GGF13 OGSA-WG session #4 - EM profile ===================================== * EM profile proposal (Hiro) - Candidate profile is an informational document - The Basic profile and Naming profile are also assumed and should be added to the slides (done) - Supporting more complicated variants: e.g., workflow - Including these in scope will probably make the profile definition take longer. - WS-Agreement - It looks like that spec might have to go through another round of public comments. It depends on the outcome of the discussion stemming from the public comment review. - Fig. 1: job submission should not be the service layer/(ref karl's presentation) - EM profile scope - JSDL can be used separately/standalone. - How far WS-Agreement is required is one of the issues that has to be resolved. - Contents (examples): - Need to be careful that this is profiling and not specification writing. So the statement "jobs are ws-resources..." should be part of a referenced specification not part of this profile. - Is placement (scheduling) out of scope of this profile or not? - It is out of scope unless someone really objects (not an issue at this moment). - Issue: Introspection, or getting information about providers - Issue: Should resource consumption tracking by part of this candidate profile or not? candidate specs: UR and RUS * Business grid project description (Hiro) * Basic execution management (Karl) - "...must support...": read this as these are the kind od jobs that globus supports at the moment - WS-Agreement: replaces GRAM rather than in addition to it. So the createAgreement operation would have the same meaning as a submitJob. - container : service provider - job: negotiated service - mgmt state: agreement - Does JSDL as framed today fit with WS-Agreement? Yes. In fact WS-Agreement is specified so that it can be composed with any other domain-specific specification. - Timeline: - The WS-Agreement specification is in revision and might take 3-6 months before it can be published. - If the specs were ready today then an EM profile that combines WS-Agreement, JSDL and adds some extra things (unclear) might be good enough but this is not feasible as things stand. - WS-Agreement entities: Difference with Hiro's figure is that the pink layer is the job itself and the agreement layer is the job factory. - 'Negotiation' is a misnomer. The simplest case and what is specified at the moment is just yes/no reply. - Chaining; could be used to model reservations - So WS-Agreement is a pattern that leaves 'everything' up to the specific domain language. Returns an EPR to the agreement. - Once a domain specific agreement is created it may also have porttypes that are domain specific. These are probably going to be the 'interesting' porttypes. No such porttypes are currently defined. - Whether data staging is in scope or not - Definition of what files are needed is probably in scope but probably not the mechanism of providing it - Also an issue whether this should be separate from the job submission; it is a form of workflow. * Naregi presentation (Saga) - Want to use EM profile (agreement) between super-scheduler and gridvm. Maybe also between client and super-scheduler but at the moment only JSDL is used between client to super-scheduler. - "...WS-Agreement based brokering is more secure..." - "secure" is used in the sense of "less fragile" here * Unicore and EM profile (Dave Snelling) - Overview of Unicore and Unigrids approach - JSDL satisfies most requirements and has a good fit. With the exception of no multi-job support. - What capabilities are needed to submit pieces and coordinate them (e.g., guarantee they start up together) * OGSA BES strawman - initiateactivityfromjsdl : looks similar to the createagreement operation - Why is activityStates defined as [][]: i.e., return a set of states? - A job may include multiple components/operations. For example, there may be multiple data staging operations which may be done concurrently while the job is still in the hold state. The status of the executable part of the job as well as of each data staging could be returned. * Conclusion - Is it premature or not to attempt an EM profile? - Specifications are still in a fluid state and likely to change. So it looks premature. - But it may still be useful to continue the discussion within OGSA-WG. The proposal is for a candidate profile and so is separate from the API (BES) discussion. - There is a logistical problem with how many people can work on this. It may be be better to postpone starting work at least until next GGF. - How about working out a rough idea of what the contents of the EM profile might look like? - It implies starting the work. And the people needed for the discussion are overextended as is. - Rough consensus that JSDL and BES are probably sufficient and that the WS-Agreement operations may be the same as the BES ones. If so, there might not be a need for this profile. In any case more work on the BES specification is needed to make things clearer before deciding if and how to proceed with an EM profile.