Discussion on Computing Share.

How to select a specific share? Two options were presented. One BES endpoint per share or add the requirement to JSDL and let the lower levels read the JSDL to select the share. Sergio also suggested adding this information to the BES message. BES group members seem to agree that this should be added to the JSDL.

Chris Smith asked if anyone had looked at the RSS schema. A metadata request to RSS will not necessarily send the query to BES.

In response to Sergio's slide on descriptive vs declaritive, Chris thinks that the desciptive approach is OK but we should be aiming for the declaritive approach.

Discussion on XQuery: Andrew didn't want know the procedure he wanted to know the goal. For XQuery we would only need the predicate. The computing service endpoint is not neessarily a URL it could be a APR. Balaz seems to think that most people didn't like the query approach. We should look a Condor as an example, requirements = true. OK,we can use XQuery but only a simple match.

JSDL needs to be free to define different query types such as LDAP, SQL, RDF etc.

How to handle backwards compatibility with new versions?

If separate element namespace is used , dialect + version, the implementation will look at the namespace.

The get status document is not such a good idea.

Action: e-mail Activity fragment schema to the BES working group.

Adding Glue to BES with #other is OK.

How do we ask for the Glue attributes? Use an extra parameter.

Profiling, should be a working group as it will produce a document.

Action: JSDL + BES should do a glue profile

Short dicussion on Glue info reterival endpoint?

Action: email the 3 working groups.

Balazs: Now we have glue what are you going to do with the information model in BES?

Andrew: BES should add the Glue information model , need further discussion between the two groups.

Session 2: Glue Overview

A question was raised on information about the expiry date for the certificate.

Session 3: Discussion

Is it possible to describe grouping of a WN? For example in EU-IndiaGrid they are not allocating a single slot but multicore parallel job on a single machines.

Solution: Execution Environment logical cpu + max slots per job. Then need to send parameter to the batch system.

JP: Being able to describe the hardware configuration is good not sure about the need for detailed state information in this scenario. JP raised a question about share priority. Laurence replied that priority is a relative number and difficult to assign a real meaning in a broader context. An example was given on the priority attribute in pbs as a usecase.

A question was raised about when one entity references another entity? Do we use the value or ID, eg computing activity The activity is an exception so a general rule is not required.

Computing service, is that a list of states? Active and pending states. It is useful to know the sate of the cluster. Go for well known states and as we can always extend. Description contains grid jobs should this be local jobs?

Action: We should remove the word grid jobs.

Discussion on the file system attributes. We should look at the WN environment variables for inspiration. Needs further discussion A question was raised on the next downtime and a suggestion for multiple downtimes instead.