco Travostino [mailto:travos@nortelnetworks.com]
Sent: Tuesday, October 23, 2001 1:57 PM
To: Robert Snively; 'Colin Kelly'; ips@ece.cmu.edu
Subject: RE: FC encapsulation

Should we then add a sentence to the 'FC Frame Encapsulation' draft (introduction?) to capture this restricted scope (class 2 & 3)? This scope is somewhat implied in the SOF table (close to the end) but not stated upfront. Like Colin, someone else may be oversold by the title.

-franco


At 07:06 PM 10/19/2001, Robert Snively wrote:

>
> Is there a fundamental reason why the FC encapsulation draft
> does not include class 1?  The Fibre Channel standards (as far
> as I can tell) do not exclude the possibility of allowing
> class 1 connections through a fabric, so regardless of the intent
> of the FCIP, IFCP, and mFCP encapsulations, I do not see why
> this draft should exclude class 1.

Yes, there really is a fundamental problem.  Class 1 switch
behavior makes a full bandwidth lossless circuit connection
between one Fibre Channel node and another.  If you push exactly
1 Gbit into it, it SHALL push exactly 1 Gbit out, totally in order and
without loss and with low latency.  It will also change
from one circuit connection
to another rather dynamically under Fibre Channel protocol
control, with extremely low connection latencies.

TCP/IP simply does not provide those capabilities, so we never
even considered including it.  Class 6 is a clone of Class 1 in
those respects.  Class 4 has the same general properties, but
with clocked fractional bandwidth, so it has the same problems.

This did not upset most Fibre Channel implementers because all
the key applications that could truly exploit those additional
capabilities provided by TCP/IP were all implemented in Class 2
and in Class 3.  The few highly specialized legacy applications
that used Class 1 were perfectly happy with no IP connectivity.

Hope that is some help,

Bob Snively                        e-mail:    rsnively@brocade.com
Brocade Communications Systems     phone:  408 487 8135
1745 Technology Drive
San Jose, CA 95110

Franco Travostino, Director Content Internetworking Lab
Advanced Technology Investments
Nortel Networks, Inc.
600 Technology Park
Billerica, MA 01821 USA
Tel: 978 288 7708 Fax: 978 288 4690
email: travos@nortelnetworks.com