W3C home > Mailing lists > Public > www-ws-desc@w3.org > February 2002

Re: Reference requirements

From: Steve Graham <sggraham@us.ibm.com>
Date: Thu, 28 Feb 2002 01:49:20 -0500
To: "Liu, Kevin" <kevin.liu@sap.com>
Cc: "'www-ws-desc@w3.org'" <www-ws-desc@w3.org>, www-ws-desc-request@w3.org
Message-ID: <OF04DEB379.3BCB2A6D-ON85256B6E.00255D53@raleigh.ibm.com>
We make use of references and stateful services in the Grid services work
we proposed with Globus at the Global Grid Forum meeting last week.


Grid Services Spec

Steve Graham
(919)254-0615 (T/L 444)
Emerging Technologies

                      "Liu, Kevin"                                                                                                     
                      <kevin.liu@sap.co        To:       "'www-ws-desc@w3.org'" <www-ws-desc@w3.org>                                   
                      m>                       cc:                                                                                     
                      Sent by:                 Subject:  Re: Reference requirements                                                    
                      02/27/2002 11:58                                                                                                 

Somebody needs to convince me that passing object by references really
sense in the web services case:
1. Does it make sense to pass by references if the services is stateless?
it a good idea to keep a web services stateful?
2. What's the benefit of passing by references vs.  passing by value across
Internet ?
3. As for WSDL, it already has some construct for expressing in/out
parameters using ParameterOrder attribute of Operation - any needs to go

Date: Wed, 20 Feb 2002 12:56:43 +0100 (CET)
From: Jacek Kopecky
To: Paul Prescod
Subject: Re: Reference requirements

 we will be very glad if we can standardize our way of supporting
remote references, and we'll welcome support from other W3C
members if we try to submit a Remote references and Instance IDs
spec as a W3C note.

 However, we will be equally glad if a different standard is
created and adopted and we can support that instead.

 In any case, the problem of remote references is two-fold:
 1) accessing different instances of a single service,
 2) referencing services and their instances.

 We can take your approach of extending the URI of the service
with the instance ID, and it would solve both problems easily,
but I feel nervous about this method because nowhere does it say
that every URI scheme used for accessing Web Services will allow
such extensions (thinking of email here).

 We can alternatively take the SOAP approach - create an
extension that will carry that additional information inside a
header. That's what we've taken. This solves the first problem
and it somehow resembles the notion of "sessions".
 The second problem we solve by passing a structure which
references a WSDL service (by its qname) and specifies the value
of the instanceID for the service, for example:
 In fact, the instanceID element in this structure can (and
probably should) be viewed as an extension of this structure
which without it serves quite sufficiently to reference whole
services. This implies there can be other parameters in such a
structure (if we don't move them to the WSDL description itself)
and I think this is the proper way.

 Oh, then there are HTTP cookies, too, but that way lay madness.

                   Jacek Kopecky

                   Senior Architect, Systinet (formerly Idoox)

On Tue, 19 Feb 2002, Paul Prescod wrote:

 > Anne Thomas Manes wrote:
 > >
 > > Systinet WASP supports remote references using a header element to
 > > the instance id.
 > >
 > > Anne Thomas Manes
 > > CTO, Systinet
 > That's great. Can we standardize it?
 > Even better...why not combine the endpoint URI and the instance ID
 > somewhat like this:
 > <http://www.manes.net/service?instance=instanceID>
 > I have other ideas beyond that but I'll let that one sink in. If you do
 > that then voila you've made it easy for every instance to be an endpoint
 > and all you need is a WSDL for it. Which puts the ball back in WSDL's
 > court. You need a way to say that the return value of a method will be a
 > URI like that and declare the WSDL that goes with that
 > instance-endpoint.
 >  Paul Prescod
Received on Thursday, 28 February 2002 01:50:57 UTC

This archive was generated by hypermail 2.4.0 : Friday, 17 January 2020 23:06:20 UTC