RE: Reliable Messaging Spec v0-078

[Ajay K Sanghi <sanghi@giasdl01.vsnl.net.in>]

Hi Rik,

This is very addicting. I don't think I have earned my bread today ;-)
Following are my post lunch thoughts:

A typical connection oriented protocol (barring PVCs) would involve:
#. Connection establishment phase
#. Data transfer phase
#. Disconnection phase

issues:
#. asynchronous/synchronous mode of communication
#. handshake mechanism for establishing connection, assigning port numbers,
session ids etc.
#. bi-directional (full-duplex) mode of communication
#. mechanism for flushing the buffered data to lower protocol
#. reliability (ordering + retries and timeouts) to be leveraged from TCP
#. flow control to be leveraged from TCP
#. multiplexing of ebXML messages to be achieved using different TCP
connections
#. ability to send command during normal data transfer phase
#. ability to negotiate parameters, listen, accept, send request, bind to
port, assign a data structure which can be referenced for this connection
etc
#. message security issues
#. binding static/dynamic port number to ebXML listener/application
#. ability of the application to query on the status of the connection
#. resync and recovery of lost ebXML sessions
#. facilities like reverse charging, d-bit, q-bit and fast select as in X.25
#. ebXML header should be different in call establishement phase and
different during data transfer phase. an ebXML id (like SVC in case of X.25)
may be used to identify the ebXML connection
#. gateways should not have to read the entire data first to relay/transmit
it to other intermediate gateway, unless a new security stamp is to be
created for the entire data
#. a state transition diagram needs to be part of the specs
#. sequence number (if used) should be reset to 1 for each new ebXML
session. it should not use any timer or no-activity for it as currently in
the specs. i think, TCP uses something like this mainly because time to live
for a segment may not have expired when it recovers from error cases but
this is not applicable to ebXML scenario

Products: -
#. companies may develop only the ebXML resync and recovery protocol stack
as a full blown product. by this i also mean that providing resync and
recovery is non-trivial and a specialis(z)ed development
#. companies like HP may develop ebXML protocol analyser so that the
communication can be monitored and debugged

Parting thought:
How much will such a system cost?

Regards,

Ajay

> -----Original Message-----
> From: Ajay K Sanghi [mailto:sanghi@giasdl01.vsnl.net.in]
> Sent: Thursday, September 28, 2000 1:32 PM
> To: Rik Drummond; ebxml-transport@lists.ebxml.org
> Subject: RE: Reliable Messaging Spec v0-078
>
>
>
> Hi Rik,
>
> The neck-break speed with which the ebXML team is working is
> simply amazing. I have been finding it difficult to keep upto date.
>
> > we decide to make this transport independent.
>
> making it transport independent is fine, but providing
> reliability over unreliable transports like UDP, in my opinion
> should be dropped. I don't think that RM specs should build "flow
> control", which it will require if reliability is to be supported
> over unreliable transports. "sequence" numbering is a different issue.
>
> >i have been
> > thinking about it
> > work directly over tcp/ip, as we are now doing with smtp and
> > http... what do
> > you think of that? and are there any concerns?
>
> I assume, you mean making ebXML as first class application (or
> application protocol) over TCP/IP, just like SMTP, HTTP and
> issues/concerns associated with the same.
>
> Yes, I would like to see ebXML sit directly on top of TCP/IP (at
> later date), with SMTP as an excellent starter.
>
> I noticed Martin using the term "logical channel" in response to
> Sanjay's email, which is a key point. In my earlier mail, I
> suggested going through BX.25 specs ("B" stands for Bell Labs),
> which are very comprehensively done - used by AT&T for their
> NEMOS and 800 DCS project (that was 1992). Basically, one LAPB
> link can contain multiple X.25 virtual circuits (logical
> channels) and each virtual circuit is dedicated to one BX.25
> session (I think - it has been a while I worked on it). Over and
> above all the reliability that X.25 provides, BX.25 was mainly
> used for session recovery (if lost virtual circuit) and
> resynchronisation (continue from the point where it let off). Key
> thing leveraged from X.25 was "flow control" besides
> "reliability". ebXML should also leverage flow control and
> reliablity from TCP.
>
> Following is just loud thinking-
> Assumption: well known port for ebXML is defined (say 1000).
> Context: <sender ip, reciever ip, sender port number, receiver
> port number> forms unique combination to connect to an instance
> of application
> Scenario:
> One ebXML receiver (server) is listening on port 1000
> One sender sending 2 ebXML messages concurrently,
> case 1: on 2 different sender port numbers
> case 2: on 1 (same) sender port number but the 2 ebXML messages
> are multiplexed
>
> case 1 does not have the problem associated with flow control.
> i.e. one ebXML  session blocking the other
>
> case 2 would involve building flow control in the ebXML RM layer
> so that the other ebXML message is not blocked - defining flow
> control would increase complexity, besides invloving negotiation
> at application protocl layer for the window size, etc.
>
> more cases can also be introduced like the receiver re-connecting
> on a different port for each ebXML message once the ebXML
> listener identifies a ebXML message request...
>
> Issues:-
> 1. should ebXML RM support auto resyncronisation and recovery
> after the session has been timed-out.
> 2. if yes, how to identify the resynchronisation point. resync
> should be  supported for "best effort" cases also, once the
> decision is yes (may be as optional feature).
> 3. sequence number is ok for resync purpose but then a
> "session-id" (logical channel) should also get added. who
> allocates the session-id? sender or receiver? since it's going to
> be common to both. Also, some more resync protocol messages have
> to be added.
> 4. Who initiates the recovery and resync process (sender or
> receiver or combination of both)? for example,sender recovers but
> receiver resyncs. Note recovering has to do with only
> identification of the lost session, while resync has to identify
> the point within the session from where data transfer is to begin.
> 5. should ebXML RM assume window size of 1, or should a window be
> defined. I think, window complexity should not be added.
> 6. does session establishment involve any negotiation for QoS.
> Can receiver deny "at most once" effort to downgrade the session
> to "best effort".
> 7. Delivery confirmation can be achieved even without sequence
> numbering but for resynchronisation, sequence numbering  (or some
> identification) with every packet is required
> 8. I find timing-out at RM level quite dicy, without the
> knowledge of timers at lower level as mentioned in my earlier
> email. I think retrying to send the data at RM level should be
> replaced by re-establishing for resync and recovery in case that
> is to be supported.
> 9. arhcitecturally, it seems there would be one ebXML listener
> and each call reqquest would be handled by a different thread
>
> these are some of the points that i can immediately think of...
>
> Thanks and regards,
>
> Ajay
>
>
>
> > -----Original Message-----
> > From: Rik Drummond [mailto:rvd2@worldnet.att.net]
> > Sent: Wednesday, September 27, 2000 9:53 AM
> > To: Ajay K Sanghi; ebxml-transport@lists.ebxml.org
> > Subject: RE: Reliable Messaging Spec v0-078
> >
> >
> > Ajay. thanks for the nice comments on the draft everyone has
> been working
> > their butts off...
> >
> > we decide to make this transport independent. i have been
> > thinking about it
> > work directly over tcp/ip, as we are now doing with smtp and
> > http... what do
> > you think of that? and are there any concerns?
> >
> > best regards, rik
> >
> > -----Original Message-----
> > From: Ajay K Sanghi [mailto:sanghi@giasdl01.vsnl.net.in]
> > Sent: Monday, September 25, 2000 6:46 AM
> > To: Jim Hughes; ebxml-transport@lists.ebxml.org
> > Subject: RE: Reliable Messaging Spec v0-078
> >
> >
> >
> > Hi Jim,
> >
> > Since this is my first communication with the ebXML group, I take this
> > opportunity to congratulate ebXML team for excellent work being
> done. In a
> > very recent Bill Gates talk in New Delhi, I asked him the question of
> > Microsoft's position w.r.t ebXML, particularly in relation with
> Biztalk. I
> > don't hink he has heard of ebXML, unless he was unable to understand my
> > Indian accent ;-) All he said was that Interoperabilty is more important
> > (agreed) and that XML will facilitate that (he was speaking on
> Microsoft's
> > .NET strategy).
> >
> > I am reading Sep 22, 2000 spec for ebXML TRP Reliable Messaging.
> > I feel that
> > Sequence Numbering is NOT required for achieveing Reliablity at
> ebXML MSH
> > layer. We must leverage reliablilty provided by the protocol themselves.
> >
> > Assumption for Reliable Messaging should only be that it should use a
> > (reliable) connection oriented protocol (CONS like X.25, TCP - they are
> > reliable, meaning "ordering" is guaranteed). However, this
> reliablity does
> > not "guarantee" that the ebXML MSH has received the message, for which a
> > "delivery confirmation" element/attribute may be added instead
> of sequence
> > number. Whenever delivery confirmation element/attribute is set
> > to YES, the
> > receiving MSH will respond with ebXML message (just the ebXML header,
> > confirming receipt). Note (Key Point) - on account of this
> > message (delivery
> > confirmation request message) riding on top of CONS, a response to this
> > message would necessarily mean that all previously sent
> messages has been
> > received. Infact, if the recieving MSH is sending back some data, this
> > delivery confirmation response can be piggybacked. There is
> absolutely no
> > need for persisting packets as suggested in the specs. This delivery
> > request/response can be thought of as a ebXML-MSH-ping.
> >
> > Suggestion for Timeout/Retry is ok provided MSH-timeout is of greater
> > interval than what it is at the tranport level
> (tranport-retries*timeout),
> > otherwise it will cause grieve. If retires are to be done by MSH, then
> > sequence numbering "may" be required for checking duplicates
> (not for the
> > purpose of delivery confirmation), since for the sending
> transport layer,
> > it's fresh data. In first phase of ebXML specs, I'll be tempted
> > to drop this
> > idea too - the logic being that if it can't be delivered even with
> > transport's retries and timeout, then there is a good chance it won't be
> > delivered with ebXML-MSH-retries as well. Sending application
> > should simply
> > abort and send the message at some other time.
> >
> > I am in favor of ebXML-MSH, but it should only be used to enable certain
> > services (as glue) that CONS transport is supporting or to
> build something
> > on top of what reliable transports support, if absolutely required. For
> > example Q-ualified data (used for out of band signalling in X.25) can be
> > used for signalling End of File for file transfers. Alternate to
> > Q-data way
> > of sending End of File would be to first send the size of file.
> > The Delivery
> > confirmation (D-bit) logic in X.25 protocols should be looked
> > into. Also, of
> > interest may be some other attributes like fast-select (sender sends
> > connection request along with small data and the receiver sends
> back data
> > along with call termination - verifone (I think) used to use this
> > for credit
> > card verification)
> >
> > However, if the goal is to provide Relibility over Unrealible
> > connectionless
> > protocols (CLNS - like UDP) then use of sequence numbering is ok.
> > I suggest
> > that X.25 protocol is looked at to see the problems assoicated
> > with sequence
> > numbering, especailly in case MSH is queuing data and not sending
> > it to the
> > application (see how RNR is used). My suggestion is that ebXML
> should not
> > address reliability over unreliable protocols.
> >
> > Just a suggestion.
> >
> > Thank you and regards,
> >
> > Ajay
> >
> > Ajay K Sanghi
> > Managing Director
> >
> > ABO Software Private Limited
> > B102 Gulmohar Park, New Delhi 110049
> > Tel: +91 11 6512816, 6512822 Fax: 6518873
> > Website: http://www.abosoftware.com
> > email: ajay@abosoftware.com
> >
> > > -----Original Message-----
> > > From: Jim Hughes [mailto:jfh@fs.fujitsu.com]
> > > Sent: Sunday, September 24, 2000 4:30 AM
> > > To: ebxml-transport@lists.ebxml.org
> > > Subject: Reliable Messaging Spec v0-078
> > >
> > >
> > > Attached are Word and PDF copies of the latest version of the Reliable
> > > Messaging Spec, for discussion on Wednesday in the F2F.
> > Shimamura-san has
> > > added much more detail on reliability issues, and of course the
> > > earlier use
> > > of Message Groups is now deleted for simplicity. Since there were many
> > > changes, I deliberately did not mark changes in this document,
> > > but you can
> > > easily see them by using Word's compare utility against the
> > > previous version.
> > >
> > > [Ralph, could you have some copies of the document available
> on Tuesday,
> > > for those that might not have access to a printer before
> > > travelling to Dallas?]
> > >
> > > Jim
> > >
> >
> >