meeting minutes --------------- GRAAP-WG meeting 3 (Technical Session 2) focus of session: Managing lifecycles for agreements (led by Kate Keahey) ---------------- We defined lifecycle as everything that occurs between creation of the agreement (service) via the factory to termination of the service, currently due to the OGSI termination time. It was pointed out that termination does not necessarily release the parties from terms of the agreement. The logical lifetime could be longer than the "physical lifetime" of the agreement service. Different lifetime states were defined: - Considered - Observed - Committed might be a new one based on the previous session's discussion Also satisfaction states for an observed agreement: - Satisfied - Violated - Completed - Partial satisfaction These states were called out in a state diagram. That led to the question of whether a violated agreement can get back to being satisfied. Consensus of discussion was yes, but that we must capture the fact that a violation has occurred because it may be needed later. How should termination time be used? One use is to limit the amount of time to reach an agreement during negotiation, but termination time itself could be a negotiable attribute. This leads to the idea of termination time as an attribute on individual terms or at least to term groups that make up the agreement though there's concern this starts to become quite complex to manage properly. A question was raised about whether there's any assertion that terms are in a consistent/non-contradictory state. The current answer is no. As related work, there was a question about how negotiation is handled in the legal field. There are likely defined processes and terminology there. But, we are not trying to fully mimic or replace a legal process here. The group does not intend to be able to anything as rigorous as would be required for legal contracts. We also discussed how agreement life time relates to the agreement target's lifetime. Must the agreement live longer? This is what we've discussed previously in the context of an audit system that holds things after finalization. What happens at finalization could be a subject of negotiation, and we may want to provide some default term definitions for this, but the group does not presently want to define a particular persistent audit service. This could look at the RUS-WG for related work here, but in many cases this is likely to be domain specific. Also, what about when the negotiation takes longer than the target's lifetime. You must renew the lifetime of the target much as you would with Kerberos tickets. We also discussed what can occur prior to an agreement. That is, what does advertising do. The notion here was around templates, and that has been discussed in some other groups such as JSDL or DAIS. This could be an easier approach than the current supportedAgreements definition. We also re-asserted the need for a commitment handshake and that without it, the spec. is broken. Asit Dan then presented work on templates ----------------------------------------- The idea is similar to car buying. You're provided with a list of options or option packages to pick and choose from. These option packages are encapsulated in a template. Templates need not enumerate all possible agreements, that's just too much detail and complexity. We want to provide hints with this mechanism, and there's consensus that we can advertise them via SDEs on the factory without adding anything to the port-types. Also pointed out that templates are not the only way for a client to get started with a provider. Perhaps there's an old agreement around they uses as a starting point for a new agreement creation. We also discussed the idea of storing templates in a centralized repository for larger scale and brokering. Issues of how the repository is managed and updated come up with this approach. Also, querying this repository is going to be difficult. A question was raised about the use of WS-Policy since it does not exist in any standards body. That's ok, though, because it is publicly available, and we can refer to any other public document just as well.