Savas:

I wondered when you'd join the debate, and as usual your comments are very much to the point.

At 06:09 PM 3/1/2005 +0000, Savas Parastatidis wrote:

Please note that I am not suggesting that WS-RF (and WS-Transfer for
that matter) is not useful in certain cases. Indeed, in the systems
management area it may make sense to use them. However, what it is being
proposed in the OGSA working group (if I understand correctly) is that
WS-RF be used as the foundation for all high-level services.

I don't think that the OGSA working group would argue for using it as a foundation for "all high-level services," but I understand that some people fear that this is intended.

For system management applications (a major focus of OGSA), I do believe that there is value, as you also note.

However, a difference
that I personally see between the WS-RF/WS-Notification and WS-Transfer/
/WS-Eventing/etc. camps is that the latter is much simpler and

"much simpler" is a relative term of course, but I think we should also remember that there isn't a lot to even the "more complex" WSRF/WS-BaseNotification. See Blackberry note below.

lightweight. Most importantly, the MS suite of specifications is not
promoted as the uniform, underlying suite of common messaging behaviours
for all high-level services as WS-RF seems to be. Instead it is some

That's not my understanding of WSRF.

patterns that are available to be used where they are appropriate. Also,
WS-Transfer is lightweight so it can be used in very-small-factor
devices.

I've seen WSRF implemented on a Blackberry, with the complete SOAP-WSRF-WSN-WSDM stack occupying 102KB. I *think* that the "small factor" issue is a red herring. If you want to run on something with just 64KB memory, you probably shouldn't be using Web services.

So, it would be interesting to understand from the WSRF architects where
they feel the WSRF approach is and is not appropriate. E.g. is it
appropriate for interactions with multiple resources? Should it be used
in the design of all high-level Grid services? What are the semantics of
the 'destroy' messages?

Here's my opinions:

* WSRF (or indeed any WS-I-compliant specification) should be used when and where it's appropriate.

* Semantics of destroy: these are defined fairly clearly in the WS-ResourceLifetime specification.

* Interactions with multiple resources: I think that's a red herring. E.g., a job factory can return EPRs to WS-Resources representing jobs, thus allowing individual jobs to be monitored and controlled, if needed. The factory can also maintain a service group representing all jobs, and then support operations that allow a client to "destroy all jobs that match a certain pattern" (for example). It's not an either/or.

_______________________________________________________________
Ian Foster www.mcs.anl.gov/~foster Math & Computer Science Div. Dept of Computer Science Argonne National Laboratory The University of Chicago Argonne, IL 60439, U.S.A. Chicago, IL 60637, U.S.A. Tel: 630 252 4619 Fax: 630 252 1997 Globus Alliance, www.globus.org