OGF23 - DMI Session - 2 June 2008 ================================= Notes: Mario Antonioletti Presentations available from: http://tinyurl.com/59yucb (if GridForge is up) Steven Newhouse gave an overview of the OGSA-DMI. Notes correspond to exceptions outside the presentation. DMI is now a proposed recommendation - the functional specification document for DMI is to be up at the OGF web site later this week. The document went to public comment at the last OGF - got some good comments - these have been addressed and the document was re-submitted. All that remains is to show that the DMI specification works. David Martin suggested that perhaps OGSA-DMI should give the document a final check to pick up any typos. Steven Newhouse countered by saying that he thinks it is not worth going through the spec until the final changes derived from the implementation phase are made. Steven described what an EPR (End Point Reference) is for people that are not familiar with the concept. The typical process for DMI: a client gets in touch with DTF (Data Transfer Factory) - it examines at the DEPRs given to it by the client and inspects these to see how data can be transferred from the source to the sink. This then creates an EPR for a Data Transfer Instance (DTI)- the client uses the DTI to get a status of the transfer. The DTI manages the transfer of data from the source to the sink. Mario asks where the data EPR comes from and as it happens Steven has a number of slides that shows where these are obtained from. In the first instance the expectation is that the DEPRs will be minted by the client in the absence of other mechanisms. ... Erwin asked about future work: is DMI thinking of managing bulk transfers. Response is that we can already specify bulk transports through the source or the sink - from source to sink but not from multiple sources to multiple sinks. Another use case was proposed - use metadata to discover data as opposed to supplying data EPR, then the DTF could use that to look for the data. The response: can use third parties to do that - feed a service metadata that then returns Data EPRs which can then be provided to a DTF. Michel Drescher takes over to talk about renderings and Interop. Currently only have a functional specification that only specifies the core behaviour. We are going to do two different renderings: plain WS (non-WSRF) rendering and a WSRF rendering. The WSRF rendering will be on top of the non-WSRF rendering (call it WS-I). With WSRF we could be OGSA-BP compliant which would give us more constraints - other issue is how WSRF compliant it is (how many of the optional operations should be supported?). Mario comments that as we are called OGSA-DMI maybe that implies that we should be OGSA-BP compliant. ... For the WSRF rendering the DMI Factory could have Resource Property Value Change Notification. Could do similar stuff for the DTI. Aligned to the Resource Management life cycle which fits in quite nicely. We already have some renderings available on GridForge. Still need to decide what WSRF rendering features we want and then define normative renderings, OGSA-DMI ask for a possible implementation from EGEE - no commitment as yet. Interop issues: security - disabled security for the ByteIO interop testing that Michel was involved in and this then happened successfully. DMI spec is silent on security other than for the access security token(s) that is\are passed - validating security is not the primary concern here - in previous interop tests security has been a major headache for as opposed to testing the specification. Can compose DMI with security but it makes specification validation difficult. Protocols used to verify the spec: ftp? Protocol security: is the same username password enough? Guideline should be to do as little as possible but as much as necessary. 4-way principle for validation: - my client vs my service - your client vs your service - my client vs your service - your client vs my service Do we want to move data between Microsoft and Fujitsu? Probably not. An aggressive time line is proposed - have something ready by the next OGF. David Martin asked about WS-I vs WSRF interoperability - WS-I vs WS-I implementations; WS-I vs WSRF; WSRF vs WSRF - this could be done. Some implementations will not support WSRF - Microsoft implementation will not be WSRF compliant; Fujitsu will be WSRF compliant. Could there be a compliant implementation that would not inter-operate? The WSRF stuff would be a superset of the WS-I stuff. The normative description in the functional spec is around shared data types. There are normative documents that will describe the WSRF/WS-I renderings. They will go through the recommendation track separately. The documents will be fairly short (excluding the XML/WSDL). Important to get this tested against these published documents. Get drafts of the rendering for comment for OGF24. Source/Sink - expect ftp server with user name and password; http as a source - could use a put for a sink ... maybe. Question as to whether GridFTP could be used because of certificate issues. Looking for implementors. Steven implemented 75% of the DMI spec in an afternoon so there is no massive barrier. Asked D-grid and Virginia explicitly if they might consider the OGSA-DMI spec.