OGSA-WG Teleconference - 30 August 2004 ======================================= * Participants Ian Foster (ANL) Andrew Grimshaw (UVa) Bill Horn (IBM) Fred Maciel (Hitachi) Susan Malaika (IBM) Bryan Murray (HP) Andreas Savva (Fujitsu) Jem Treadwell (HP) Jeffrin Von Reich (HP) Minutes: Andreas Savva * Agenda bashing - Agreed on the agenda that Andrew sent out before the call - Objectives - What are we trying to do? - Other standards we MUST work with - Discuss straw man document and received comments - (Ian's, Bryan, Peter's, OGSA-DAI document sent by Dave Berry only to Andrew (others haven't seen it)) - Line item and philosophy - Distinguish DESIRABLE properties versus REQUIRED properties - Is there an existing STANDARD that meets the requirements? - (There are many existing implementations) - Next steps. * Define the problem to solve - Initial target - Not trying to do new grid specific standard - It should useful to the wider community and not just the Grid - Difficult to make sure that wider community will accept something developed just within GGF - Need a lot of marketing effort to get buy-in - Identify where work should be done to be adopted by the wider community. - Not aiming to just come up with a specific naming scheme - Make the case for the need for a naming scheme and possibly do some prototype work on proposal - And sell it to major players - For both political and technical reasons: need to make sure major players are involved to ensure success - Do we anticipate a single set of requirements and solution or a range of solutions? - Initial step is try to build a case for naming - Determine required vs desirable properties - There are probably going to be some core requirements - In the end classify requirements using MUST/SHOULD/CAN categories - Also have to show that these are not addressed in existing standards - (And convince people that there is a need for a base name scheme vs N schemes) * Other related/required standards - Anything that does references: e.g., WS-Addressing, or the related but different proposal from Oracle (WS-MessageDelivery?) - WSRF subgroup trying to unify approaches of using WS-Addressing - It's an open question whether WSRF (RR) can (be tweaked to) fulfill naming requirements. - There is no RR spec publically available right now - A suggestions was to do a wider search for similar efforts within other standards work, e.g., W3C. - (Commercial area has been slow to see the need for naming.) * Document walkthrough Ref: https://forge.gridforum.org/projects/ogsa-wg/document/ogsa-naming/en/2 ** Introduction - The document has no motivation section yet. - Andrew will add it before distributing it to a wider audience - Does naming need to be defined separately from metadata? - Metadata mapping (e.g., directory) is higher level (third level) - Here deal with bottom two levels: addresses and abstract names - This point (3 levels) should be made clear in the introduction. It goes to defining the problem to be addressed. - Mapping directly human name to physical/address? - Some implementations might want that; some existing systems do it. - But some flexibility is lost in doing so - Use cases - Need to come up with as many as possible - (To identify as broad set of requirements as possible; and then classify them) ** Terms - Make clearer distinction, if any, between service and resource. - Need for extensibility because it is unlikely that one proposal can address all future requirements. - Security and naming connection: - Should discuss more the need to support different authentication schemes; might have to conform to different ones. - "...number information." - 'cardinality' is probably more correct - Refers to replicants (clients can't distinguish between them) - Resource identity and WS-Security terminology - (Not sure yet how to phrase this) - Binding time: when and how often to refresh mapping ** Desirable properties - Try to tie desirable properties (back) to use cases - Alias - If an id matches they are the same; if not they might still be; and need for more checks. - (This is the same as WSDM) - Location portability: - Name maps to same resource regardless where it is used - Conference room counter-example; find a room based on location of inquiry. (i.e., ability to get service nearby.) - Andrew thinks this should be done at a higher level; above naming - E.g., PDA querying its location and then asking for room nearby - Leads to question of how to find one's location? Typically broadcasting is used. - Not clear that this property falls in desirable or not. - But does sound like a use case to consider. - Transparencies: Agreed that not all are needed in all cases - Trust: - Don't require trust but the ability to support trust. - (Current (grid) environment is in a sense more trusted than the (potential) global one.) - A desirable rather than required property - Agreed on need to be more specific on performance, etc. - Binding: - Bind whenever you need to and get the 'right' answer - Ability to detect stale bindings (e.g., by receiving faults) - And how to detect that if a binding is returned then it does map to the desired entity - Ties into security/authentication - Free (no licensing): - Handle.net one example of a nice approach that is unfortunately not royalty free for commercial use. - Consensus that RF is a requirement for something as basic as naming - Additional commenton Versioning: - Will add in but not sure if it is required; could be done at higher level. - 'Same' document vs different documents * Related work - Handle.net: lots of nice properties. More of a directory service than a naming scheme - Life Science Identifier (LSID) - Persistent identifier (identifying proteins and related annotations across time/organizations) - IBM reference implementation - WSRF/RR - RR document not available yet - "Scalable Naming in Global Middleware" (citeseer reference by Ian) - WSDM definition: UUID. Deals with identification rather than naming scheme. * Comments on Bryan's paper Ref: http://www-unix.gridforum.org/mail_archive/ogsa-wg/2004/08/msg00055.html - Rather than Grid community retooling the problem is for WS community retooling. - Can 'fake' it by just mapping what the clients think are urls (so that the clients are not aware of changes) - The big problem is not the client-side but getting buy-in to make this functionality part of the software infrastructure. * Is there a standard that meets the requirements? - No * Next steps: 1. Andrew to re-draft doc - Merging some of Bryan's stuff where appropriate - Keep within OGSA for now or seek broader audience? - Consensus to do another draft, making a more compelling case before going wider - (Bryan is involved in WSDM/WSRF so there is some input from other groups already) - WSDM identified some required pieces. E.g., identity. - But not trying to solve the entire problem. - If solution is WSRF based then it could be picked up by WSDM too since they are WSRF/WS-Addressing currently based - In the next version get people to comment whether each property is required or desirable, etc. 2. Andrew to consult with Hiro and Dave Snelling (WSRF co-chair) on next steps - Talk to Dave about the state of the RR spec. - Also thinking of creating an 'ogsa-naming' mailing list; possibly other timeslot for future calls. 3. Do a review of name schemes and how far they address requirements detailed in the paper. [And at a later point:] 4. Go to standards bodies, possibly first talking to major vendors - Andrew has some contacts: - IBM: Steve Graham, Jeff Frey - MS: .Net group - WSRF in OASIS: - One deliverable is RR. Could submit the naming proposal to that group - Of course this still does not get buy in from major companies not involved in WSRF (e.g., MS)