W3C home > Mailing lists > Public > www-ws@w3.org > June 2003

RE: Debating on the usefulness of supporting a) Stateful Web Service Instances b) Stateful Interaction

From: Savas Parastatidis <Savas.Parastatidis@newcastle.ac.uk>
Date: Thu, 19 Jun 2003 21:02:57 +0100
Message-ID: <BC28A9E979C56C44BCBC2DED313A447001D7568E@bond.ncl.ac.uk>
To: "marco" <events@oxfordsociety.org>, <www-ws@w3.org>
Cc: <umit.yalcinalp@oracle.com>, "Jim Webber" <jim.webber@arjuna.com>, "Steve Graham" <sggraham@us.ibm.com>, "Krishna Sankar" <ksankar@cisco.com>



> Today standard web services don't support:
> a) The concept of stateful service instance
> b) Stateful interaction
> -  Object passing, neither by value nor by reference


Stateful interactions are possible today through a variety of ways. The
approach that OGSI has adopted is to use specific grid service instances
to capture stateful interaction semantics. Effectively, a new web
service, called a grid service instance, is created to represent the
state associated with an interaction.

The Grid Service Instance concept also allows resources to be exposed
through a web service interface. For example, you could expose a
database table as a grid service instance and then refer to it using a
Grid Service Handle (which must be first resolved to a Grid Service

The entire approach adopted by OGSI brings an O-O flavour to Web
services. People seem to like this and it seems to work. Personally, I
don't like it, but that's just me.

You can think of other approaches to achieve the same behaviour. As I
said, it's a matter of choice on how well an approach fits to SOA.

Stateful interactions are possible through context exchanges. Look at
the OASIS BTP specification, WS-Coordination and WS-Transactions. In
these specifications, the stateful interactions are identified by the
exchange of context information rather by specific endpoints.
Furthermore, the approaches say nothing about stateful services. The
context information exchanged is rich enough for the interested parties
to reason about the particulars of an interaction.

There is something to be said about how you refer to state and how you
get pass-by-reference semantics (if you really want them). Well, you
could use URIs or more complex XML-based structures to identify state
instances (not service instances). That could be one approach.

There is something to be said on the standardisation front of such
approaches. The OGSI working group has done a great job in standardising
one approach for the grid application domain (although, as I said, the
Object-Orientation feel is not to my liking... I prefer a context-based

Received on Thursday, 19 June 2003 16:03:34 UTC

This archive was generated by hypermail 2.3.1 : Tuesday, 6 January 2015 20:37:09 UTC