OGSA Teleconference - 23 October 2006 ===================================== * Participants Mike Behrens (R2AD, LLC) Chris Kantarjiev (Oracle) Hiro Kishimoto (Fujitsu) Fred Maciel (Hitachi) Tom Maguire (EMC) Andreas Savva (Fujitsu) Alan Sill (TTU) Dave Snelling (Fujitsu) Ellen Stokes (IBM) Paul Strong (eBay) Minutes: Andreas Savva * Summary of new actions AI-1023a: Tom and Ellen will do a CIM Profile Introduction for Oct.30 (45-60 minutes session) AI-1023b: Tom and Ellen will start working on an introductory example for the F2F. - The exact topic is not decided. It may be good to make sure that it aligns with the XQuery examples that Michel is using. (Tom will contact Michel.) - Michel and Fred should be involved in this work. AI-1023c: Tom volunteered to review Hiro's HPC Cluster use case slides and make appropriate changes to indicate where the CIM Profile would fit in and how it would link into the Reference Model. AI-1023d: Paul will add more items to list on slide 6 of the HPC Cluster use case AI-1023e: Hiro will send out references to relevant OGF specs, e.g., BES and CDDLM AI-1023f: Paul will review relevant OGF specs and give feedback to Hiro for the next version of the use case slide deck. AI-1023g: Hiro will look at other parts of slide 5 and identify further possible refinements of this use case. * Action item review - AI-1005a: Alan has written a draft charter and will be approaching the new ADs. Opinions vary on whether to do a BOF at GGF19 or try and form a WG directly. Doing a BOF would help clarify deliverables. Alan will send the link to the charter draft to the list for comments. (Closed) - AI-0925c: Hiro has sent email to EGA contacts. (Closed) * F2F update Considered and rejected adding a session for the EGA Refererence Model/HPC Cluster use case. Many expected participants cannot join in person. * Information/Service separation discusion - Discussion based on Tom's email sent to the list. - The term 'Profile' is overloaded. The WS-I (also OGF/OGSA) meaning is different from the DMTF/SNIA usage. - Missing link is the connection between the refined (profiled) information model and service; currently this is not done for OGSA profiles (ref BES) - Are these approaches(profiles) really different, or is it just a difference in formats? There are similarities in purpose but they result in significantly different profiles. For example, the WS-I or OGSA profiles do not specialize information models and link that specialization to a service interface. - (An example of such a specialization would be if a relationship that in the model is defined as 0..N is profiled to be appear exactly twice.) - One end result after exploring this area could be additions to the DMTF work register. - Ellen pointed out that CIM does not have a reasonable service model. The existing service model is basically a placeholder and needs refinement before Tom's proposal can be taken forward. - (The motivation of the 'middle bubble' (intermediate model for matchmaking) of the EMS figure is along the same lines.) - Though normative work is the goal some educational work on what is a CIM profile is needed. AI-1023a: Tom and Ellen will do a CIM Profile Introduction for Oct.30 (45-60 minutes session) AI-1023b: Tom and Ellen will start working on an introductory example for the F2F. - The exact topic is not decided. It may be good to make sure that it aligns with the XQuery examples that Michel is using. (Tom will contact Michel.) - Michel and Fred should be involved in this work. AI-1023c: Tom volunteered to review Hiro's HPC Cluster use case slides and make appropriate changes to indicate where the CIM Profile would fit in and how it would link into the Reference Model. - One discussion point was how to represent dags in the Reference Model---they seem to represent dependencies on aspects of the underlying information model. In a sense a different 'slice' of the information models in DMTF. * HPC Cluster use case & EGA reference model [Slide 4] - Added BLAST binary and CHARMM binary and dependencies - Hiro was not sure how to express this in the model. - This slide shows something closer to an instance model rather than an information model. In an information model there would be some kind of relationships expressed; whiler here there is only one kind of relationship and not the type usually used in an information model - There is a relationship to the Application information model. Some explanation of that model might be useful. It was not clear what kind of applications the Information Model covers (packaged application, perhaps implicitly?) and whether it could also cover business functions (probably). [Slide 6] - Hiro started a list of expected functions AI-1023d: Paul will add more items to list on slide 6 of the HPC Cluster use case Discussed the differences between batch job submission and stateful services operating on (users') data and why the latter paradigm is more important here. [Slide 7] - Noted that not all service transitions would have to be realized (some transitions can be 'silent' or implicit) [Slide 9] - Shows an attempted mapping between the two state models. Though there does not seem to be anything contradictory some BES states would probably map to multiple Reference model states. E.g., bes:Pending mapping to both unconfigured and inactive. - Discussed bes:finished as a separate state. The motivation is that after the job completes some information might remain until the user retrieves it (or possibly longer). The service being managed has disappeared and the remaining state lifetime is separate from that of the running state. [Slide 11] - There seems to be some mismatch in the two models. CDDLM makes distinctions between 'initialized' and 'running' state while the Reference model only has 'initialized'. - It may be that the cddlm lifecycle captures two aspects: the lifecycle of the binary and some part of the lifecycle of the service/component - As a basic use case, assuming CDDLM may be more than needed. For example with BES it may be sufficient to assume manual deployment. ** Next steps Discussed whether the HPC Cluster use case is too simple. Though some aspects look simple (submit/run) other aspects (pre-, post-) should provide a good target for further refinement. AI-1023e: Hiro will send out references to relevant OGF specs, e.g., BES and CDDLM AI-1023f: Paul will review relevant OGF specs and give feedback to Hiro for the next version of the use case slide deck. AI-1023g: Hiro will look at other parts of slide 5 and identify further possible refinements of this use case. Next review is scheduled for 2 weeks later. * Glossary review - Discussion focused on 'enterprise' related terms. - Agreed that the term 'enterprise' should be defined and preferably in a way that motivates the definition of other related terms. - Decided after lengthy debate to stay with a simple definition (that may have some unstated exceptions). - Agreed that e-Science should also be added - Refined definitions of 'enterprise computing' and 'enterprise application' - Revised definitions are in the working copy of the Glossary. Jem will uploaded after further refinement.