Minutes of the OGSA Data Design Team, 26th January 2005 0) Ongoing actions. - Simon to mail the list about different types of metadata: data resource vs. service. - Dave to mail Allen about the new proposed time slot. - Dave to double-check AG arrangements for the London meeting with Steven. - Dave to mail Andrew about BOF for BasicFiles. 1) Early discussion * Roll call Susan Malaika, IBM Mario Antonioletti, EPCC Andrew Grimshaw, UVa Mark Morgan, UVa Fred Maciel, Hitachi Ted Anderson, IBM The previous minutes were approved. The agenda was accepted. 2) Update on outstanding actions - Simon to mail the list about different types of metadata: data resource vs. service. This is ongoing. - Dave to check with Dave Snelling when Basic Profile is ready for review. The profile is on GridForge and Dave is encouraging people to contribute. - Discuss time slot in e-mail. The people on the call were happy to move the call half-an-hour earlier. Dave to mail Allen for his view. 3) Architecture issues - Dave has been working on the straw man document, with input from a small group of people. This is beginning to take a coherent shape and he hopes to distribute it more widely after next week's London meeting. The telcons are useful in resolving issues that arise while writing this straw man. - Can we use the BasicFiles interface for both data (bytes) and information (structured data)? Dave explained the issues. Fred suggested two-level architecture. Susan explained Info-D's approach. Talks about published information, brokered via subscribers to consumers, using same model as WS-Notification. Mark explained that the BasicFiles interface is low-level, for reading bytes. Malcolm confirmed that DAIS is assuming an interface with similar operations but working on elements. But OGSA-DAI has run into problems with reading elements that contain images, which then need byte-level reading. Need method of moving fluently between two interfaces. Mark says that Java & C# libraries allow this. - Is data transfer a service? - Are we transferring data (bits) or information (with some semantic content)? Malcolm: clear that we need a number of different interfaces. Do we specify that operations in one interface affect those in another interface, e.g. if both are reading from the underlying stream. Mark: All session info should be kept on client side. Malcolm: How would that link to database running a cursor over a materialised query result? The database may not provide a mechanism for exposing the cursor. Dave: What if the data source were a stream? Mark: The BasicFiles design is very file-centric. Susan: What is the data transfer model under discussion? Dave explained the model, which is pictured in the documents from the metadata team. Susan: Similar to producers and consumers in Info-D. Can we use this as use cases for Info-D? Malcolm: Long-term aim to abstract details of information transfer so that m/w can optimise solutions. In short-term, have to allow applications more control. So currently applications control data movement. Susan: Info-D are trying to allow for the more detailed control as well as the abstraction. Malcolm: Concerned that services are heavyweight, with layers of management etc. Want them to collapse to shared-memory interface when source and sink on same machine. In most cases, the data producer will be the service that controls the information transfer service. Doesn't prevent them passing this over to an intermediary. - How do we specify data types? This discussion was postponed. - Other issues Malcolm: W3C have released three specs related to data transfer: - XML Optimised Binary (XOP), - Message transfer Optimisation mechanism (MTOM) and - Resource Representation SOAP Header Block See http://www.w3.org/2005/01/xmlp-pressrelease. - Do we have volunteers to write subsections of the document? Mark (was) volunteered to write some notes on RNS. Susan volunteered to very short DAIS section and pass it over to Mario. Malcolm asked whether it is only resources that want to be named. Counter-argument is that everything that should be named is a resource. But can a single entry in a log be a resource? Intention is to allow naming of everything: can a part of a resource be named? * Flag as issue. * Action to chase BOF for BasicFiles: Dave to mail Andrew. 5) Wrap up DONM - F2F in London, Feb 2nd.