OGSA Teleconference February 26, 2004 ===================================== * Participants Latha Srinivasan (HP) Andrew Grimshaw (UVa) Jem Treadwell (HP) David Snelling (Fujitsu) Andreas Savva (Fujitsu) Hiro Kishimoto (Fujitsu) Fred Maciel (Hitachi) * Summary of Actions Action: UC reviewers should send comments to Jeffrin by Monday (3/1). Action: Andrew to present the naming proposal at GGF10 - To do presentation based on EPRs and show how mapping is done. - Make clear this proposal does not address identity yet. Action: Andrew to talk with GFS people (and Dave Martin) about generalizing the GFS proposal. (Hierarchical naming scheme that at the lowest level does not necessarily map to files.) * Early Discussion ** Use cases - Checkpoint next Monday. (Arranged timeslot in call.) Action: UC reviewers should send comments to Jeffrin by Monday (3/1). ** OGSA document pen - Andrew will have some PE additions ready for the OGSA document early next week. - Andreas has the pen at the moment, email him to get the latest document. * WS-Naming - Andrew gave an overview of naming schemes in distributed systems. - Multi-level naming schemes are common in distributed systems. - Maximum levels: 3 - One level is typically human-space - In OGSI, GSH is(can be) an abstract name. GSHs do not exist in WSRF. - Legion example: - LOID: abstract name similar to GSH - 'Context space' has a number of operations - int insert("name", loid) - int remove("name") - LOID lookup("name") - LOID_list multilookup(regexp) - context spaces: directed graph of context object - Hierarchical, borrowed a lot from filesystems (e.g., ".." would also work.) - Something like Unix procfs (could check what interfaces or attributes a 'service' supports and also invoke operations on it.) - e.g., command-line that takes GSH or directory path (might not be able to disinguish between them without extra information). - There is no global/universal context. - So if I have no context how do I get one? (Legion: a setup script has a configuration for setting the root contenxt. Defined two context by default: root and current context. So bootstrapping is out of band.) - Contexts are hierarchical. But it is possible to have disconnected hierarchies (islands) or may have arbitrary links between the hierarchies. No need for absolute names. - Problem: WSRF collapsed GSH and GSR into RenewableReferences (RR). No longer have anything like an abstract name (GSH). - Also RRs cannot be compared - (OGSI GSH at least had textual equality.) - Also would be nice to be able to ask a service 'are you this thing?" - This is not in the OGSA (and it wasn't in OGSI either). - Dave gave a summary of the RR proposal: - Mechanism to associate identifier to resource (resource is something like host, file, fileserver, etc.) - WS that fronts resource provides the 'identifier'. The identifier is not guaranteed unique. - EPR referenceProperty contains resource identity. - EPR also contains - address of WS. - Policy component that defines how the EPR can be renewed. i.e., Resolver information that can return new information about the EPR (one new EPR). - In brief: RR is a special case of an EPR - (EPRs are defined in the WS-Addressing specification) - (An EPR that contains an endpoint that associates a web service to a resource becomes a resource qualified EPR) - (An EPR may also contain a policy component; and the policy may be renewable.) - (Is there any guarantee that EPRs are permanent? Might renew an EPR number of times and end up with an EPR that is completely different.) - (EPRs might end up being quite large.) - The basic mechanism is there, but the naming semantics we want are not there. - Proposal is for OGSA to specify extra semantics so that clients/invokers can assume those semantics. - One possibility is "name" to EPR translation (or some definition of the nature of resource qualified name). - But the second possibility does not handle directory like schemes. - Going from abstract name to EPR is probably easy/simple; not so the other way. - ('renewable' is something added to the open policy section. It is an extension.) - By codifying usage could define abstract names as possible extensions (to the EPRs) without requiring more stuff. - Abstract name vs identity - identity is different from abstract name. We should define how we determine it. Action: Andrew to present the naming proposal at GGF10 - To do presentation based on EPRs and show how mapping is done. - Make clear this proposal does not address identity yet. Action: Andrew to talk with GFS people (and Dave Martin) about generalizing the GFS proposal. (Hierarchical naming scheme that at the lowest level does not necessarily map to files.) * AOB ** Enterprise Grid BoF: - People interested in attending but not enough (up to date) information about this BoF. - Kickstarted by CeC; Seems to have come out of GMAC (partly) - Under architecture area - A number of proposals for chairs; not fixed yet. - It's a research group; more open ended, longer timescale.