(this is the abbreviated version of the minutes. See the "notes" files for more details as well as the slide sets used by the presenters. All are available on GRAAP's GridForge site). meeting minutes of the 1st GRAAP-WG meeting at GGF9: ---------------------------------------------------- Agenda: 1. overview of activities since GGF8 2. intro to WS-Agreement 3. series of speakers for domains 4. plans for the rest of this GGF Meeting Goal: make people understand WS-Agreement and its applicability to their specific domain. Jim gave an overview of current state of the group and an introduction to WS-Agreement. Since GGF8, we had: - a face-to-face - telecons - collection of issues with current draft - more interaction with other groups, particularly JSDL. We renamed the spec. to WS-Agreement from OGSI-Agreement for political reasons and to reflect the generality of the approach beyond the grid. OGSI ran into problems with "Grid" in their name in this respect. The overview described terminology used and the roles played by different parties. Discussion made it clear that there are really four roles, initiator and provider for the agreement need not be the same as client and server for the service that is operating under the agreement. There is also relationship between the lifetime of the agreement, and that agreed upon. It was also pointed out the agreement need not be relative to another grid service. It could be about anything. These two aspects were shown in a grid with examples of agreement lifecycle and target lifecycle as well as agreement target being a grid service or not with examples for each quadrent to make the point clear. We discussed the negotiation model based first on factories, then on interactions with the service representing the agreement. We followed with a series of speakers emphasizing different aspects of WSA or different use models Asit Dan spoke on the importance of defining "guarantee semantics" which really means making the conditions of the agreement clear, and making them monitorable and therefore enforceable. Penalties for failing to meet the agreement must also be made clear. He discussed the need for specifying these guarantees either through a new term type that plays this role explicitly or by augmenting every term with this sort of information. These options should be taken up by the group to better support these concepts. Kate Keahey spoke about the National Fusion Collaboratory. In this environment, there are hard deadlines for running codes based on the next batch of experiments. Simulations need to be completed in time to tune the next pulse of the experiment sequence. Agreements should be a mode for specifying these deadlines and guarantees. Ultimately, we need to be able to compose agreements and specify the assurance level for the agreement. Jon Maclaren spoke about a project starting at Manchester on brokers and superschedulers for the grid. This should support workflows and dependencies. Ultimately, the goal is to provide a service that spans the spectrum of pure batch to pure reservation. The goal is to push away from reservation as it leds to gaps in the schedule and therefore lower utilization. There was discussion that, ultimately, some of this is based on who's needs are being met. Batch is for the provider as it leads to best utilization, reservation is for consumer as it gives them the most flexibility. The approach to this middle ground is a negotiation approach. This looks more like "old scheduling" such as found in airlines and so on. The project's goal is to unify "old scheduling" with compute scheduling into a single approach, getting the benefits of both. They're working with a group from Nottingham which understands "old scheudling" and are forming a workshop to get the groups in discussion. There was some discussion around the idea that the two approaches could be merged via linked agreements where one master agreement might govern one domain and other, sub-agreements could be made to handle the other domain in accordance with what was agreed to previously. Steven Newhouse talked about the GESA-WG and it's relation to WSA. GESA has currently defined a negotiation model for pricing of services. It builds upon OGSI mechanisms for lifecycle and advertising via SDEs. Terms here would include things like pricing models, currency, cost and a definition of what the "consumable" is. An important result of this work is that a commitment phase is really needed to reach agreement in these sorts of scenarios. There was also discussion that it must be possible to negotiate in two dimensions at the same time. For example, cost of access to differnet resources, or cost of alternative configurations. There was also some discussion that particular negotiation models might be useful such as specific types of auctions. How do we capture these interaction models? Chris Smith talked about early experiences with WSA in the context of Platform's CSF. The goal of CSF is interaction between multiple resource managers. He enumerated the different types of service and agreements they created. Thus far, all agreements are relatively static in that all terms are required, and cannot be negotiated. He described the agreement terms used in CSF, but also the need to standardize these such as via JSDL-WG. Discussion also pointed out that we really need a framework for implementing WS-Agreement as presently everything is being hand coded, but there's belief that there is lots of common code that could be factored out into a common framework/class library, etc. Ali Anjomshoaa spoke briefly about JSDL and its relation to WSA. JSDL still has a charter of doing an XML spec for job description. WSA popped up and is something the group should consider, and there is one proposal already made in this direction which is to be discussed at one of the JSDL sessions. We ended with a summary of upcoming sessions related to the work of the group, and a request by Jim to get people thinking about applying WSA to their domains.