Minutes of the Data Design Team telcon, 15th September 2004. 0) Summary of actions * Action: Mario to update metadata document by Thursday, and Dave to forward to interested parties * Action: add slot for Peter Clarke and David Martin to talk at start of F2F * Action: Dave to update F2F agenda and remind people of location 1) Early discussion Roll call: David Martin, IBM USA Simon Laws, IBM UK Andrew Grimshaw, U. Virginia Dave Berry, NeSC (chair and note taker) Jim Cox, Boeing Corp Mario Antonioletti, EPCC Peter Kunszt, CERN Allen Luniwewski, IBM USA 2) Update on plans for next FTF & GGF Simon volunteered to present DAIS. Mario will present metadata. Actions as above. 3) Technical discussion -- updates on ongoing work. * Data federation (Allen Luniewski) Nothing has changed since last week; no feedback received. Simon: Will there be a "federation factory" that takes a set of services plus info about interfaces and produces a federation? Allen: Federation isn't static - can add new resources and access methods. Question about how this links to WSRF. Allen's model doesn't add a new Web Service, just adds more WSDL portTypes on the existing Web Services. This may or may not be a new endpoint. Clarification: The policy interface specifies the policy of the federating service. not of the services that are being federated. Clarification: A service need not know that it is being federated. Hence a service does not provide information about the explicit federating services that it can co-operate with. But metadata of a service might provide information for potential federators. Clarification: federation is a form of derivative data products. Clarification: It is up to the implementation how to handle changing schemas. The most we can provide is general guidance. We would like to find policy options that are common to any federation. * Contexts, Files, Caches (Andrew Grimshaw) The document has been updated to: - clarify that document discusses service interfaces. Assumes OGSA naming. - JNDI: OGSA WG is very wary of proprietary interfaces and language-specific interfaces. Can use JNDI it as input for our design. - Removed duplicate functions -- always work with paths and not just file names. - Added explanation of stateless sessions, and explaining with there is no "creat" operation. - Cache interface is about cache management. - New text in progress about containers. A file container is a thing that holds a bunch of file services. Andrew suggested that GGF should create a data area WG to pursue this. We discussed whether it is possible or practical to keep a list of everything that points to a service, e.g. in metadata? Andrew noted that this was very useful in Legion but can lead to dangling pointers. Might not have permission to update the services. Peter noted that EGEE found it was better to handle standard "stat" and ACL metadata separately from more general metadata. We could possibly add metadata about what sort of device the data is stored on. Access control can be added to a context via ACLs. Contexts should not store detailed policies on entries in the context. * Metadata (Mario, Simon, et al) Slightly updated version of document available. Basic model is to find service handle from metadata. Need to focus on what metadata should cover within a data context. Haven't yet identified what metadata exists specifically for data. Andrew suggested that we can define a base capability and then add a query notion on top. Logically metadata can be kept with resource (not necessarily really kept there, provided we can find it there). Logical view needn't correspond to implementation strategy. Then can build metadata catalog on top of this (either pull or push model). Peter noted that EGEE keeps metadata in catalogs and this has been a mixed experience. 4) Glossary items (15 min) Held over for a future meeting 5) Wrap up DONM - GGF12