OGSA Teleconference - 9 October 2006 ==================================== * Participants Jay Unger (IBM) Jem Treadwell (HP) Paul Strong (eBay) Ellen Stokes (IBM) Dave Snelling (Fujitsu) Andreas Savva (Fujitsu) Fred Maciel (Hitachi) Hiro Kishimoto (Fujitsu) Chris Kantarjiev (Oracle) Chris Jordan (SDSC) Alan Clark () Mukund Buddhikot (SUN) Michael Behrens (R2AD, LLC) Minutes: Andreas Savva * Summary of new actions AI-1009a: Review slide 6 again and clarify what the problem is (Reference Model Design Team) AI-1009b: Review the Reference Model 'Manage Job' in more detail. Specifically look at it from the perspective of an enterprise application (e.g., a 3-tier app) (Reference Model Design Team) AI-1009c: Hiro to ask Jay to provide his comments for Slide 6 - Perspective of submitter vs platform provider and the hierarchical model connection AI-1009d: Jem will produce a topic discussion list for the converged glossary * October 5th minutes accepted with minor corrections - Correct Abdeslem Djaoui's name - Add Andreas Savva to participants - Correct AI misnumbering * Oct 2 minutes approval postponed - Many people were not aware that the Oct.2 minutes had already been posted. * EGA Reference Model ** HPC Cluster Use Case Explained by Hiro - [slide 2] Note that this description is more general than the scope of the *HPC Profile*---this is an HPC cluster use case - [slide 4] The "Abstract Application Definition" appears twice in the stack - There is no problem doing this from the view of the EGA reference model. (It is not necessary to have a 1-to-1 mapping of things on the stack.) - In this case the "HPC cluster" level is probably better called "virtualized platform". Agreed to make this correction. - [slide 5] Provisioning, etc, are noted as out-of-scope of the HPC use case. They do happen but they are done (typically manually) at very rare occasions---e.g., at installation or upgrading of environment---and might occur a long time before or after the actual HPC use case of submitting a job. - There is no actual model conflict here but that is perhaps due to the HPC work not focusing on these aspects even though there are a number of assumptions. (And the assumptions should be made explicit.) - One consequence of this analysis might be to start looking at some other use cases; and so leverage other work from the Reference Model. Staying only on the HPC use case might provide enough coverage. - The works otherwise seem complementary - "Managed Job" in the case of the EGA reference model is probably a much richer concept than in the HPC use case - A longer lived application vs one-off invocation - Interactive vs non-interactive; but this may be less true for HPC applications nowadays - If one can thinks in terms of aggregation of HPC jobs as higher level services then it would fit better into the EGA model - [slide 6] Raises the issue of difference in job vs application state model - Reviewed Paul's suggested alternate figure sent to the mailing list. Agreed it is a good way of looking at things: workload - job +- hpc cluster - node +- binary - This relationship occurs frequently. It is the same for database/database service/database software. (An instance of a database service is composed through managing the things it depends on.) - There is a danger here of trying to do a very elaborate model (like CIM). It is important to know where to draw the line and stop. - OGSA position so far is close to "stop once the service is identified"; the Reference Model on the other hand takes a few more steps towards CIM; but aims to have a broad rather than than a deep view - There is no assertion by the Reference Model that there is only one possible graph - Another difference is what the submitter of the job cares about (needed characteristics/attributes) vs what the environment/platform provider cares about (which includes the hierarchical model needed to instantiate things) - (With the reference model it is enough to focus on the current node; no need to try and dig deep in the hierarchy.) - Some preliminary work is being done in OGSA in the context of the EMS scenarios---the JSDL and CDL connection. - There is a need to discuss and refine the next version of this figure; and incorporate Jay's comments - [slide 5] Need more discussion on what "Submit Job" means - Also "Manage Job" in the Reference Model is much richer than what the HPC/BES model are doing - This is essentially the difference between (discrete) invocations of an HPC job and taking care of an enterprise application that once up has to be taken care of, resources provided appropriately to maintain the required service level, etc. AI-1009a: Review slide 6 again and clarify what the problem is (Reference Model Design Team) AI-1009b: Review the Reference Model 'Manage Job' in more detail. Specifically look at it from the perspective of an enterprise application (e.g., a 3-tier app) (Reference Model Design Team) AI-1009c: Hiro to ask Jay to provide his comments for Slide 6 - Perspective of submitter vs platform provider and the hierarchical model connection ** Glossary review Jem has produced an initial draft with EGA terms highlighted. Agreed that the target is a converged glossary by GGF19. (The HPC discussion is higher priority.) It is important to know where there are differences or discussion points. Agreed that the discussion should proceed by email and the occasional time slot on a teleconference. AI-1009d: Jem will produce a topic discussion list for the converged glossary * AI Review [Only items with updates are listed] AI-0928b: Closed (see next item) AI-0928c: Closed. Obsolete since we have agreed to provide normative text AI-0925b: Closed. Draft was posted to gridforge AI-0921b: Closed. Started writing charter. AI-0921d: Closed. AI-0914f: Closed. Picked location and launched zoomerang. AI-0803e and AI-0803f: No update. - Hiro volunteered to talk to the CDDLM-WG. * OGSa Roadmap Review Chris Jordan gave an overview of current state, followed by some discussion on the status of various documents. - There have been many contributed updates, in particular from Jem, Hiro and Andreas. - Still missing updates for some documents, especially HPC Profile and Naming. - With respect to the state of documents Dave S mentioned that the status of the WSRF and WSN specs has changed since the publication of Roadmap 1.0. - For the Roadmap however the various specification owners (that actually refer to WSRF or WSN) have to update their documents for the Roadmap tables to change. - Chris J is leaving the update of the tables for last. - Status of HPC documents - The HPC Use Cases have finished their public comment period on Sep.20. - Not sure about the HPC Profile - Naming documents: Not sure. The profile does seem close to completion. - Review progress at the next Roadmap call * Other business - Jem has updated the OGSA Wikipedia entry to bring it up to date with the publication of OGSA 1.5 and Basic Profile 1.0 - Thursday's call is cancelled