OGSA Teleconference - 14 July 2004 ================================== * Participants Bill Horn (IBM) Hiro Kishimoto (Fujitsu) Fred Maciel (Hitachi) Andreas Savva (Fujitsu) Frank Siebenlist (ANL) Latha Srinivasan (HP) Jem Treadwell (HP) Jeffrin Von Reich (HP) Minutes: Jem Treadwell and Andreas Savva * Minutes July 12th 2004: Accepted * F2F Update Most people OK; Jem, Abdeslem and Dave Snelling can't make it. Hiro asked Heather for WSDM people to come to our meeting, but has not heard anything yet. Hiro will follow up. Proposal is for a 2.5 day meeting - Latha asks how many people might be able to make it - Hiro: so far 10, assume 10-15. - Is the F2F agenda fixed yet? No still tentative, so let Hiro know any suggestions. Focus will be on the v2 document. * Resource Model discussion - Latha: did we decide that we would have a single model, and should we first have a CIM representation of resources? - Fred: didn't decide - we had a lot of contributions from Andrea, which is why there may be some bias toward CIM, but we haven't decided or said that in the document. - Would be good if we can synchronize. - Hiro: to achieve the goal of sharing resources we may have to create a simple common model for the Grid. - Fred: A single model is utopian; desirable but not possible. - Hiro: depends on definition - Fred: problem is not complexity or size; it's consensus---we're not going to achieve consensus at this point. - Fred: Ch 4 of CMM discusses semantics and renderng. Rendering is how to put the model on the wire - e.g. XML. MIB is semantic. When I talk about models, I talk about the semantics. Second level is how to access it. Industry didn't get consensus on one resource model, so WSDM is standardizing using a model-agnostic approach - For example: How do you map CIM operations and attributes to WS-RF, to Web services etc. - simple example, map CIM attributes to resource properties. - So CIM could be rendered onto WSDM. At the Gap Analysis session at GGF11, Heather said don't go through CIM, you map directly to WSDM. - What is our challenge in OGSA? - Fred: which resource models do we need, so which will we adopt. - Bill: Advocating a non-prescriptive approach so that one can pick and choose? - Fred: no, hope we can be reasonably clear - use SNMP for this etc., and then - (Bill:) render through WSDM? Fred, yes. - Bill: so go through the models that are of interest, pick and choose strategic models - Fred: if we want interoperability we must pick the models, map to unicore or GLUE, CDL, CDDLM, JSDL, this kind of thing - they have resources also - they have attributes, but not operations, so if they've received a request, how do I map what's in the request to our model so that one understands the other. One more gotcha: if we're doing deployment - reservation, deployment & monitoring, so we use a resource model for each of these, we will need the resource models to be in sync, can map to each other. Not sure if the concept of identities should be common - e.g., are we talking about a running instance in a Java VM or are we talking about the whole host. - Hiro: JSDL and CDDLM are now creating their specs, so if we can propose an abstract model they can be aligned. JSDL is defining a language for job description, not resource description, so if we can provide a model they would be happy to use ours, not define our own. - Jeffrin: I think Fred is trying to saying that if you think of managing manageable entities you can think of an enormous number of things - schedules, provisioning statements etc.---all kinds of things will need to be managed in OGSA, so fundamentally we have too many models - too many things to be managed. In the case of deployment, while CDDLM could provide the CDL [the mechanism needs to be managed?] too many things that need to be managed---might be worthwhile to think in terms of not trying to match every model to a universal model, but saying that everything that needs to be managed needs a model, and think what needs to be done - but don't know how this can be done! Different layers of abstraction. - There was disagreement if it is easier to handle multiple models or multiple renderings of a single model. - Is this the only way to achieve interoperability? Is it possible to avoid choosing a specific model and do mappings at a higher level instead? - How well adopted is CIM? - There is some adoption on the systems side but not so much on the network side. And very little on the applications side. - For this reason (Jeffrin) seems more interested in mapping between models rather than assuming there is a single model. (It's going to be multiple models so) if you choose one thing above everything else then it becomes just another standard that nobody uses. A first candidate within OGSA to use CIM is EMS. But Fred is worried that we are talking about EMS as an example for resource modeling when there are no EMS people on this call. - Also it is not clear that OGSA has progressed enough to the point of being able to feed in requirements to CIM - CIM is complex and it is not always clear how or where to extend the model - Next step: Hiro to try to get more EMS people to join the next call - EMS Container & Job as two initial modeling topics * Use case document public comments review - Process discussion: Discuss within group first, reach consensus and then post replies to public comment tracker 1. "Service utilization" - 'platform service': This is no longer consistent with the current OGSA document. Perhaps the term should be changed; but in any case it does not sound like a major problem. - Perhaps add some explanation on what the template should be describing. Action: Relevant people (OGSA-WG) should review document and come up with a proposal on how to handle this issue. 2. "Advance" vs "Advanced" reservation Discussed whether 'advanced' actually means something else (e.g., complex reservation) Action: Jeffrin will review document and kick off discussion on whether the two terms mean different things. 2-1. Disagree with suggestion to list requirements for general workflow engines 2-2. Time to process reservation: Hiro agreed that wording is not very good and should be changed. Action: Jeffrin to edit and Hiro to check 3. Agreed to change 'enterpise' to 'organization' though there should not be a problem either way. 4. This is a term defined by this use case. No extra explanation is needed. 5. Abstract: Agreed to re-phrase. Jeffrin to do. 6. Agreed to - Fix figure captions to make consistent - Bibliography per section is ok, but fix inconsistent formatting - Section refs are incorrect: Search for 'section' keyword and make into real cross refs to the section numbers 7. English corrections: ok. 8. 7.2.2 service description: - The comment is not clear; ask for more explanation. - Workflow as GS: Not sure if this has to be drawn out in this case but adding a reference is a possibility. Need more information about 'taverna' and what they think its relevance is. - Collaboration: Reject. It seems inappropriate to add implementations. - Machine readability: Agree in principle but not sure what need to be added. - Collaboration: Add a sentence about 'dynamic nature' and 'description language' - Other comments: - Jeffrin to check again who is the original author of the mutual authorization use case and add to contributors - Acknowledgment: Should 'work supported by' be kept? - Confirm with relevant people first(e.g., Ian Foster) - Should refactor the use case document to make it consistent with the OGSA doc. But it is not clear how much work this will be.