Minutes of the OGSI-WG Session Global Grid Forum 5 22-23/July/2002 Session 1: Update, Status, and Planning 22/July/2002 15.00 - 16.30 Minute take: Frank Siebenlist Steve Tuecke presented slides: http://www.gridforum.org/ogsi-wg/minutes/2002-07-22_ggf5_update.pdf The following items are discussion points related to these slides: * Slide 11: A Timestamp for Distributed Computing Steve explained that the "orphaned" spec was revitalized recently, because of its importance for OGSI. It is more generically useful than just for ogsi, which brought up the question whether ogsi would be the appropriate place for this spec. The agreed conclusion was that as long as there is not better place, we'll keep it in the ogsi-wg. * Slide 12: Grid Compliance Service General agreement that defining a set of compliance tests could be good for this group. Steve Graham: Caution. Do a scoping exercise. How expensive will this be? Not cheap to build something like this. Brian Carpenter: There could be legal liability implication of specifying "compliance". Ian Foster: Argue strongly against thinking about compliance tests that go into technical spec. Can have parallel efforts to do this, but not in core group. Maybe compliance specification is too much - maybe better informational "best practices". Look at the web services world and how they have their own interoperability organization WS-Interoperability. Steve Tuecke: Suggestion to put a more complete draft together by people interested in this area, and revisit this matter later once we have something more concrete to evaluate. Rob Baxter "volunteered" to take the lead (that he already seemed to have). Also need to ask guidance from GGF on the legal implications. * Slide 13: Java OGSI Hosting Environment Agreed by show of hands that it should be handled in a separate working group. * Slide 14: OGSA Security Document Proposed and agreed to have its own ogsa-security working group. (Also discussed at OGSA Roadmap BOF, and security area meeting.) It won't collide with other security related WGs, like GSI and CP, because the charter will be ogsa specific. * Slide 15: Grid Service Primer Need a document that includes motivations, approaches, decisions. The intended audience is the service developers community. It shouldn't be implementation specific, but more how the patterns, like notification and factories, are applied. Who will step up? Kate Keahey "volunteered" May not need to be part of ogsi - but needs place to publish. Leave it to the authors how they want to proceed and where they want to publish. * Secure Grid Naming Protocol Andrew reminded us that SGNP was also posted to wg email list, and should be discussed. Discussion about whether naming is well enough addressed within the WG? Should ogsi define just one naming scheme or should it be generic to accommodate more flavors? The Handle system was mentioned as a possible alternative. Further discussion deferred to next ogsi-wg session. * Need more time between new drafts and meetings for people to digest There was a claim that GGF has a 14 day cut-off that should be adhered to. Steve: To my knowledge, there is no such rule, and I cannot find any such rule in the GGF document process document. There is a logistics-driven cutoff if you want documents printed for a GGF, but that is not the issue here. I will follow-up with Charlie to get clarification. * Slide 22: Implementations of OGSI 1) Thomas Sandholm (ANL): Globus Toolkit OGSI Technology Preview Thomas gave a short overview and status update. The technology preview release will be the basis for next Globus Toolkit (GT3) 2) David Snelling (Fujitsu): Unicore OGSI Demonstrator Unicore complex batch job workflow. (www.unicore.org) Move on top of Webservices (wrapped Unicore servlet implementations) Totally parallel effort. No shared code with #1. * Slide 24: Working Group Chair Co-chair Jeff Frey (IBM) went on 6 months leave. Chairs have nearly total control, which would a reason for a co-chair. Steve himself admits that he could use help. Co-chair was recommended by members by show of hands. Co-chair should come from industry was generally supported. * Slide 25: Weekly Phone Call Suggestion to use a tool like "placeware" to go with the call. W3C WSDL WG uses a IRC channel in parallel with the call. Concensus is to have weekly call at the proposed time. * Slide 26: Mini ogsi-wg Meeting Concensus is to have an interim meeting at Argonne (Chicago) in late August or early September. In session 2, 6 to 3 vote for Sept 4th over August 28th. ---------------------------------------------------------------- Session 2: Technical discussions 22/July/2002 17.00 - 18.00 Session 3: Technical discussions 23/July/2002 13.00 - 14.30 Minute takers: John Rofrano, Josh, Steve Tuecke Both sessions were debates on handles and references (section 4.7 of the Grid Service Specification, draft 3) Resolution: Leave handle as a URI. Resolution: Pull http(s) resolver protocols out to separate specification document. Considerable discussion about "same Grid service instance". Andrew will follow-up on email list with concrete language. Resolution (issue 26): There is a notion of a handle that is known to an instance. There may also be "handles" that are not known to the instance. This needs to be clarified in the spec. Issue 27: Choices for handles (as known to instances): 1) exactly one handle, in exactly one uri scheme (straw poll: 8) 2) at most one handle, in each uri scheme (straw poll: 14) 3) any number of handles, in each uri scheme (straw poll: 8) Resolution: Provisionally go with #2, but do not completely close. Onus is on counter proposals (on ogsi-wg mail list) for #1 and #3 to be compelling. If #2 or #3, how does and instance decide which handle to give out to whom? Resolution: This is not mandated in spec. Questions still to be answered: 1) Can the set of handles that are known by an instance change during the lifetime of that instance? 2) Can clients request or require that new handles be added to the list that is known by the instance? 3) Must a handle be resolvable? For all time? 4) What is the relationship of a "handle" with "identity"? 5) Should there be a isThisHandleYou(GSH) operation?