David,
I do not think that any illustrative remarks Marty might
have provided about how a MSH
might store its configuration data forms a
basis for showing some technical shortcoming
in CPPs or CPAs.
The essential point is that the CPP and CPA schemas define
standardized patterns to exchange(!) collaboration protocol
profiles and agreements. The internal or distributed "storage"
of that information (along with other configuration information
the MSH will need) is not something we are specifying. So
the alternative storage options you raise
are irrelevant to the basic purpose
of CPPs and CPAs-- which is the specification of standardized
ways to exchange information about collaboration protocols.
This standardization is hoped to be one step in a continuing
effort to make configuration of collaborations more automated
and less onerous (less costly) for IS departments.
Likewise, while we do in effect insist that there be one
"primary key" to a collaboration profile record in its internal
representation (whatever that might be) which is, of
course, the cpaid, we do not assert that the cpaid is the only
primary key, or even that the cpaid is the primary key
that any MSH must use as a primary key
in its implementation.
So I do not believe that your point that we are somehow constraining
software implementations is one that has a very substantial foundation.
There may be reasons to not be interested in trying to make
peer-to-peer collaboration configuration easier. One would be that
you are interested in alternative non peer-to-peer architectures based
on "nanny intermediaries." That might explain a lack of interest
in CPPs and CPAs, though I think even nanny architectures will
ultimately
benefit from things similar to CPPs and CPAs, such as the descendents
of WSDL, or the joint offspring of our pioneer CPPA and WSDL ancestors.
But I do not think that your lack of interest in such technologies
should be regarded as a technical problem to solve by those of us who
are interested in pursuing these technologies.
Dale Moberg
-----Original Message-----
From: Burdett, David
Sent: Wed 7/25/2001 9:50 AM
To: 'Martin W Sachs'
Cc: 'Arvola Chan'; David Fischer; ebXML Msg; Pete Wenzel
Subject: RE: PIP IDs
Marty
Apologies for the delay in responding ... I've been busy
elsewhere.
I agree with everything you say about how databases work and how
they
**could** be used to solve the problem. It's just that you are
pre-supposing
a particular internal design for a solution when there are other
alternatives. I also disagree that the information **has** to be
entered
into each partners system.
For example, a community of users, e.g. RosettaNet could define
a set of
three or four standard "agreements" that everyone should use for
messaging
which exclude partner specific information, e.g. URLs and
certificates. The
URL and other information could then be recorded in a registry
(UDDI?) which
a trading partner could look up to determine where to send a
message. This
information could also be cached.
Once the URL was determined, then a message could be sent where
the CPAId
referenced one of the "standard" agreements. If the sender of
the message
was not previously to the message recipient then they could look
up the
information in the registry and decide what to do with the
message.
Marty, I am not saying that your approach is wrong, it's just
that there are
alternaive equally valid approaches which the current definition
of the CPA
does not easily support. I think we should support both. Do you
agree?
David
-----Original Message-----
From: Martin W Sachs [mailto:]
Sent: Friday, July 20, 2001 7:17 AM
To: Burdett, David
Cc: 'Arvola Chan'; David Fischer; ebXML Msg; Pete Wenzel
Subject: RE: PIP IDs
David,
The usual approach to digesting information from CPAs is to have
a database
that has an entry for each CPA. Hence your problem about all
the messages
from company A to company B going to the same endpoint doesn't
exist. For
each message, the middleware will look at the CPAId and get the
right
information out of the database. Of course, once a conversation
starts, a
smart system will cache the configuration information for that
conversation
so it doesn't have to go back to a disk-based database on every
message.
Database technology has existed for generations that can execute
query or
an update against multiple selected rows.
The rule in your second paragraph is certainly support when all
those
messages are associated with the same CPA. For the more general
case that
you have in mind, as I said, database technology is well up to
it. ...and
don't forget that the changes you have in mind are very
infrequent compared
to the frequency of messages in a large, active system.
Your third paragraph: I disagree. The current perception is
that whether
there is a CPA or not, the configuration information has to be
entered into
each partner's system and managed there. Some people believe
that the CPA,
with suitable tools, is a lot easier to use than the manual
methods.
************************************************************************
****
*********
Martin W. Sachs
IBM T. J. Watson Research Center
P. O. B. 704
Yorktown Hts, NY 10598
914-784-7287; IBM tie line 863-7287
Notes address: Martin W Sachs/Watson/IBM
Internet address: mwsachs @ us.ibm.com
************************************************************************
****
*********
"Burdett, David" <> on 07/19/2001
04:51:03 PM
To: Martin W Sachs/Watson/IBM@IBMUS
cc: "'Arvola Chan'" <>, David Fischer
<>, ebXML Msg
<>, Pete Wenzel
<>
Subject: RE: PIP IDs
Marty
I agree that a sensible CPA tool should digest the CPA into a
table that
allows look up in an efficient manner. The problem is that I am
not sure
that it would (or could) actually do it. For example, if you
wanted to use
a
very simple routing rule, e.g. "all the messages that company
xyz sends to
IBM go to one URL", then the CPA processor would have to
**infer** this
rule
from analyzing all the CPAs that define the individual
collaborations
between xyz co and IBM. This type of inference processing is
both hard to
do
and un-reliable since the inference processor cannot assume that
the next
CPA it receives will not follow the rule. Secondly, if you
wanted to change
the URL for all these messages then you would have to change all
the CPAs
rather than just change the single simple rule.
To solve this problem, you need to be able to express
**directly** a rule
in
its simplest form, (e.g. all messages from xyz co for IBM go to
one URL).
However the current CPA structure doesn't support this type of
rule
definition (Is that correct?).
Finally, the current perception of a CPA is, IMO, that it is the
**only**
way to specify the information required for ebXML Messaging to
work when
clearly there could be more appropriate ways of specifying the
information
for some circumstances. Hence my reasons for wanting to be able
to use
ebXML
Messaging without CPAs.
Finally, I do think that CPAs are a good idea especially where
you have
point-to-point communication, they're just not the only good
idea ;)
Regards
David
-----Original Message-----
From: Martin W Sachs [mailto:]
Sent: Thursday, July 19, 2001 12:48 PM
To: Burdett, David
Cc: 'Arvola Chan'; David Fischer; ebXML Msg; Pete Wenzel
Subject: RE: PIP IDs
David,
Please forgive my repetition, but...
Your two points:
look up the CPA from the CPA ID and route the message to
the endpoint
identified, or
look up the Party, Service and Action in a table to identify
where to
send the message to.
are really one and the same since any sensible CPA tool will
digest the CPA
into a table that allows looking up the routing information in
the most
efficient manner. The only difference between them is where the
table
comes from.
Regards.
Marty
************************************************************************
****
*********
Martin W. Sachs
IBM T. J. Watson Research Center
P. O. B. 704
Yorktown Hts, NY 10598
914-784-7287; IBM tie line 863-7287
Notes address: Martin W Sachs/Watson/IBM
Internet address: mwsachs @ us.ibm.com
************************************************************************
****
*********
"Burdett, David" <> on 07/19/2001
03:17:41 PM
To: "'Arvola Chan'" <>, Martin W
Sachs/Watson/IBM@IBMUS,
David Fischer <>
cc: ebXML Msg <>, Pete Wenzel
<>
Subject: RE: PIP IDs
Thanks Arvola !! You have saved me from writing an email. I
absolutely
concur with what you say ... you can have one "service"
supporting multiple different business transactions. If we do
this it also
means that we can have one conversation involving multiple
different
business transactions, e.g. a 3A4 (Create Purchase Order)
followed by a
3A8 (Change Purchase Order) where the change purchase order was
for the
order created by the create purchase order.
Another use case is where you can have the same service used in
multiple
different collaborations for example consider the following two
collaborations:
Example 1, Invoice Payment:
1. BT1 - Present Invoice
2. BT2 - Make Payment
Example 2, Pay Refund:
1. BT1 - Request Refund
2. BT2 - Make Payment
Both of these involve making a payment which, in Business
Transaction
terms, would be carried out the same way.
Where I think it gets really critical to separate the two is
when you want
to route messages. If Party, Service and Action are always the
same,
irrespective of the business process in which they are being
used then you
can use these three fields for routing purposes in a consistent
way.
On reflection this I think is one of the issues I have with
always
requiring a CPA as defined in the CPA/CPP spec as you can
actually quite
reasonably do routing in two different ways:
look up the CPA from the CPA ID and route the message to
the endpoint
identified, or
look up the Party, Service and Action in a table to identify
where to
send the message to.
The former allows you to vary the endpoint very flexibly and
alter where
you send a message to depending on the business process you are
in, the
latter allows you to have simpler rules but with less
flexibility.
I think both use cases are valid ... and required ..
David
-----Original Message-----
From: Arvola Chan [mailto:]
Sent: Thursday, July 19, 2001 9:04 AM
To: Martin W Sachs; David Fischer
Cc: Burdett, David; ebXML Msg; Pete Wenzel
Subject: Re: PIP IDs
Marty,
Up to now, RosettaNet PIPs are either request-response
(two-actions) or
notification (one-action) style business processes. Earlier
versions of
PIP 3A4 are an exception in the sense that PIP 3A4 covers
Create Purchase
Order (request-response), Change Purchase Order
(request-response) and
Cancel Purchase Order (request-response) interactions.
Recently, PIP 3A4
has been split into 3A4 (Create Purchase Order), 3A8 (Change
Purchase
Order), and 3A9 (Cancel Purchase Order) in order to achieve
some degree of
uniformity across PIPs (I believe). Therefore, I think it is
reasonable to
equate existing RosettaNet PIPs with BPSS Business
Transactions.
In the RosettaNet message header, there are separate elements
to identify
the PIP ID, the PIP action and the Service. Multiple PIPs may
be
implemented by the same service, e.g., there may be a Buyer
service
implementing PIPs 3A4, 3A8, 3A9 from the buyer perspective, and
a Seller
service implementing the same PIPs from the seller perspective.
I don't think we should equate PIP ID with Service and action
with "the
particular business transaction within the PIP". Otherwise, we
will not be
able to capture the role information, e.g., the ability to
distinguish a
Buyer Service from a Seller Service, and a request action from
a response
action.
From the RosettaNet point of view, it will be desirable if we
can have
distinct Service, PIP (equivalently BPSS Business Transaction),
and action
elements in the message header. Alternatively, we can use the
Service
element to capture role information (e.g., Buyer vs Seller),
and use the
Action element to capture the PIP ID. Whether we are dealing
with a
request action or a response action will have to be inferred
from the
Service element.
Regards,
-Arvola
Arvola Chan ()
TIBCO Software (on loan to RosettaNet)
+1-650-846-5046 (US-Pacific)
----- Original Message -----
From: "Martin W Sachs" <>
To: "David Fischer" <>
Cc: "Burdett, David" <>; "ebXML
Msg" <
>
Sent: Thursday, July 19, 2001 8:07 AM
Subject: Re: PIP IDs
In my opinion it makes more sense for Service to point to the
PIP ID and
action to point to the particular business transaction within
the PIP. In
other words one execution of a PIP is a conversation. I
believe that some
of the PIPs include multiple business transactions.
Regards,
Marty
************************************************************************
****
*********
Martin W. Sachs
IBM T. J. Watson Research Center
P. O. B. 704
Yorktown Hts, NY 10598
914-784-7287; IBM tie line 863-7287
Notes address: Martin W Sachs/Watson/IBM
Internet address: mwsachs @ us.ibm.com
************************************************************************
****
*********
David Fischer <> on 07/19/2001 10:45:54
AM
To: "Burdett, David" <>
cc: ebXML Msg <>
Subject: PIP IDs
David, in the F2F you mentioned the need for a new element to
contain
industry specific business process identifiers such as a
RosettaNet PIP
identifier. Could this be done with Service/Action where the
Service would
be something like RNet and the Action something like PIP3A1
(Request
Quote)?
<eb:Service>urn:services:RNet</eb:Service>
<eb:Action>PIP3A1</eb:Action>
Regards,
David Fischer
Drummond Group.
------------------------------------------------------------------
To unsubscribe from this elist send a message with the single
word
"unsubscribe" in the body to:
------------------------------------------------------------------
To unsubscribe from this elist send a message with the single
word
"unsubscribe" in the body to:
------------------------------------------------------------------
To unsubscribe from this elist send a message with the single
word
"unsubscribe" in the body to:
------------------------------------------------------------------
To unsubscribe from this elist send a message with the single
word
"unsubscribe" in the body to:
