Re: State Alignment and Standard Signals

MessageHi Jean-Jacques

JJ: So here is a scenario that could happen in business (either by mistake or wrongfully):

- You send me an invoice and I don't "respond". Then I call you and you claim you never received it. How would you strategy work in that case?


This would simply be by using an acknowledgement mechanism - my example was just illustrating that we could use the observed behaviour of the fact that the receiving participant has now performed the next task in sequence, and thereby implicitly acknowledged receipt of the message. If there is a significant time period between the receiving participant getting the message and then sending its response, then it would be for the CDL designer to ensure there was an additional 'acknowledgement' message prior to the more significant response.


JJ: - You can also have (very common scenarios - see my article: http://www.ebpml.org/state.htm article) the case where the response can only come, days or weeks after the request (e.g. an RFQ). It is a very good practice to indicate that you received the RFQ and you were able to process it, it does not mean that you are in the position to respond just yet. Imagine how disastrous it could be for both parties if the RFQ fell through the cracks and you waited on a timeout on the response to detect it.


The main point I wanted to make from David's previous email was that I don't think we should be encoding legal significance of a message in CDL, especially not as different message types. I think this is a business level consideration that should be catered for in the business protocol - precisely because of the potential delays between different stages of the transaction - or the fact that the business transaction may be performed across multiple choreographies.


Regards
Gary

  ----- Original Message ----- 
  From: Jean-Jacques Dubray 
  To: Gary Brown ; david.burdett@commerceone.com ; tony.fletcher@choreology.com ; public-ws-chor@w3.org 
  Cc: Robin.Milner@cl.cam.ac.uk ; kohei@dcs.qmul.ac.uk ; yoshida@doc.ic.ac.uk 
  Sent: Thursday, July 15, 2004 5:26 PM
  Subject: RE: State Alignment and Standard Signals


  Gary:

   

  The "paper world" is really hard to emulate electronically. It is far easier to forge an XML text file than a piece of paper. It is also very easy to blame something like an Enigmatec infrastructure for not having delivered an XML document to my ERP system than the postmaster for not delivering a registered piece of mail. 

   

  So here is a scenario that could happen in business (either by mistake or wrongfully):

  - You send me an invoice and I don't "respond". Then I call you and you claim you never received it. How would you strategy work in that case?

  - You can also have (very common scenarios - see my article: http://www.ebpml.org/state.htm article) the case where the response can only come, days or weeks after the request (e.g. an RFQ). It is a very good practice to indicate that you received the RFQ and you were able to process it, it does not mean that you are in the position to respond just yet. Imagine how disastrous it could be for both parties if the RFQ fell through the cracks and you waited on a timeout on the response to detect it.

   

  So at the end of the day, we are either in the business of Distributed computing or in the business of B2B, but applying DC tricks to B2B will lead to a vastly suboptimal solution. When you apply DC internally, you tend to detect all these situations and find an ad-hoc solution. When you problem is to get a 1000 or 10000 companies to interoperate over some message interchanges definition, you cannot afford to have little tricks figured two by two. You need a holistic solution.

   

  I hope you understand a little more the value a little signal can have.

   

  CQFD.

   

  Jean-Jacques

   


------------------------------------------------------------------------------

  From: public-ws-chor-request@w3.org [mailto:public-ws-chor-request@w3.org] On Behalf Of Gary Brown
  Sent: Wednesday, July 14, 2004 11:41 PM
  To: david.burdett@commerceone.com; tony.fletcher@choreology.com; public-ws-chor@w3.org
  Cc: Robin.Milner@cl.cam.ac.uk; kohei@dcs.qmul.ac.uk; yoshida@doc.ic.ac.uk
  Subject: Re: State Alignment and Standard Signals

   

  Hi David

   

  As Monica pointed out in a previous email, "we are implementing the technical interactions, as the 
  business aspects are outside of WS-CDL".

   

  I would suggest that the legal status of a message (signal/acknowledgement) is a business level consideration. For example, if participant A sends a request to participant B, and the CDL defines a sequence that indicates that a response is then sent from B to A following the receipt of this request, then that implies participant B has received and processed the message.

   

  <sequence>

      <interaction A->B />

      <interaction B->A />

  </sequence>

   

  As Participant B has sent a follow-up communication that could only have resulted as a consequence of receiving the request, it would not be able to argue that it hadn't received the request.

   

  I am not sure what value defining a specific message type would add.

   

  Regards

  Gary

   

   

  ----- Original Message ----- 

    From: david.burdett@commerceone.com 

    To: tony.fletcher@choreology.com ; public-ws-chor@w3.org 

    Cc: Robin.Milner@cl.cam.ac.uk ; kohei@dcs.qmul.ac.uk ; yoshida@doc.ic.ac.uk 

    Sent: Thursday, July 15, 2004 1:24 AM

    Subject: State Alignment and Standard Signals

     

    Tony

     

    I thought it might be worthwhile putting on record the comment I made on the call on Tuesday in that I think that there are two different "state alignment" problems to be solved: "real" state alignment, and standard signals.

     

    REAL STATE ALIGNMENT

    The first is the problem you discuss below that there is a general requirement at various points in a choreography that each participant has a common shared understanding of the state of the other participants in terms of the messages that have (or have not) been received ... I hope I am paraphrasing/simplifying your requirement as described below.

     

    STANDARD SIGNALS

    The second is the problem that Anders Tell described in his email [1] that I responded to in my email [2]. Anders talks about a need that a recipient of a message "legally" accepts that he has received the message. In this case, the message is more like a signal that informs the sender of the original message what their "state" is with respect to legal acceptance of the message. There are also other signal messages that can occur, for example to indicate that a message has been received, i.e. a simple Ack, or has been "Accepted for Processing" as standards like BPSS suggest.

     

    SOLVING THE PROBLEMS

    To solve the problem you are suggesting, then we need to continue discussing the state alignment approaches you describe below.

     

    To solve Anders problem and also the issues I think Jean-Jacques was raising, we could define some "standard" Message Content Types that have specific semantics, message flow patterns and behaviors associated with them and then recommend use of these standard types when choreography designers have a need to use them. Note that these would specify the types and not the representation of those messages in XML which could potentially be done in different ways.

     

    Thoughts?

     

    David

     

    [1] http://lists.w3.org/Archives/Public/public-ws-chor/2004Jul/0009.html

    [2] http://lists.w3.org/Archives/Public/public-ws-chor/2004Jul/0012.html

      -----Original Message-----
      From: public-ws-chor-request@w3.org [mailto:public-ws-chor-request@w3.org]On Behalf Of Tony Fletcher
      Sent: Tuesday, July 13, 2004 3:56 AM
      To: public-ws-chor@w3.org
      Cc: Robin Milner; 'Kohei Honda'; Nobuko Yoshida
      Subject: Tony's nightmare - wake me up please

      Dear Colleagues,

       

      I hope this is a nightmare and someone can wake me up and re-assure me that everything is all right really!  I know that Nick (who could answer this concern as one of our main language designers) is on holiday for awhile now, so I have copied this message to your resident experts as well, though Steve, Gary or somebody else may well be able to answer simply.  I started writing this email a few days ago before Gary Brown's message on 'Perform' came out.  Having now had a look at that I wonder if my concern is a more extreme form of his.

       

      Following discussions with Peter last week I have a major concern about how we are designing the choreography description language.

       

      I am no mathematician, but my understanding is that pi-calculus describes each 'node' (a partner according to our current definitions) as a process and the allowed message sequences (I think mathematicians call these 'traces') are generated by mathematically 'composing' each node process with the others and letting the maths tell you what the various allowed message exchanges are (etc.)

       

      Now I am more familiar with state machines, and I will therefore continue the discussion based on these, but I think the same principles apply to process algebras.

       

      Consider a relationship between two nodes that we wish to describe.  The usual way to describe this fully suing state machines is to develop a state machine for each end and see how they interact.  Suppose one state machine has N states and the other M (N and M will be close to the same integer value but not necessarily equal).  So we have had to use N + M (approx 2*N) states to describe this relationship precisely.  However, there are altogether N*M (approx N squared) states that the system can be in.  Now depending on the whether each of our state machines are relatively simple and only have entries on the diagonal cells (or nearly so) or a very complicated with entries in most cells the number of valid, reachable states may be near to 2N or to N squared).  The answer is usually a lot more than 2N even if less than the worst case.

       

      Now it is possible, in principle (I think) to have a single state machine that describes the permitted message sequences for the relationship but this is usually only attempted for very simple cases due to the state 'explosion' described above.  This single state machine has to represent all the valid states which is between N+M and N*M and usually well above N+M for any 'interesting' protocol.

       

      In the current version of CDL we do not seem to be describing the process at each node then letting these interact with the others, but describing the interactions directly.  My concern is that this approach will suffer from state space explosion when one tries to include every conceivable message sequence that could happen, and that actually the current CDL may be fine for describing message sequence charts in XML (which is useful but not I thought what we were attempting to achieve) which are fine for illustrating message flows, but do not, in general, cover every possible case the protocol designer needs to allow for, but will become unwieldy / impossible to use for a complete description.

       

      I hope the point is clear and that someone can answer it.

       

      Best Regards     Tony

      A M Fletcher

       

      Cohesions  (TM)

       

      Business transaction management software for application coordination       www.choreology.com

       

      Choreology Ltd., 68 Lombard Street, London EC3V 9LJ     UK

      Tel: +44 (0) 1473 729537   Fax: +44 (0) 870 7390077  Mobile: +44 (0) 7801 948219

      tony.fletcher@choreology.com     (Home: amfletcher@iee.org)