RE: RFC1006 extension ISO Transport Class 2 Non-use of Explicit Flow Control over TCP

[Yanick <[e-mail]>]

Hi Harald,

I have included all the mail we have exchanged so far on the Informational 
RFC - ISO Transport Class 2 Non-use of Explicit Flow Control over TCP - 
RFC1006 Extension.

I have done this so that Allison Mankin, Jon Postel, Dan Harrington and 
Jim Bound can follow the conversation.

You described the following scenario in your last mail:

#- Send ED TPDU over Expedited channel
#- Send DT TPDU over Normal channel
#- IP packet loss hits the ED TPDU, causing delay
#- DT TPDU over Normal channel is delivered ahead of ED TPDU, which is
#  illegal according to ISO

We discussed at great length this scenario. And we also thougth that EA 
was the way to "absolutly guaranty" ED delivery ahead of DT. BUT having EA 
also meant that ED needed to carry sequence number and that flow control 
needed to be enabled (at least on the Expedited data virtual channels).

We decided to keep/make the extension as straightforward as possible and 
as close to ISO 8073 and RFC1006 philosophy as possible. Also what we 
really needed is separate virtual data channels for Normal and Expedited and 
that Expedited data channel not be blocked by a congested normal data channel.
One other (and real) goal was also the "ease of implementation" of this 
extension for all existing RFC1006 implementation.

So what I propose is that I add a paragraph clarifying the above.

#Apparently there are other IETF members that have voiced problems with the 
#document, but since I am not a member of any mailing list where this has
#been discussed, I don't know if the current document has been changed to
#reflect any of these comments.

I am not aware of any other "voiced problems". Can you please be clearer?

Let me know and best regards,
Yanick


From:	VBORMC::"[e-mail]" "MAIL-11 Daemon"  4-MAY-1995 11:24:38.68
To:	Yanick <taec::pouffary>
CC:	[e-mail]
Subj:	Re: Submission Informational RFC

Yanick,
thanks for your prompt response.

I am sorry that I start technical discussions so late in the game, but I in 
fact did not discover the internet-draft at all - my usual line of work is 
applications, and my interest in this topic is because of ancient history (I  
did a thesis work on OSI transport over IP back in 1984!)

The expedited problem is not quite as simple - the problem is the sequence
of events:

- Send ED TPDU over Expedited channel
- Send DT TPDU over Normal channel
- IP packet loss hits the ED TPDU, causing delay
- DT TPDU over Normal channel is delivered ahead of ED TPDU, which is
  illegal according to ISO

In order to prevent this from happening, one has to stop sending DT TPDUs
until the ED TPDU is delivered, which means that you need an EA TPDU to
discover that it is in fact delivered, so that you can start sending
DT TPDUs again.

Apparently there are other IETF members that have voiced problems with the 
document, but since I am not a member of any mailing list where this has
been discussed, I don't know if the current document has been changed to
reflect any of these comments.

        harald A


% ====== Internet headers and postmarks (see DECWRL::GATEWAY.DOC) ======
% Received: by vbormc.vbo.dec.com; id AA14922; Thu, 4 May 95 11:13:55 +0200
% Received: from domen.uninett.no by mail1.digital.com; (5.65 EXP 4/12/95 for V3.2/1.0/WV) id AA32325; Thu, 4 May 1995 02:20:39 -070
% Received: from dale.uninett.no by domen.uninett.no with SMTP (PP)  id <[e-mail]>; Thu, 4 May 1995 11:15:21 +020
% Received: from dale.uninett.no (localhost [127.0.0.1])  by dale.uninett.no (8.6.9/8.6.9) with ESMTP id FAA07702; Thu, 4 May 1995 05:15:16 -04
% Message-Id: <[e-mail]>
% X-Mailer: exmh version 1.5.3 12/28/94
% From: [e-mail]
% To: Yanick <taec::pouffary>
% Cc: [e-mail]
% Subject: Re: Submission Informational RFC
% In-Reply-To: Your message of "Wed, 03 May 1995 16:00:22 +0700." <[e-mail]>
% Mime-Version: 1.0
% Content-Type: text/plain; charset="us-ascii"
% Date: Thu, 04 May 1995 11:15:16 +0200
% Sender: [e-mail]

From:	TAEC::POUFFARY     "Yanick"  3-MAY-1995 16:10:38.26
To:	VBORMC::"[e-mail]"
CC:	VBORMC::"[e-mail]",POUFFARY
Subj:	Re: Submission Informational RFC

Hi Harald,

Thanks for your input. I just came in the office today (I am on vacation till 
next tuesday). I want to briefly answer your questions now. We can discuss 
more when I come back from vacation May 9th.

##1) Has the port 399 been officially assigned, or is it a "grab"?

Yes it has on 19-NOV-1994. See the mail below
+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
From:	VBORMC::"[e-mail]" 19-NOV-1994 01:16:13.90
To:	taec::pouffary
CC:	[e-mail]
Subj:	Re: Request for Well-known TCP port number assignment

Yanick,

We have assigned port number 399 to ISO-TSAP Class 2,
with you as the point of contact.

Joyce
+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

#2) I would *much* prefer that the document start out with "this is the
#   documentation for how DECNET/OSI works over TCP/IP".

Well we could but really this extension is very generic. ISO transport 
identify its user by transport service access point ID (TSAP-ID). 

#- Expedited follows an idea that Marshall used for an 
#  *earlier* RFC1006 version, but abandoned; there is no guarantee that
#  the expedited PDU does not arrive *later* than normal data sent at
#  the same time, and this is required by the ISO protocol

To guaranty delivery of interrupt data the second channel is not torn 
down until the normal data channel is. The need is to have separate virtual 
data channels. Interrupt Data transmission cannot be blocked by a congested 
Normal data channel.

#- Multiplexing is recommended against, but not forbidden; this combined
#  with no explicit flow control means that multiplexing can happen
#  and will lead to some *fun* reactions on congestion

This is why we don't recommended for performance reason to do multiplexing.
But if one want to do it one can always do it.

#- The idea that one can tell from the TPDUs whether this is RFC1006
#  or this new protocol is simply wrong, unless one watches every
#  packet from session establishment onwards

That is not true. Connect-request identify the class and the end-user TSAP-ID.
Once the connection is established no one needs to monitor it.

#- There is no description on what to do if someone suggests that we
#  negotiate down to class 0 on establishment; is this allowed?

This is not allowed. Class negotiation is done according to ISO 8073 rules.
When using TP0 over TCP/IP there is only class 0 can be negotiated.
When using TP2-No-Flow-Control over TCP/IP no other class can be negotiated.
This is because 0 is not an alternate of 2 unless it is explicitly specified
in the connect request TPDU.

#All in all, I feel very good about DEC wanting to document its protocol,
#but I don't like the present submission very much.
#Would it be appropriate to suggest that he publish it as an Internet-draft,
#so that other people can take a look at it to ensure clarity before it
#is published as Informational?

It was published for six months. Sorry if you missed it.

+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
Internet-Draft is available from the on-line Internet-Drafts
directories.

       Title     : ISO Transport Service over TCP RFC1006 extension
       Author(s) : Y. Pouffary
       Filename  : draft-pouffary-tcp-00.txt
       Pages     : 10
       Date      : 10/25/1994

Internet-Drafts are available by anonymous FTP.  Login with the
username "anonymous" and password "guest".  After logging in,
Type "cd internet-drafts".
     "get draft-pouffary-tcp-00.txt".

Internet-Drafts directories are located at:

     o  US East Coast
        Address:  ds.internic.net (198.49.45.10)

     o  US West Coast
        Address:  ftp.isi.edu (128.9.0.32)

     o  Pacific Rim
        Address:  munnari.oz.au (128.250.1.21)

     o  Europe
        Address:  nic.nordu.net (192.36.148.17)

Internet-Drafts are also available by mail.

Send a message to:  [e-mail]. In the body type:
     "FILE /internet-drafts/draft-pouffary-tcp-00.txt".
+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

Looking forward to more cooperation.
Best regards
Yanick

From:	VBORMC::"[e-mail]" "MAIL-11 Daemon"  3-MAY-1995 10:49:42.62
To:	[e-mail]
CC:	[e-mail], taec::pouffary
Subj:	Re: Submission Informational RFC

Jon,
1) Has the port 399 been officially assigned, or is it a "grab"?
2) I would *much* prefer that the document start out with "this is the
   documentation for how DECNET/OSI works over TCP/IP".

There are a number of problems with the protocol described, including:
- Expedited follows an idea that Marshall used for an 
  *earlier* RFC1006 version, but abandoned; there is no guarantee that
  the expedited PDU does not arrive *later* than normal data sent at
  the same time, and this is required by the ISO protocol
- Multiplexing is recommended against, but not forbidden; this combined
  with no explicit flow control means that multiplexing can happen
  and will lead to some *fun* reactions on congestion
- The idea that one can tell from the TPDUs whether this is RFC1006
  or this new protocol is simply wrong, unless one watches every
  packet from session establishment onwards
- There is no description on what to do if someone suggests that we
  negotiate down to class 0 on establishment; is this allowed?

All in all, I feel very good about DEC wanting to document its protocol,
but I don't like the present submission very much.
Would it be appropriate to suggest that he publish it as an Internet-draft,
so that other people can take a look at it to ensure clarity before it
is published as Informational?

             Harald A

% ====== Internet headers and postmarks (see DECWRL::GATEWAY.DOC) ======
% Received: by vbormc.vbo.dec.com; id AA13229; Wed, 3 May 95 10:39:04 +0200
% Received: from domen.uninett.no by mail1.digital.com; (5.65 EXP 4/12/95 for V3.2/1.0/WV) id AA17837; Wed, 3 May 1995 01:45:49 -070
% Received: from dale.uninett.no by domen.uninett.no with SMTP (PP)  id <[e-mail]>; Wed, 3 May 1995 10:38:47 +020
% Received: from dale.uninett.no (localhost [127.0.0.1])  by dale.uninett.no (8.6.9/8.6.9) with ESMTP id PAA03117; Mon, 1 May 1995 15:10:08 -04
% Message-Id: <[e-mail]>
% From: [e-mail]
% To: [e-mail]
% Cc: [e-mail], taec::pouffary
% Subject: Re: Submission Informational RFC
% In-Reply-To: Your message of "Fri, 28 Apr 1995 11:18:51 PDT." <[e-mail]>
% Mime-Version: 1.0
% Content-Type: text/plain; charset="us-ascii"
% Content-Id: <[e-mail]>
% Date: Mon, 01 May 1995 15:10:08 -0400
% Sender: [e-mail]

From:	VBORMC::"[e-mail]" 28-APR-1995 20:28:46.59
To:	[e-mail]
CC:	taec::pouffary
Subj:	Submission Informational RFC


Hi.

The RFC Editor has received this independent submission of a document to
be published as an informational RFC.  However, the subject matter seems
to cover a topic close to the IETFs heart and so the RFC Editor is asking
the IESG to consider reviewing this document.  If the IESG does not raise
any objections in two weeks this will be published as a informational RFC.
The two week time out ends 10-May-95.

--jon.


----- Begin Included Message -----

From [e-mail] Fri Apr 28 09:20:23 1995
Date: Fri, 28 Apr 95 18:05:12 MET DST
From: Yanick <[e-mail]>
To: [e-mail]
Cc: [e-mail]
Subject: Submission Informational RFC 

Request for Comments                                     Yanick Pouffary
April 1995



    ISO Transport Class 2 Non-use of Explicit Flow Control over TCP
                           RFC1006 extension



Status of this Memo

   This memo provides information for the Internet community.  This memo
   does not specify an Internet standard of any kind.  Distribution of
   this memo is unlimited.

Table of Contents

      Status of this Memo.............................................1
      1. Introduction - General recommendations.......................2
      2. The protocol.................................................3
      2.1 TCP service as a Network Service - The Primitives...........3
      2.2 Connection Establishment....................................4
      2.3 Data Transfer...............................................5
      2.4 Connection Release..........................................5
      3. Packet Format................................................6
      4. DIGITAL DECnet over TCP/IP...................................7
      Acknowledgements................................................8
      References......................................................8
      Author' Address.................................................8
























Y. Pouffary                                                     [Page 1]


RFC            ISO Transport Class 2 Non-use of Explicit      April 1995
               Flow Control over TCP - RFC1006 Extension


1. Introduction - General recommendations

   This document is an extension to RFC1006, a standard for the Internet
   community. The document does not duplicate the protocol definitions
   contained in RFC1006 and in International Standard ISO 8073.  It
   supplements that information with the description of how to implement
   ISO Transport Class 2 Non-use of Explicit Flow Control on top of TCP.

   The document should be used in conjunction with the RFC1006 and ISO
   8073.

   The RFC1006 standard defines how to implement ISO 8073 Transport
   Class 0 on top of TCP. This memo defines how to implement ISO 8073
   Transport Class 2 Non-use of Explicit Flow Control on top of TCP.
   Like ISO Transport Class 0, Class 2 Non-use of Explicit Flow Control
   provides basic connection with minimal overhead.

   A Transport protocol class is selected for a particular Transport
   connection based upon the characteristics of the lower layers and the
   requirements of the upper layer. Use of class 2 Non-use of Explicit
   Flow Control is suitable when the use of separate virtual data
   channels for normal and expedited Data are desirable or when an
   explicit disconnection of the Transport connection is desirable.

   Hosts which choose to implement this extension are expected to listen
   on a well-known TCP port number 399.

   It is recommended that the well-known RFC1006 TCP port 102 not be
   used. This recommendation is done to minimise impact to an existing
   RFC1006 implementation.

   The memo also describes the use of this extension within the DIGITAL
   Network Architecture (DNA).


















Y. Pouffary                                                     [Page 2]


RFC            ISO Transport Class 2 Non-use of Explicit      April 1995
               Flow Control over TCP - RFC1006 Extension


2. The protocol

   The protocol specified by this memo is fundamentally equivalent to
   the protocol ISO 8073 Transport Class 2 Non-use of Explicit Flow
   Control, with the following extensions:

   - Expedited Data service is supported.

   - Splitting and Recombining may be used for Expedited Data
     transmission.

   - The Network Service used is provided by TCP.

   The ISO 8073 Transport protocol Class 2 allows Multiplexing. It is
   recommended that this capability not be use for performance reasons.

   The ISO 8073 Transport protocol exchanges information between peers
   in discrete units of information called transport protocol data units
   (TPDUs).  The protocol defined in this memo encapsulates these TPDUs
   in discrete units called TPKTs.  The structure of these TPKTs and
   their relationship to TPDUs are discussed in the next sections.



2.1 TCP service as a Network Service - The Primitives

   The mapping between the TCP service primitives and the service
   primitives expected by ISO 8073 Transport when operation over
   Connection-oriented network service is straightforward.

   Note: The following description of the mapping is a repeat from the
   RFC1006 standard.


   network service                 TCP
   ---------------                 ---
   CONNECTION ESTABLISHMENT

           N-CONNECT.REQUEST       open completes

           N-CONNECT.INDICATION    listen (PASSIVE open) finishes

           N-CONNECT.RESPONSE      listen completes

           N-CONNECT.CONFIRMATION  open (ACTIVE open) finishes

   DATA TRANSFER

           N-DATA.REQUEST          send data


Y. Pouffary                                                     [Page 3]


RFC            ISO Transport Class 2 Non-use of Explicit      April 1995
               Flow Control over TCP - RFC1006 Extension


           N-DATA.INDICATION       data ready followed by read data

   CONNECTION RELEASE

           N-DISCONNECT.REQUEST    close

           N-DISCONNECT.INDICATION connection closes or errors


   Mapping parameters between the TCP service and the network service is
   also straightforward:


   network service                 TCP
   ---------------                 ---
   CONNECTION ESTABLISHMENT

           Called address          server's IP address (4 octets)

           Calling address         client's IP address (4 octets)

           all others              ignored

   DATA TRANSFER

           NS-user data (NSDU)     data

   CONNECTION RELEASE

           all parameters          ignored





2.2 Connection Establishment

   The principles used in connection establishment are based upon those
   described in ISO 8073, with the following extensions.

   - Connection Request and Connection Confirmation TPDUs may negotiate
     the use of Expedited Data transfer using the negotiation mechanism
     specified in ISO 8073.

   - Connection Request and Connection Confirmation TPDUs must not
     negotiate the Use of Explicit Flow Control.

   To perform an N-CONNECT.REQUEST action, the TS-peer performs an
   active open to the desired IP address using a well know TCP port 399.


Y. Pouffary                                                     [Page 4]


RFC            ISO Transport Class 2 Non-use of Explicit      April 1995
               Flow Control over TCP - RFC1006 Extension


   When the TCP signals either success or failure, this results in an
   N-CONNECT.INDICATION action.

   To await an N-CONNECT.INDICATION event, a server listens on a well
   know TCP port 399.  When a client successfully connects to this port,
   the event occurs and an implicit N-CONNECT.RESPONSE action is
   performed.



2.3 Data Transfer

   The elements of procedure used during transfer are based upon those
   presented in ISO 8073, with the two following extensions.

   - Expedited Data may be supported (if negotiated during connection
     establishment).

     In Non-Use of Explicit Flow Control Expedited Data require no
     Expedited Data Acknowledgement.

   - Splitting and Recombining may be used for Expedited Data
     transmission.

     The procedure of Splitting and Recombining allows a transport
     connection to make use of multiple TCP connections.
     TCP connections created for Splitting purposes should also use
     the primitives described in 2.1.

     It is recommended to only create a second TCP connection for
     Expedited Data when transmission of Expedited Data is requested.

     Expedited Data must only be sent over an outgoing TCP connection.
     This second TCP connection must not be shared among transport
     connections and must remain established until the transport
     connection is terminated, at which time it must be closed.

   To perform an N-DATA.REQUEST action, the TS-peer constructs the
   desired TPKT and uses the TCP send data primitive.

   To trigger an N-DATA.INDICATION action, the TCP indicates that data
   is ready and a TPKT is read using the TCP read data primitive.



2.4 Connection Release

   The elements of procedure used during a connection release are
   identical to those presented in ISO 8073.


Y. Pouffary                                                     [Page 5]


RFC            ISO Transport Class 2 Non-use of Explicit      April 1995
               Flow Control over TCP - RFC1006 Extension


   Disconnect Request and Disconnect Confirmation TPDUs are exchanged.
   Upon receipt of a Disconnect Confirmation TPDU a N-DISCONNECT.REQUEST
   action is performed to simply close the TCP connection.

   If the TCP connection has been closed or has failed, this generates
   an N-DISCONNECT.INDICATION event.



3. Packet Format

   A fundamental difference between TCP and the network service expected
   by ISO transport is that TCP manages a continuous stream of octets,
   with no explicit boundaries.

   The protocol described in RFC1006 uses a simple packetization scheme
   in order to delimit TPDUs. Each packet, termed a TPKT, consists of
   two parts: a packet-header and a TPDU.

   We use the same scheme described in RFC1006 for this extension.
   There is no need to change the version number. The ISO transport TPDU
   sufficiently describes the transport protocol class being used.

   The format of the packet-header described below is a repeat from
   RFC1006.

           0                   1                   2                   3
           0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
          +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
          |      vrsn     |    reserved   |          packet length        |
          +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

         where:

         vrsn                         8 bits

         This field is always 3 for the version of the protocol described in
         this memo.

         packet length                16 bits (min=7, max=65535)

         The packet length is the length of the entire packet in octets,
         including packet-header.


   The format of the ISO transport TPDU is defined in ISO 8073.





Y. Pouffary                                                     [Page 6]


RFC            ISO Transport Class 2 Non-use of Explicit      April 1995
               Flow Control over TCP - RFC1006 Extension


4. DIGITAL DECnet over TCP/IP

   DECnet over TCP/IP is implemented using the DECnet Session Control
   layer over this RFC1006 extension protocol.

   The informational RFC defined in this document provides the Transport
   Service functionality required by DECnet Applications while operating
   over TCP/IP.

   The next paragraph is a brief summary of the role of the DECnet
   Session Control Layer. For further details, refer to the DIGITAL DNA
   Session Control Layer Specification.

   The DECnet Session Control Layer makes a Transport Service available
   to End Users of a network. This layer is concerned with system-
   dependent functions related to creating, maintaining, and destroying
   Transport Connections.  Separate virtual data channels, known  as
   "Normal"  and  "Expedited",  are provided to End Users. DECnet
   Session Control must be guaranteed independence of these channels by
   the Transport Layer. Expedited Data transmission cannot be blocked by
   a congested normal data channel. DECnet Session Control requires that
   all data in transit be delivered before initiating the release of the
   Transport Connection.

   DECnet, DNA, and the DIGITAL logo are trademarks of Digital Equipment
   Corporation.

























Y. Pouffary                                                     [Page 7]


RFC            ISO Transport Class 2 Non-use of Explicit      April 1995
               Flow Control over TCP - RFC1006 Extension


Acknowledgements

   Bill Duane, Jim Bound, David Sullivan, Mike Dyer, Matt Thomas, Dan
   Harrington and many other members of the DECnet engineering team.



References

   [ISO8072]  ISO. "International Standard 8072.  Information Processing
   Systems -- Open Systems Interconnection: Transport Service
   Definition."

   [ISO8073]  ISO. "International Standard 8073.  Information Processing
   Systems -- Open Systems Interconnection: Transport Protocol
   Specification."

   [ISO8327]  ISO. "International Standard 8327.  Information Processing
   Systems -- Open Systems Interconnection: Session Protocol
   Specification."

   [RFC791]   Internet Protocol. Request for Comments 791 (MILSTD 1777)

   [RFC793]   Transmission Control Protocol. Request for Comments 793
   (MILSTD 1778)

   [RFC1006]  ISO Transport Services on Top of the TCP. Request for
   Comments 1006



Author' Address

       Yanick Pouffary
       End Systems Networking             Phone: [phone]
       Digital Equipment Corporation      Fax:   [phone]
       Centre Technique (Europe)          Email: [e-mail]
       B.P. 027
       950 Routes des colles
       06901 Sophia antipolis, France











Y. Pouffary                                                     [Page 8]


----- End Included Message -----


% ====== Internet headers and postmarks (see DECWRL::GATEWAY.DOC) ======
% Received: by vbormc.vbo.dec.com; id AA08060; Fri, 28 Apr 95 20:19:00 +0200
% Received: from venera.isi.edu by mail1.digital.com; (5.65 EXP 4/12/95 for V3.2/1.0/WV) id AA20059; Fri, 28 Apr 1995 11:22:40 -070
% Received: from zen.isi.edu by venera.isi.edu (5.65c/5.61+local-21) id <AA20419>; Fri, 28 Apr 1995 11:18:33 -070
% Date: Fri, 28 Apr 1995 11:18:51 -0700
% From: [e-mail]
% Posted-Date: Fri, 28 Apr 1995 11:18:51 -0700
% Message-Id: <[e-mail]>
% Received: by zen.isi.edu (5.65c/4.0.3-4) id <AA09355>; Fri, 28 Apr 1995 11:18:51 -070
% To: [e-mail]
% Subject: Submission Informational RFC
% Cc: taec::pouffary