Minutes of the GGF6 OGSI-WG Meeting Chicago 15/Oct/2002 Session 1: Overview and Update ============================== Steve Tuecke (Slides available on the web site) Documents: Grid Service Spec v4 (Major working document), Timestamp for Distributed Computing (Tabled by the working group), Handle URI Definitions (May come into existence within this WG, e.g. http(s), SGNP). This group does have more to do here before the actual creation of these documents. Physiology of the Grid (Several options are possible, but likely to move to OGSA-WG for final clean up) Primer (Status uncertain) Compliance Service: Pending a document proposal before this WG would take it on. A prototype is in process along with OGSA-DAI project. Possibly before this group again at GGF7. Other OGSA related activity (OGSA, OGSA Security, DAIS, Usage Record, Resource Usage, Java Containers) W3C and WSDL Evolution (v1.2) This WG pushed for Open Content and portType extension. Plans are that they will be there in 1.2. Also work proceeding on SOAP, OASIS (General Security issues), Project Liberty (Identity Federation). Implementations (Globus Tech preview toward GT3; Unicore will do something but not at the same time scale as GT3 Alpha Jan 15th 2003; IBM will be providing something within WebSphere via a common implementation with GT3 through the community source process; Python hosting environment implementation coming on the Spring time scale; G-SOAP implementation from Florida with in GridLab project. Activities Since GGF5 (9 Teleconferences, 1 three day interim meeting, broad participation). Document Schedule: GGF6: Draft 4 Nov 11; Draft 5 Nov 14-15: Interim meeting Dec 15: Submit to GGF document process GGF7: Approval as GGF Document. Version 1.1: There may be a need for this, so we are planning for it through minimal temporary solutions. Question: Drop or delay? Leaning toward toward drop rather than delay. Bugzilla: The document is finished when bugizza is empty. Weekly phone calls will continue through December. Interim Meeting 2: at ANL November 14-15. IPR Statements: Copyright issues cleared with initial authors of GSS. Open statement from IBM for RF status with respect to GSS. The status of WS-Inspection specifically is still being pursued. Sun reiterated a RF approach to any GGF document. We will rely on GGF to make this status clear for this and other specifications. WSDL 1.2 Timing: Will probably slip past the 4Q/02. Options: Delay with W3C: Shoe horn into WSDL 1.1: Subset functionality into WSDL 1.1 and adapt: Go as if WSDL 1.2 was complete: Press on W3C between now and the Interim meeting: - This seems to be the consensus at this point. Technical issues for later meetings, see program for times. Session 2: Notification and Events ================================== Background: Today we have ServiceData which is a 'view' on the service's state. Nothing is said specifically about coherency. Access to the data is possible through findServiceData and Notification. There is also lifetime management on the ServiceData. There is a notion of 'edge' versus 'level' triggered notifications. The current spec only supports level triggering. Note that all ServiceData is subject to access controls. There is a difference between the reliability of the service data changes being observed by the service and the reliable delivery of those changes to the client. Part of this stems from the push/pull model. The tie between push and pull model could be tighter in the model. Where do we go from here? Shel F: Do we need reliable delivery in the spec? In this context we do need to keep the distinction between reliable delivery and availability of complete data distinct. Steve G: Asynchronous messaging is critical to Grid Service world. Patterns are the key. Notifications are tied to ServiceData. Made a general proposal for decoupling messaging from the message content. Note that the pattern needs to be kept separate from the binding. Andrew G: Thinking in terms of Events as the base concept with a subtype for notification streams. Shel F: One way to describe the event model is through queuing. Steve G: A publish/subscribe model + Notifications + point-to-point/queuing model is probably what we need. ??: JMS has three entities: producer, provider, and consumer. How does this work in the Grid Service world? Peter V: How does this map to portTypes? Steve G: Don't know yet. Karl C: The point-to-point is captured by the method invocation, but [Steve G] it can't capture the queuing aspect. Malcolm A: Agreement from the perspective of OGSA-DAI. Dennis G: Agree too. Action proposition: Go away and write something that might work with JMS and related issues. This was replaced with a general call for use cases that will allow us to understand these issues better. Susan M: Described the trigger model from the DataBase world. On a first pass all data would need to be exposed by SDE's in order to get the triggers expressed. This is clearly not workable (or safe), therefore, the Grid Service Specification is not rich enough. ??: The WSDL group has some primitives may be defined within WSDL for event delivery etc.. Steve T: Current WSDL 1.1 has four models interaction models. The group is thinking about resolving to something simpler, but Steve is skeptical about time scales matching our needs. We could push it to W3C. Resolution: On the call two weeks from now we will discuss use cases and straw solutions. Registration/Collection: Background: Two parts of the spec have caused some confusion, registration of existence and a SDE element containing the 'contents'. But the nature of the contents was not necessarily clearly defined and the notion of a registry carried too much semantic meaning. There is a current proposal to model only a Collection concept. Primitives: Collection and Registration which are argued to be separate. Some question remains over what the correct factoring is. All possible solutions have been proposed to the list. Andrew G: What is registration primitive for; is it really fundamental enough to be part of the Grid Service Spec? Steve G: Origins are in discovery. Steve T: There may be several higher level functions that might all need this functionality. Jeff N: Noted that the Collection concept had been considered in the early days of the draft, including aggregation of services. Steve T: Registries probably aren't central, but collections are fundamental. Steve G: What constitutes a collection of Grid Services seems to be fundamental, e.g. to apply a GSH to a collection of services. Supported by Jeff N. Steve G: Richer semantics are provided by extending the collection portTypes with more functionality. Dennis G: Agrees, but would push for increased simplicity, or at least consider it carefully before proceeding. Shel F: Pointed out that there were several meanings for aggregation and that we should decide which we want to capture in the spec (implicitly they are not compatible with each other). Steve G: Reiterated that the higher level semantics are provided by the higher level porttypes. Next session will start with "What does it mean to be a collection." Actions: Action: All to post use cases for Notification/Events to the list. Action: Steve Tuecke to approach the Steering Committee to put us on track for a doing a version 1.1. Action: Keith to post pointers to the Python implementation. Action: Massimo Cafaro to provide gSOAP pointers onto the list. - Done Action: Kate to provide an update on the primer. Pending issues (from this meeting only - see earlier minutes for additional items): Is there a returned XML element coming back with a subscription containing additional information about the subscription?